Up next in 10
In this episode of the Global AI October sessions we focus on how to get started in the field of AI.
Show More Show Less View Video Transcript
3:24
no it's not that it's just like the
3:26
no it's not that it's just like the
3:26
no it's not that it's just like the the projection from many to many to me
3:29
the projection from many to many to me
3:29
the projection from many to many to me seems really
3:30
seems really
3:30
seems really hard
3:36
[Laughter]
3:44
fantastic
3:58
thanks
4:11
yeah it's a great question i mean we
4:13
yeah it's a great question i mean we
4:13
yeah it's a great question i mean we really focus on making sure
4:14
really focus on making sure
4:14
really focus on making sure we have an offering uh for developers
4:17
we have an offering uh for developers
4:17
we have an offering uh for developers really of all skill levels and so there
4:19
really of all skill levels and so there
4:19
really of all skill levels and so there are a couple things that we we focus on
4:21
are a couple things that we we focus on
4:21
are a couple things that we we focus on um one is our suite of cognitive
4:23
um one is our suite of cognitive
4:23
um one is our suite of cognitive services where these are
4:25
services where these are
4:25
services where these are sophisticated ai models that microsoft
4:27
sophisticated ai models that microsoft
4:27
sophisticated ai models that microsoft has built so if you want to add
4:29
has built so if you want to add
4:29
has built so if you want to add say speech to text directly to your
4:30
say speech to text directly to your
4:30
say speech to text directly to your application you can do it calling a web
4:32
application you can do it calling a web
4:32
application you can do it calling a web service and so really simple to get
4:34
service and so really simple to get
4:34
service and so really simple to get started on that
4:35
started on that
4:35
started on that and then sort of at the other end we go
4:37
and then sort of at the other end we go
4:37
and then sort of at the other end we go all the way to azure machine learning
4:39
all the way to azure machine learning
4:39
all the way to azure machine learning which lets you build your own custom
4:41
which lets you build your own custom
4:41
which lets you build your own custom models and call them directly from
4:43
models and call them directly from
4:43
models and call them directly from your applications and we focus there on
4:45
your applications and we focus there on
4:45
your applications and we focus there on making sure it works for all skill
4:46
making sure it works for all skill
4:46
making sure it works for all skill levels whether you're
4:48
levels whether you're
4:48
levels whether you're a novice who's working with uh with a
4:53
novice who's working with automated
4:54
novice who's working with automated
4:54
novice who's working with automated machine learning or someone who's
4:55
machine learning or someone who's
4:55
machine learning or someone who's experienced using the designer or
4:57
experienced using the designer or
4:57
experienced using the designer or even a notebook you can get started from
5:08
there
5:29
yeah we really try and make sure that
5:31
yeah we really try and make sure that
5:31
yeah we really try and make sure that you know wherever you are in your ai
5:33
you know wherever you are in your ai
5:33
you know wherever you are in your ai journey that we've got something that
5:34
journey that we've got something that
5:34
journey that we've got something that will work for you
5:50
always uh we're constantly innovating
5:52
always uh we're constantly innovating
5:52
always uh we're constantly innovating and developing new things
5:54
and developing new things
5:54
and developing new things uh there are a couple of exciting things
5:56
uh there are a couple of exciting things
5:56
uh there are a couple of exciting things that i'd love to talk to you more about
5:57
that i'd love to talk to you more about
5:57
that i'd love to talk to you more about three in particular
5:58
three in particular
5:58
three in particular first is metrics advisor metrics advisor
6:01
first is metrics advisor metrics advisor
6:01
first is metrics advisor metrics advisor really brings
6:02
really brings
6:02
really brings real-time monitoring um you know based
6:05
real-time monitoring um you know based
6:05
real-time monitoring um you know based on ai to all of your data streams and so
6:07
on ai to all of your data streams and so
6:08
on ai to all of your data streams and so from that monitoring we'll show you the
6:09
from that monitoring we'll show you the
6:10
from that monitoring we'll show you the anomalous points and you can get
6:11
anomalous points and you can get
6:11
anomalous points and you can get alerting and root cause analysis in in
6:13
alerting and root cause analysis in in
6:13
alerting and root cause analysis in in real time
6:14
real time
6:14
real time um but more than talk about these things
6:16
um but more than talk about these things
6:16
um but more than talk about these things this is why seth is here seth maybe you
6:18
this is why seth is here seth maybe you
6:18
this is why seth is here seth maybe you could help us out and
6:19
could help us out and
6:19
could help us out and show us a demo of that and see it in
6:20
show us a demo of that and see it in
6:20
show us a demo of that and see it in action that is correct
6:22
action that is correct
6:22
action that is correct i am here as the demo monkey
6:26
i am here as the demo monkey
6:26
i am here as the demo monkey and i'm excited to do that so let me go
6:28
and i'm excited to do that so let me go
6:28
and i'm excited to do that so let me go over and start to share my screen here
6:31
over and start to share my screen here
6:31
over and start to share my screen here i think this is the right one you know
6:34
i think this is the right one you know
6:34
i think this is the right one you know you never know which screen you're gonna
6:35
you never know which screen you're gonna
6:35
you never know which screen you're gonna share it's always so
6:37
share it's always so
6:37
share it's always so i think it's this one yes
6:41
i think it's this one yes
6:41
i think it's this one yes i think all right eric how much time do
6:43
i think all right eric how much time do
6:43
i think all right eric how much time do you want me to do this in
6:45
you want me to do this in
6:45
you want me to do this in uh i don't know i think we're probably
6:46
uh i don't know i think we're probably
6:46
uh i don't know i think we're probably behind so you'll have to do the uh the
6:48
behind so you'll have to do the uh the
6:48
behind so you'll have to do the uh the medium fast version fantastic so
6:51
medium fast version fantastic so
6:51
medium fast version fantastic so basically metrics advisor what it does
6:53
basically metrics advisor what it does
6:53
basically metrics advisor what it does is it takes time series data
6:55
is it takes time series data
6:55
is it takes time series data such as the one you see here this is
6:57
such as the one you see here this is
6:57
such as the one you see here this is time series data based on multiple
6:58
time series data based on multiple
6:58
time series data based on multiple dimensions
6:59
dimensions
6:59
dimensions with multiple metrics basically tickets
7:02
with multiple metrics basically tickets
7:02
with multiple metrics basically tickets that come
7:02
that come
7:02
that come out of california and washington new
7:04
out of california and washington new
7:04
out of california and washington new york we have hours views answers and
7:06
york we have hours views answers and
7:06
york we have hours views answers and comments
7:07
comments
7:07
comments and basically metrics advisor will take
7:09
and basically metrics advisor will take
7:09
and basically metrics advisor will take this stream of data
7:10
this stream of data
7:10
this stream of data and it will create for you something
7:13
and it will create for you something
7:13
and it will create for you something that looks like this
7:14
that looks like this
7:14
that looks like this it will show you all of the actual data
7:17
it will show you all of the actual data
7:18
it will show you all of the actual data as it goes not only by some but also by
7:21
as it goes not only by some but also by
7:21
as it goes not only by some but also by all of the dimensions so for example
7:23
all of the dimensions so for example
7:23
all of the dimensions so for example these are all of me scroll down here
7:25
these are all of me scroll down here
7:25
these are all of me scroll down here these are all the tickets that are of
7:27
these are all the tickets that are of
7:27
these are all the tickets that are of type high
7:28
type high
7:28
type high right and it's all the locations and so
7:31
right and it's all the locations and so
7:31
right and it's all the locations and so notice there's a ton of anomalies and
7:32
notice there's a ton of anomalies and
7:32
notice there's a ton of anomalies and all we had to do is
7:33
all we had to do is
7:33
all we had to do is literally just map the data so let me go
7:35
literally just map the data so let me go
7:35
literally just map the data so let me go into one particular data point so let me
7:37
into one particular data point so let me
7:37
into one particular data point so let me go into this one here
7:39
go into this one here
7:39
go into this one here and you're going to see that you can go
7:40
and you're going to see that you can go
7:40
and you're going to see that you can go to something called an incident hub
7:42
to something called an incident hub
7:42
to something called an incident hub that will describe the entire
7:45
that will describe the entire
7:45
that will describe the entire thing basically without you having to do
7:48
thing basically without you having to do
7:48
thing basically without you having to do anything
7:48
anything
7:48
anything so here you see i found an issue at this
7:51
so here you see i found an issue at this
7:52
so here you see i found an issue at this time in day and what it's telling me is
7:53
time in day and what it's telling me is
7:53
time in day and what it's telling me is that
7:53
that
7:54
that there was an anomaly with the data
7:55
there was an anomaly with the data
7:55
there was an anomaly with the data across the dimensions of
7:57
across the dimensions of
7:57
across the dimensions of high value tickets that were of type
7:59
high value tickets that were of type
7:59
high value tickets that were of type unix that happened in new york and you
8:01
unix that happened in new york and you
8:01
unix that happened in new york and you can actually go in here
8:03
can actually go in here
8:03
can actually go in here and diagnose and see what type of
8:07
and diagnose and see what type of
8:07
and diagnose and see what type of contribution it has to the anomaly so
8:09
contribution it has to the anomaly so
8:09
contribution it has to the anomaly so for example in this case
8:10
for example in this case
8:10
for example in this case when you see this this delta ratio says
8:13
when you see this this delta ratio says
8:14
when you see this this delta ratio says that it's 63.84
8:17
that it's 63.84
8:17
that it's 63.84 above the expected value for this
8:20
above the expected value for this
8:20
above the expected value for this particular
8:21
particular
8:21
particular time series data and there's a ton of
8:22
time series data and there's a ton of
8:22
time series data and there's a ton of cool things that you can do with this
8:24
cool things that you can do with this
8:24
cool things that you can do with this for example
8:24
for example
8:24
for example you can drill into the metrics to see
8:27
you can drill into the metrics to see
8:27
you can drill into the metrics to see what how each of them contribute for
8:29
what how each of them contribute for
8:29
what how each of them contribute for example for ta
8:30
example for ta
8:30
example for ta you can see that here you can see where
8:31
you can see that here you can see where
8:31
you can see that here you can see where the anomaly was and then to finish up
8:34
the anomaly was and then to finish up
8:34
the anomaly was and then to finish up the coolest thing about this is that you
8:36
the coolest thing about this is that you
8:36
the coolest thing about this is that you can also do something
8:37
can also do something
8:38
can also do something called a hook so that anytime there is
8:40
called a hook so that anytime there is
8:40
called a hook so that anytime there is an incident
8:41
an incident
8:41
an incident for example an anomaly in this case it
8:44
for example an anomaly in this case it
8:44
for example an anomaly in this case it can
8:44
can
8:44
can alert you send a query to an api send
8:47
alert you send a query to an api send
8:48
alert you send a query to an api send you an email
8:48
you an email
8:48
you an email etc it's pretty amazing basically you
8:50
etc it's pretty amazing basically you
8:50
etc it's pretty amazing basically you start with data
8:51
start with data
8:51
start with data you hook it up and to metrics advisor
8:54
you hook it up and to metrics advisor
8:54
you hook it up and to metrics advisor and then you have
8:55
and then you have
8:55
and then you have immediate insight and alerting available
8:58
immediate insight and alerting available
8:58
immediate insight and alerting available to you
8:58
to you
8:58
to you hopefully that was short enough
9:02
hopefully that was short enough
9:02
hopefully that was short enough 63.84 that was very precise i thought
9:05
63.84 that was very precise i thought
9:05
63.84 that was very precise i thought that was an
9:06
that was an
9:06
that was an excellent uh description of it um yeah
9:09
excellent uh description of it um yeah
9:09
excellent uh description of it um yeah metrics advisor is great it's built on
9:11
metrics advisor is great it's built on
9:11
metrics advisor is great it's built on technology that we
9:12
technology that we
9:12
technology that we really developed internally at microsoft
9:14
really developed internally at microsoft
9:14
really developed internally at microsoft to monitor a lot of our own internal
9:16
to monitor a lot of our own internal
9:16
to monitor a lot of our own internal systems
9:16
systems
9:16
systems and a lot of the real-time monitoring
9:18
and a lot of the real-time monitoring
9:18
and a lot of the real-time monitoring for that and so just making that
9:20
for that and so just making that
9:20
for that and so just making that available
9:21
available
9:21
available um you know and packing up their metrics
9:23
um you know and packing up their metrics
9:23
um you know and packing up their metrics advisors so really excited to see that
9:25
advisors so really excited to see that
9:25
advisors so really excited to see that um the next thing i'd like to talk about
9:27
um the next thing i'd like to talk about
9:27
um the next thing i'd like to talk about though is you know given
9:28
though is you know given
9:28
though is you know given it's currently the kovit 19 era and
9:31
it's currently the kovit 19 era and
9:31
it's currently the kovit 19 era and everyone is
9:32
everyone is
9:32
everyone is you know wanting to make sure that they
9:33
you know wanting to make sure that they
9:33
you know wanting to make sure that they they know where they are in space and
9:35
they know where they are in space and
9:35
they know where they are in space and how they relate to other people
9:36
how they relate to other people
9:36
how they relate to other people uh another capability we announced at
9:38
uh another capability we announced at
9:38
uh another capability we announced at ignite is uh spatial analysis
9:40
ignite is uh spatial analysis
9:40
ignite is uh spatial analysis and so this is a computer vision model
9:43
and so this is a computer vision model
9:43
and so this is a computer vision model that really
9:44
that really
9:44
that really looks at the number of people in a
9:46
looks at the number of people in a
9:46
looks at the number of people in a particular room the distance between
9:47
particular room the distance between
9:48
particular room the distance between them whether they're going into
9:49
them whether they're going into
9:49
them whether they're going into particular spaces the dwell time that
9:51
particular spaces the dwell time that
9:51
particular spaces the dwell time that they're spending
9:52
they're spending
9:52
they're spending um and it's really helped in a number of
9:54
um and it's really helped in a number of
9:54
um and it's really helped in a number of places with uh
9:55
places with uh
9:55
places with uh with reopening and places that want to
9:57
with reopening and places that want to
9:57
with reopening and places that want to make sure that managing social
9:58
make sure that managing social
9:58
make sure that managing social distancing or mass wearing and things
10:00
distancing or mass wearing and things
10:00
distancing or mass wearing and things like that
10:02
like that
10:02
like that so really another exciting service
10:04
so really another exciting service
10:04
so really another exciting service that's really relevant
10:05
that's really relevant
10:05
that's really relevant based on you know the types of
10:06
based on you know the types of
10:06
based on you know the types of applications you can develop now
10:09
applications you can develop now
10:09
applications you can develop now um yeah we can go back to uh as a
10:11
um yeah we can go back to uh as a
10:11
um yeah we can go back to uh as a self-described
10:12
self-described
10:12
self-described demo monkey himself maybe you can walk
10:14
demo monkey himself maybe you can walk
10:14
demo monkey himself maybe you can walk us through a little bit more on what
10:15
us through a little bit more on what
10:16
us through a little bit more on what uh spatial analysis has to offer
10:18
uh spatial analysis has to offer
10:18
uh spatial analysis has to offer absolutely in this case
10:19
absolutely in this case
10:19
absolutely in this case instead of showing you like the
10:22
instead of showing you like the
10:22
instead of showing you like the underlying underlying part of it i
10:24
underlying underlying part of it i
10:24
underlying underlying part of it i wanted to show you what it actually
10:25
wanted to show you what it actually
10:25
wanted to show you what it actually does visually so there's three things
10:28
does visually so there's three things
10:28
does visually so there's three things that he mentioned the first one is
10:30
that he mentioned the first one is
10:30
that he mentioned the first one is person counting imagine you're in a
10:31
person counting imagine you're in a
10:31
person counting imagine you're in a situation where you have to limit the
10:33
situation where you have to limit the
10:33
situation where you have to limit the amount of people in a particular room
10:35
amount of people in a particular room
10:35
amount of people in a particular room this part of the service spatial
10:37
this part of the service spatial
10:37
this part of the service spatial analysis will tell you
10:39
analysis will tell you
10:39
analysis will tell you not only where the people are but notice
10:41
not only where the people are but notice
10:41
not only where the people are but notice that you can see the lines that it kind
10:42
that you can see the lines that it kind
10:42
that you can see the lines that it kind of tracks
10:44
of tracks
10:44
of tracks where they're going that's the first one
10:46
where they're going that's the first one
10:46
where they're going that's the first one so just how many people are in a room is
10:48
so just how many people are in a room is
10:48
so just how many people are in a room is awesome but what if we want to know
10:50
awesome but what if we want to know
10:50
awesome but what if we want to know how far away they are from each other
10:53
how far away they are from each other
10:53
how far away they are from each other that's the second
10:54
that's the second
10:54
that's the second thing it will tell you how far away from
10:56
thing it will tell you how far away from
10:56
thing it will tell you how far away from that and you can also set alerts
10:58
that and you can also set alerts
10:58
that and you can also set alerts to tell you when certain criteria are
11:01
to tell you when certain criteria are
11:01
to tell you when certain criteria are met so that's the second one so the
11:02
met so that's the second one so the
11:02
met so that's the second one so the first one is
11:03
first one is
11:03
first one is how many people are there the second one
11:05
how many people are there the second one
11:05
how many people are there the second one is how far away they are
11:07
is how far away they are
11:07
is how far away they are and then another one which is even more
11:08
and then another one which is even more
11:08
and then another one which is even more interesting is this notion of when
11:11
interesting is this notion of when
11:11
interesting is this notion of when people are in or outside of a polygon
11:13
people are in or outside of a polygon
11:14
people are in or outside of a polygon that you
11:14
that you
11:14
that you define which is even more specific so
11:17
define which is even more specific so
11:17
define which is even more specific so imagine being able to use
11:18
imagine being able to use
11:18
imagine being able to use these three in combination we have
11:20
these three in combination we have
11:20
these three in combination we have customers for example in new york rxr
11:22
customers for example in new york rxr
11:22
customers for example in new york rxr which is a very large company
11:25
which is a very large company
11:25
which is a very large company that handles manages buildings they now
11:27
that handles manages buildings they now
11:27
that handles manages buildings they now have this
11:28
have this
11:28
have this running on premises because obviously
11:30
running on premises because obviously
11:30
running on premises because obviously you don't want this information
11:31
you don't want this information
11:31
you don't want this information going out but they can run in a
11:33
going out but they can run in a
11:33
going out but they can run in a container uh spatial analysis
11:35
container uh spatial analysis
11:35
container uh spatial analysis and they know who's coming they know
11:37
and they know who's coming they know
11:37
and they know who's coming they know when people are coming in how many
11:38
when people are coming in how many
11:38
when people are coming in how many people are in there if they're following
11:40
people are in there if they're following
11:40
people are in there if they're following social distancing practices etc it's a
11:43
social distancing practices etc it's a
11:43
social distancing practices etc it's a pretty
11:44
pretty
11:44
pretty amazing service so again person counting
11:47
amazing service so again person counting
11:47
amazing service so again person counting how far
11:47
how far
11:47
how far away they are and whether they are in or
11:49
away they are and whether they are in or
11:49
away they are and whether they are in or out of a particular polygon imagine what
11:51
out of a particular polygon imagine what
11:51
out of a particular polygon imagine what you can do with all three together
11:53
you can do with all three together
11:53
you can do with all three together yeah it's really great and we've worked
11:54
yeah it's really great and we've worked
11:54
yeah it's really great and we've worked with a number of customers in a number
11:56
with a number of customers in a number
11:56
with a number of customers in a number of different
11:57
of different
11:57
of different uh you know fields as well even in
11:59
uh you know fields as well even in
11:59
uh you know fields as well even in manufacturing where they want to make
12:01
manufacturing where they want to make
12:01
manufacturing where they want to make sure people are staying out of
12:02
sure people are staying out of
12:02
sure people are staying out of particular areas or do they have hard
12:03
particular areas or do they have hard
12:03
particular areas or do they have hard hats on or the like and so
12:05
hats on or the like and so
12:05
hats on or the like and so lots of applications where being able to
12:07
lots of applications where being able to
12:07
lots of applications where being able to sort of get good analytics you know in
12:09
sort of get good analytics you know in
12:09
sort of get good analytics you know in retail i want to know how long are
12:10
retail i want to know how long are
12:10
retail i want to know how long are people spending in line and do i need
12:12
people spending in line and do i need
12:12
people spending in line and do i need more you know check out people or the
12:14
more you know check out people or the
12:14
more you know check out people or the like um
12:15
like um
12:15
like um just from the cameras in the that are
12:17
just from the cameras in the that are
12:17
just from the cameras in the that are already installed in many stores so lots
12:19
already installed in many stores so lots
12:19
already installed in many stores so lots of interesting things that can be done
12:20
of interesting things that can be done
12:20
of interesting things that can be done using spatial analysis
12:22
using spatial analysis
12:22
using spatial analysis um the next thing that the the third
12:24
um the next thing that the the third
12:24
um the next thing that the the third thing that i want to talk about
12:25
thing that i want to talk about
12:26
thing that i want to talk about is uh the azure machine learning
12:29
is uh the azure machine learning
12:29
is uh the azure machine learning um the designer so this is a capability
12:31
um the designer so this is a capability
12:31
um the designer so this is a capability that makes it really easy for people to
12:34
that makes it really easy for people to
12:34
that makes it really easy for people to uh develop uh sophisticated ai models
12:37
uh develop uh sophisticated ai models
12:37
uh develop uh sophisticated ai models and uh you know this is a capability
12:39
and uh you know this is a capability
12:39
and uh you know this is a capability that we just recently gave at ignite
12:41
that we just recently gave at ignite
12:41
that we just recently gave at ignite and uh you know again more than me
12:43
and uh you know again more than me
12:43
and uh you know again more than me talking about it i think designer is
12:45
talking about it i think designer is
12:45
talking about it i think designer is something that is
12:45
something that is
12:45
something that is is best seen and so uh why don't you
12:48
is best seen and so uh why don't you
12:48
is best seen and so uh why don't you take it away again
12:50
take it away again
12:50
take it away again absolutely so for those that don't know
12:51
absolutely so for those that don't know
12:51
absolutely so for those that don't know azure machine learning service is a
12:53
azure machine learning service is a
12:53
azure machine learning service is a comprehensive
12:54
comprehensive
12:54
comprehensive suite or studio is a comprehensive suite
12:57
suite or studio is a comprehensive suite
12:57
suite or studio is a comprehensive suite of amazing things that you can do in the
12:58
of amazing things that you can do in the
12:58
of amazing things that you can do in the enterprise with machine learning
12:59
enterprise with machine learning
12:59
enterprise with machine learning one of the cool things is this thing
13:01
one of the cool things is this thing
13:02
one of the cool things is this thing called the
13:03
called the
13:03
called the designer and what's cool about it is you
13:05
designer and what's cool about it is you
13:05
designer and what's cool about it is you can actually
13:06
can actually
13:06
can actually draw out draw out any particular
13:09
draw out draw out any particular
13:09
draw out draw out any particular complex scenario with machine learning
13:12
complex scenario with machine learning
13:12
complex scenario with machine learning for example
13:13
for example
13:13
for example like selecting columns in a data set
13:14
like selecting columns in a data set
13:14
like selecting columns in a data set clean missing data so the data portion
13:17
clean missing data so the data portion
13:17
clean missing data so the data portion and you can also train your models and
13:19
and you can also train your models and
13:19
and you can also train your models and score them
13:20
score them
13:20
score them from here you can also publish your
13:22
from here you can also publish your
13:22
from here you can also publish your models which is really cool and you have
13:23
models which is really cool and you have
13:24
models which is really cool and you have a ton of data sets that you can use to
13:25
a ton of data sets that you can use to
13:25
a ton of data sets that you can use to start
13:26
start
13:26
start your training with to figure it out and
13:28
your training with to figure it out and
13:28
your training with to figure it out and then you have a number of
13:29
then you have a number of
13:29
then you have a number of amazing tools like for example feature
13:31
amazing tools like for example feature
13:31
amazing tools like for example feature selection
13:32
selection
13:32
selection statistical functions to summarize data
13:34
statistical functions to summarize data
13:34
statistical functions to summarize data and then the thing that people are most
13:36
and then the thing that people are most
13:36
and then the thing that people are most interested in
13:37
interested in
13:37
interested in the machine learning algorithms that you
13:38
the machine learning algorithms that you
13:38
the machine learning algorithms that you can use to train these things
13:40
can use to train these things
13:40
can use to train these things so if you're just getting started with
13:42
so if you're just getting started with
13:42
so if you're just getting started with machine learning but you want to do
13:43
machine learning but you want to do
13:43
machine learning but you want to do something
13:43
something
13:43
something a little bit more advanced than calling
13:46
a little bit more advanced than calling
13:46
a little bit more advanced than calling cognitive services because you have a
13:47
cognitive services because you have a
13:47
cognitive services because you have a specialized
13:48
specialized
13:48
specialized thing that you want to do the designer
13:50
thing that you want to do the designer
13:50
thing that you want to do the designer is a great way to get started so you can
13:52
is a great way to get started so you can
13:52
is a great way to get started so you can graphically
13:53
graphically
13:53
graphically see what you're doing how you're doing
13:55
see what you're doing how you're doing
13:55
see what you're doing how you're doing it and you can also
13:56
it and you can also
13:56
it and you can also publish it and create pipelines for
13:59
publish it and create pipelines for
13:59
publish it and create pipelines for inferencing and training yeah i find
14:02
inferencing and training yeah i find
14:02
inferencing and training yeah i find this super useful and
14:03
this super useful and
14:03
this super useful and and that notion of pipelines too i mean
14:06
and that notion of pipelines too i mean
14:06
and that notion of pipelines too i mean that's what the machine learning
14:07
that's what the machine learning
14:07
that's what the machine learning workflow really is that it's a workflow
14:09
workflow really is that it's a workflow
14:09
workflow really is that it's a workflow a series of steps that the more you
14:11
a series of steps that the more you
14:11
a series of steps that the more you automate and the more you can sort of
14:12
automate and the more you can sort of
14:12
automate and the more you can sort of turn them into a pipeline
14:14
turn them into a pipeline
14:14
turn them into a pipeline they become repeatable which helps you
14:16
they become repeatable which helps you
14:16
they become repeatable which helps you with ml ops and that way i can sort of
14:17
with ml ops and that way i can sort of
14:17
with ml ops and that way i can sort of do the same thing over and over
14:19
do the same thing over and over
14:19
do the same thing over and over uh they become scriptable automatable
14:21
uh they become scriptable automatable
14:21
uh they become scriptable automatable and then you can sort of bash them and
14:23
and then you can sort of bash them and
14:23
and then you can sort of bash them and do multiple models at the same time
14:25
do multiple models at the same time
14:25
do multiple models at the same time there's a lot of capabilities that can
14:27
there's a lot of capabilities that can
14:27
there's a lot of capabilities that can really be lit up just by this simple
14:29
really be lit up just by this simple
14:29
really be lit up just by this simple expression of hey here's how my data
14:31
expression of hey here's how my data
14:31
expression of hey here's how my data flows and how i go and create my model
14:33
flows and how i go and create my model
14:33
flows and how i go and create my model so
14:33
so
14:33
so a lot of really great things so thanks
14:35
a lot of really great things so thanks
14:35
a lot of really great things so thanks for sharing that with us
15:00
safe to assume that no one else can hear
15:01
safe to assume that no one else can hear
15:01
safe to assume that no one else can hear willem as well
15:03
willem as well
15:03
willem as well yeah i'm not hearing as well maybe we
15:05
yeah i'm not hearing as well maybe we
15:05
yeah i'm not hearing as well maybe we should we should just keep doing demos
15:07
should we should just keep doing demos
15:07
should we should just keep doing demos because
15:08
because
15:08
because we can't oh very
15:12
we can't oh very
15:12
we can't oh very now i got you you're back now yeah
15:58
yeah it's another great area that we've
16:01
yeah it's another great area that we've
16:01
yeah it's another great area that we've uh we've really embraced that you know
16:03
uh we've really embraced that you know
16:03
uh we've really embraced that you know when we think about ai
16:04
when we think about ai
16:04
when we think about ai it brings a lot of advantages but it's
16:06
it brings a lot of advantages but it's
16:06
it brings a lot of advantages but it's up to us as the creators
16:30
know that those diagnoses are fair
16:32
know that those diagnoses are fair
16:32
know that those diagnoses are fair across
16:33
across
16:33
across you know races genders lots of different
16:35
you know races genders lots of different
16:35
you know races genders lots of different things like that
16:36
things like that
16:36
things like that and so what they use is toolkits that
16:37
and so what they use is toolkits that
16:37
and so what they use is toolkits that we've come out with things like fair
16:39
we've come out with things like fair
16:39
we've come out with things like fair learn and interpret ml
16:40
learn and interpret ml
16:40
learn and interpret ml which are directly integrated into
16:42
which are directly integrated into
16:42
which are directly integrated into machine learning to evaluate the
16:43
machine learning to evaluate the
16:43
machine learning to evaluate the different models that they have and
16:45
different models that they have and
16:45
different models that they have and really
16:46
really
16:46
really see going to work in the different
16:48
see going to work in the different
16:48
see going to work in the different environments across different categories
16:50
environments across different categories
16:50
environments across different categories of users and so
16:51
of users and so
16:51
of users and so uh you know this is something that you
16:53
uh you know this is something that you
16:53
uh you know this is something that you know not only do we make available to
16:55
know not only do we make available to
16:55
know not only do we make available to our customers but
16:56
our customers but
16:56
our customers but we practice it ourselves as we go and
16:58
we practice it ourselves as we go and
16:58
we practice it ourselves as we go and develop all of our own models
16:59
develop all of our own models
16:59
develop all of our own models we put a series of development practices
17:01
we put a series of development practices
17:01
we put a series of development practices in place really ensure
17:03
in place really ensure
17:03
in place really ensure that we feel confident that what we're
17:04
that we feel confident that what we're
17:04
that we feel confident that what we're developing is being done responsibly
17:10
yeah personally i'm a huge fan of the
17:12
yeah personally i'm a huge fan of the
17:12
yeah personally i'm a huge fan of the responsible ai and it turns out that
17:14
responsible ai and it turns out that
17:14
responsible ai and it turns out that microsoft we have six very
17:17
microsoft we have six very
17:17
microsoft we have six very specific principles so for example if
17:19
specific principles so for example if
17:19
specific principles so for example if you
17:20
you
17:20
you if you type in responsible ai you google
17:22
if you type in responsible ai you google
17:22
if you type in responsible ai you google with bing
17:23
with bing
17:23
with bing responsible ai microsoft right you will
17:25
responsible ai microsoft right you will
17:26
responsible ai microsoft right you will see like the
17:26
see like the
17:26
see like the we have responsible ai principles that
17:29
we have responsible ai principles that
17:29
we have responsible ai principles that are super important
17:30
are super important
17:30
are super important they're six they include fairness
17:32
they're six they include fairness
17:32
they're six they include fairness reliability and safety privacy and
17:34
reliability and safety privacy and
17:34
reliability and safety privacy and security inclusiveness
17:35
security inclusiveness
17:35
security inclusiveness transparency and obviously
17:37
transparency and obviously
17:37
transparency and obviously accountability and these are all
17:39
accountability and these are all
17:39
accountability and these are all super important just to add to what eric
17:40
super important just to add to what eric
17:40
super important just to add to what eric is saying we are super
17:42
is saying we are super
17:42
is saying we are super primed on doing this these are our ai
17:44
primed on doing this these are our ai
17:44
primed on doing this these are our ai principles uh willem can can you talk to
17:46
principles uh willem can can you talk to
17:46
principles uh willem can can you talk to us now
17:53
cool so we're gonna take a five minute
17:54
cool so we're gonna take a five minute
17:54
cool so we're gonna take a five minute break but thank you so much eric
17:57
break but thank you so much eric
17:57
break but thank you so much eric you have been amazing the demos were on
17:59
you have been amazing the demos were on
17:59
you have been amazing the demos were on point
18:00
point
18:00
point they were some of the best demos i think
18:01
they were some of the best demos i think
18:02
they were some of the best demos i think i've ever seen seth as always it's great
18:03
i've ever seen seth as always it's great
18:03
i've ever seen seth as always it's great to uh
18:04
to uh
18:04
to uh talk to you and and hey great to talk
18:06
talk to you and and hey great to talk
18:06
talk to you and and hey great to talk again to the global ai community it's
18:08
again to the global ai community it's
18:08
again to the global ai community it's uh you know while we had a couple of
18:09
uh you know while we had a couple of
18:09
uh you know while we had a couple of hiccups along the way it's always a
18:11
hiccups along the way it's always a
18:11
hiccups along the way it's always a great community for me to engage with so
18:13
great community for me to engage with so
18:13
great community for me to engage with so really really excited to be here thanks
18:14
really really excited to be here thanks
18:14
really really excited to be here thanks again
18:26
awesome
22:03
so like this this test
22:12
so like this this test
22:12
so like this this test test okay
22:15
test okay
22:15
test okay cool
22:31
let's take this test again test test
22:49
oh the noise is
23:17
so i'm hoping the sign interpreters can
23:18
so i'm hoping the sign interpreters can
23:18
so i'm hoping the sign interpreters can still understand
23:20
still understand
23:20
still understand uh yes i'm i can count backwards that
23:23
uh yes i'm i can count backwards that
23:23
uh yes i'm i can count backwards that sounds all right
23:27
at least my neural network is working
23:29
at least my neural network is working
23:29
at least my neural network is working still i guess
23:30
still i guess
23:30
still i guess we're coming back online soon
23:34
we're coming back online soon
23:34
we're coming back online soon and i fixed it and he didn't do anything
23:37
and i fixed it and he didn't do anything
23:37
and i fixed it and he didn't do anything and the good thing is
23:38
and the good thing is
23:38
and the good thing is hank fixed it he's with me in the studio
23:41
hank fixed it he's with me in the studio
23:41
hank fixed it he's with me in the studio today
23:42
today
23:42
today and he didn't do anything so it must be
23:44
and he didn't do anything so it must be
23:44
and he didn't do anything so it must be the ai
23:45
the ai
23:45
the ai no i turned teams all that over here oh
23:49
no i turned teams all that over here oh
23:49
no i turned teams all that over here oh this is a classic case of turning it off
23:51
this is a classic case of turning it off
23:51
this is a classic case of turning it off and on again
23:53
and on again
23:53
and on again that's even better
23:57
that's so cool and seth is back as well
23:59
that's so cool and seth is back as well
23:59
that's so cool and seth is back as well hey seth
24:00
hey seth
24:00
hey seth welcome back
24:08
yeah veronica is joining us pretty soon
24:16
hello
24:18
hello
24:18
hello i'll come back so we have this weird
24:21
i'll come back so we have this weird
24:21
i'll come back so we have this weird thing happening seth
24:23
thing happening seth
24:23
thing happening seth where we try turning it off and on again
24:24
where we try turning it off and on again
24:24
where we try turning it off and on again and it works
24:29
yeah it still works in 2020 and i'm so
24:33
yeah it still works in 2020 and i'm so
24:33
yeah it still works in 2020 and i'm so happy that it does that
24:35
happy that it does that
24:35
happy that it does that okay we're back online hey veronica
24:38
okay we're back online hey veronica
24:38
okay we're back online hey veronica welcome to
24:39
welcome to
24:39
welcome to welcome to the show
24:43
glad to have you so hi hey
24:47
glad to have you so hi hey
24:47
glad to have you so hi hey hi we do have sounds well we're back
24:50
hi we do have sounds well we're back
24:50
hi we do have sounds well we're back people
24:50
people
24:50
people and it's all working again we turned
24:53
and it's all working again we turned
24:53
and it's all working again we turned stuff off and on again
24:55
stuff off and on again
24:55
stuff off and on again and we're back back with veronica and
24:57
and we're back back with veronica and
24:57
and we're back back with veronica and she is going to talk to us about
24:59
she is going to talk to us about
24:59
she is going to talk to us about actually she's going to talk again about
25:02
actually she's going to talk again about
25:02
actually she's going to talk again about anomalies and
25:03
anomalies and
25:03
anomalies and the metrics advisory so um
25:07
the metrics advisory so um
25:07
the metrics advisory so um yeah so veronica what sort of work do
25:09
yeah so veronica what sort of work do
25:09
yeah so veronica what sort of work do you do during the work day
25:12
you do during the work day
25:12
you do during the work day so officially i'm a senior software
25:15
so officially i'm a senior software
25:15
so officially i'm a senior software engineer
25:16
engineer
25:16
engineer at liberty mutual in boston i am
25:19
at liberty mutual in boston i am
25:19
at liberty mutual in boston i am currently working from home
25:20
currently working from home
25:20
currently working from home and most of the people here
25:23
and most of the people here
25:23
and most of the people here um but uh i am
25:27
um but uh i am
25:27
um but uh i am and um mvp microsoft would be an ai
25:31
and um mvp microsoft would be an ai
25:31
and um mvp microsoft would be an ai so i do lots of machine learning and ai
25:34
so i do lots of machine learning and ai
25:34
so i do lots of machine learning and ai work and research on the side um
25:38
work and research on the side um
25:38
work and research on the side um so i'm super excited to share what i
25:41
so i'm super excited to share what i
25:41
so i'm super excited to share what i figure
25:41
figure
25:42
figure out cool cool
25:45
out cool cool
25:45
out cool cool that's impressive now veronica i i met
25:48
that's impressive now veronica i i met
25:48
that's impressive now veronica i i met you
25:49
you
25:49
you but like it was how many years ago it
25:52
but like it was how many years ago it
25:52
but like it was how many years ago it was i think i was in boston
25:53
was i think i was in boston
25:53
was i think i was in boston wasn't it yeah it was i think a couple
25:57
wasn't it yeah it was i think a couple
25:57
wasn't it yeah it was i think a couple of years ago
25:58
of years ago
25:58
of years ago um and i think scott got me was here
26:01
um and i think scott got me was here
26:01
um and i think scott got me was here with his uh red shirt tour
26:05
with his uh red shirt tour
26:05
with his uh red shirt tour and from there actually remember that
26:08
and from there actually remember that
26:08
and from there actually remember that you can
26:11
you can
26:11
you can yeah i mean that was a pretty cool uh
26:13
yeah i mean that was a pretty cool uh
26:13
yeah i mean that was a pretty cool uh event but here's a question for you
26:16
event but here's a question for you
26:16
event but here's a question for you tell us about your journey into getting
26:18
tell us about your journey into getting
26:18
tell us about your journey into getting into ai
26:19
into ai
26:19
into ai how did that happen what got you excited
26:21
how did that happen what got you excited
26:21
how did that happen what got you excited about it um
26:24
about it um
26:24
about it um it is an interesting story nothing
26:27
it is an interesting story nothing
26:27
it is an interesting story nothing crazy but i was working
26:31
crazy but i was working
26:31
crazy but i was working at university of massachusetts medical
26:33
at university of massachusetts medical
26:34
at university of massachusetts medical school
26:35
school
26:35
school several years ago and there we were
26:38
several years ago and there we were
26:38
several years ago and there we were trying to figure out
26:39
trying to figure out
26:40
trying to figure out a better solution for some of
26:43
a better solution for some of
26:43
a better solution for some of our issues and we had an app
26:46
our issues and we had an app
26:46
our issues and we had an app where it basically was an app
26:50
where it basically was an app
26:50
where it basically was an app for meditation where people
26:53
for meditation where people
26:53
for meditation where people needed to uh submit their
26:56
needed to uh submit their
26:56
needed to uh submit their moves before meditation and that
26:59
moves before meditation and that
26:59
moves before meditation and that meditation and it was just horrible
27:02
meditation and it was just horrible
27:02
meditation and it was just horrible experience
27:04
experience
27:04
experience it was just a drop down basically where
27:06
it was just a drop down basically where
27:06
it was just a drop down basically where if you choose okay
27:07
if you choose okay
27:07
if you choose okay i feel good i feel bad and when i saw
27:10
i feel good i feel bad and when i saw
27:10
i feel good i feel bad and when i saw the drop down i'm like okay
27:12
the drop down i'm like okay
27:12
the drop down i'm like okay my mood is down right away all right so
27:15
my mood is down right away all right so
27:16
my mood is down right away all right so we decided to figure out how we can
27:18
we decided to figure out how we can
27:18
we decided to figure out how we can improve it
27:19
improve it
27:19
improve it and that was my first interaction with
27:22
and that was my first interaction with
27:22
and that was my first interaction with cognitive services
27:24
cognitive services
27:24
cognitive services and i just fell in love with cognitive
27:27
and i just fell in love with cognitive
27:27
and i just fell in love with cognitive services i started learning more about
27:29
services i started learning more about
27:29
services i started learning more about them so
27:30
them so
27:30
them so we ended up using those content services
27:33
we ended up using those content services
27:33
we ended up using those content services some of the visual
27:34
some of the visual
27:34
some of the visual services i think at that time face api
27:38
services i think at that time face api
27:38
services i think at that time face api was separate
27:38
was separate
27:38
was separate so we used that one and then
27:42
so we used that one and then
27:42
so we used that one and then i was so excited so i started learning
27:44
i was so excited so i started learning
27:44
i was so excited so i started learning more and more about cognitive services
27:46
more and more about cognitive services
27:46
more and more about cognitive services and that's how i got
27:47
and that's how i got
27:47
and that's how i got into machine learning so now i am
27:51
into machine learning so now i am
27:51
into machine learning so now i am trying to build custom models using
27:53
trying to build custom models using
27:53
trying to build custom models using canal.net
27:54
canal.net
27:54
canal.net and other tools it's fun story
27:59
and other tools it's fun story
27:59
and other tools it's fun story that's really cool so basically you
28:01
that's really cool so basically you
28:01
that's really cool so basically you started with you wanted to actually
28:02
started with you wanted to actually
28:02
started with you wanted to actually solve a problem
28:04
solve a problem
28:04
solve a problem you found some cognitive services and
28:06
you found some cognitive services and
28:06
you found some cognitive services and then that led you down the path to
28:07
then that led you down the path to
28:07
then that led you down the path to building custom models so
28:09
building custom models so
28:09
building custom models so tell me about the transition between
28:11
tell me about the transition between
28:11
tell me about the transition between calling a service
28:12
calling a service
28:12
calling a service and then building your own model was
28:14
and then building your own model was
28:14
and then building your own model was that a was that a big jump is that
28:16
that a was that a big jump is that
28:16
that a was that a big jump is that something that all of us could do
28:19
um i think anyone can do that i don't
28:22
um i think anyone can do that i don't
28:22
um i think anyone can do that i don't have
28:22
have
28:22
have a phd in machine learning or in
28:26
a phd in machine learning or in
28:26
a phd in machine learning or in specific education i don't even have
28:29
specific education i don't even have
28:29
specific education i don't even have like any specific
28:30
like any specific
28:30
like any specific background in building machine learning
28:32
background in building machine learning
28:32
background in building machine learning models
28:34
models
28:34
models but uh starting with um pre-builds
28:37
but uh starting with um pre-builds
28:37
but uh starting with um pre-builds uh cognitive services and then use the
28:41
uh cognitive services and then use the
28:41
uh cognitive services and then use the custom
28:41
custom
28:41
custom quality of services and then move to
28:44
quality of services and then move to
28:44
quality of services and then move to something like ml.net since i'm a dotnet
28:48
something like ml.net since i'm a dotnet
28:48
something like ml.net since i'm a dotnet developer
28:49
developer
28:50
developer for me that transition wasn't
28:53
for me that transition wasn't
28:53
for me that transition wasn't super hard but definitely i
28:57
super hard but definitely i
28:57
super hard but definitely i had to read some books about machine
28:59
had to read some books about machine
28:59
had to read some books about machine learning and how
29:01
learning and how
29:01
learning and how um what techniques i supposed to use
29:03
um what techniques i supposed to use
29:03
um what techniques i supposed to use what tools available
29:05
what tools available
29:05
what tools available and uh what what is responsible ai
29:08
and uh what what is responsible ai
29:08
and uh what what is responsible ai and you know just lots of different
29:11
and you know just lots of different
29:11
and you know just lots of different stuff
29:13
stuff
29:13
stuff that's really cool so
29:16
that's really cool so
29:16
that's really cool so here's another question for you and
29:18
here's another question for you and
29:18
here's another question for you and hopefully you can hear me everyone was
29:19
hopefully you can hear me everyone was
29:19
hopefully you can hear me everyone was saying they could hear the air in my
29:20
saying they could hear the air in my
29:20
saying they could hear the air in my room i
29:21
room i
29:21
room i see this this sounds this sounds bad and
29:23
see this this sounds this sounds bad and
29:23
see this this sounds this sounds bad and now i'm fixing it so
29:25
now i'm fixing it so
29:25
now i'm fixing it so thank you for all i'm looking at the
29:26
thank you for all i'm looking at the
29:26
thank you for all i'm looking at the chat making sure um
29:29
chat making sure um
29:29
chat making sure um so the question i have for you is
29:32
so the question i have for you is
29:32
so the question i have for you is when you once you jump to building your
29:34
when you once you jump to building your
29:34
when you once you jump to building your own models
29:35
own models
29:35
own models when is a good time to like how
29:38
when is a good time to like how
29:38
when is a good time to like how do i know when i should build my own
29:40
do i know when i should build my own
29:40
do i know when i should build my own models versus whether i should just use
29:42
models versus whether i should just use
29:42
models versus whether i should just use a service
29:45
um i would recommend to check available
29:49
um i would recommend to check available
29:49
um i would recommend to check available services first since that's
29:52
services first since that's
29:52
services first since that's already pre-built for you by
29:56
already pre-built for you by
29:56
already pre-built for you by researchers and actually specialists who
29:59
researchers and actually specialists who
29:59
researchers and actually specialists who work on that all their life um
30:03
work on that all their life um
30:03
work on that all their life um and if you don't see something that is
30:05
and if you don't see something that is
30:05
and if you don't see something that is suitable for your specific project
30:08
suitable for your specific project
30:08
suitable for your specific project then you can start thinking about custom
30:10
then you can start thinking about custom
30:10
then you can start thinking about custom solution
30:11
solution
30:11
solution and then based on that solution or
30:14
and then based on that solution or
30:14
and then based on that solution or amount of data that you have
30:17
amount of data that you have
30:17
amount of data that you have you can figure out what tools you need
30:19
you can figure out what tools you need
30:19
you can figure out what tools you need to use
30:22
to use
30:22
to use cool and so here's my question to you uh
30:24
cool and so here's my question to you uh
30:24
cool and so here's my question to you uh when you when you started building your
30:26
when you when you started building your
30:26
when you when you started building your own model
30:27
own model
30:27
own model what was the thing that was hardest for
30:29
what was the thing that was hardest for
30:29
what was the thing that was hardest for you or like what what was something that
30:31
you or like what what was something that
30:31
you or like what what was something that you thought
30:32
you thought
30:32
you thought or it would like how would you help me
30:34
or it would like how would you help me
30:34
or it would like how would you help me if i was moving into that realm what was
30:36
if i was moving into that realm what was
30:36
if i was moving into that realm what was like one
30:37
like one
30:37
like one one tip you have for me um
30:41
one tip you have for me um
30:41
one tip you have for me um so definitely reading about best
30:45
so definitely reading about best
30:45
so definitely reading about best practices
30:45
practices
30:45
practices and building uh models in order to
30:48
and building uh models in order to
30:48
and building uh models in order to understand
30:49
understand
30:49
understand how that all actually works and then
30:52
how that all actually works and then
30:52
how that all actually works and then roi ml is awesome it does
30:56
roi ml is awesome it does
30:56
roi ml is awesome it does lots of heavy lifting for you and then
30:59
lots of heavy lifting for you and then
30:59
lots of heavy lifting for you and then you are
31:00
you are
31:00
you are getting really great models with less
31:04
getting really great models with less
31:04
getting really great models with less effort
31:07
that's cool looks like we have another
31:08
that's cool looks like we have another
31:08
that's cool looks like we have another guest willem
31:26
so yeah marian is here as well
31:30
so yeah marian is here as well
31:30
so yeah marian is here as well so maybe this is a good question for you
31:32
so maybe this is a good question for you
31:32
so maybe this is a good question for you marianne yeah i can try it
31:34
marianne yeah i can try it
31:34
marianne yeah i can try it out i think i can build up my prior
31:37
out i think i can build up my prior
31:37
out i think i can build up my prior prior thoughts that i think i'm really
31:41
prior thoughts that i think i'm really
31:41
prior thoughts that i think i'm really thinking about starters so um
31:50
um so but but always with a
31:54
um so but but always with a
31:54
um so but but always with a sight mark that if you really want to
31:57
sight mark that if you really want to
31:57
sight mark that if you really want to use the model then
31:58
use the model then
31:58
use the model then get in touch with somebody who really
31:59
get in touch with somebody who really
32:00
get in touch with somebody who really knows the algorithm behind it because i
32:01
knows the algorithm behind it because i
32:02
knows the algorithm behind it because i think it's always good to check
32:03
think it's always good to check
32:03
think it's always good to check but the nice thing is that without
32:05
but the nice thing is that without
32:05
but the nice thing is that without having
32:06
having
32:06
having too much knowledge you can you can
32:08
too much knowledge you can you can
32:08
too much knowledge you can you can directly start out of the box and we
32:10
directly start out of the box and we
32:10
directly start out of the box and we will see that also
32:11
will see that also
32:11
will see that also later on that basically using existing
32:14
later on that basically using existing
32:14
later on that basically using existing technology existing models
32:18
technology existing models
32:18
technology existing models and then it's very easy to start
32:21
and then it's very easy to start
32:21
and then it's very easy to start and i think as a developer i think it's
32:24
and i think as a developer i think it's
32:24
and i think as a developer i think it's also nice that
32:25
also nice that
32:26
also nice that that you can grow yourself so you can
32:28
that you can grow yourself so you can
32:28
that you can grow yourself so you can also start with
32:30
also start with
32:30
also start with using out-of-the-box stuff and then
32:33
using out-of-the-box stuff and then
32:33
using out-of-the-box stuff and then going to custom going to writing your
32:35
going to custom going to writing your
32:35
going to custom going to writing your own code
32:36
own code
32:36
own code improving things but yeah
32:39
improving things but yeah
32:40
improving things but yeah i would take it step by step i
32:43
i would take it step by step i
32:43
i would take it step by step i would recommend you to do some
32:54
statistics
33:10
exactly and just make sure that that
33:13
exactly and just make sure that that
33:13
exactly and just make sure that that that you have
33:14
that you have
33:14
that you have somebody inside i would say that can
33:16
somebody inside i would say that can
33:16
somebody inside i would say that can maybe help you a little bit
33:18
maybe help you a little bit
33:18
maybe help you a little bit just to be sure because um but
33:21
just to be sure because um but
33:21
just to be sure because um but the servers are getting better and
33:22
the servers are getting better and
33:22
the servers are getting better and better but at the end you have to know
33:25
better but at the end you have to know
33:25
better but at the end you have to know what you do i mean you cannot use
33:27
what you do i mean you cannot use
33:27
what you do i mean you cannot use everything for everything right
33:29
everything for everything right
33:29
everything for everything right so a little bit of advice normally helps
33:32
so a little bit of advice normally helps
33:32
so a little bit of advice normally helps and um i would say look for a nice
33:34
and um i would say look for a nice
33:34
and um i would say look for a nice mentor but just you know
33:35
mentor but just you know
33:35
mentor but just you know start building because i think that's
33:37
start building because i think that's
33:37
start building because i think that's the way you learn
33:39
the way you learn
33:39
the way you learn that's awesome so there's a couple
33:40
that's awesome so there's a couple
33:40
that's awesome so there's a couple questions in the chat that i want to
33:42
questions in the chat that i want to
33:42
questions in the chat that i want to throw out to you two experts that are
33:44
throw out to you two experts that are
33:44
throw out to you two experts that are here well all of you experts that are
33:45
here well all of you experts that are
33:45
here well all of you experts that are here
33:46
here
33:46
here the first one is from gothier as a
33:48
the first one is from gothier as a
33:48
the first one is from gothier as a beginner in ai
33:49
beginner in ai
33:49
beginner in ai is it useless to work with ai locally or
33:51
is it useless to work with ai locally or
33:51
is it useless to work with ai locally or should we use the cloud for big
33:53
should we use the cloud for big
33:53
should we use the cloud for big processing needed for ai that's the
33:55
processing needed for ai that's the
33:55
processing needed for ai that's the first one
33:57
first one
33:57
first one so could you repeat a question or can we
33:59
so could you repeat a question or can we
33:59
so could you repeat a question or can we read it somewhere
34:00
read it somewhere
34:00
read it somewhere absolutely so if as a beginner in ai is
34:03
absolutely so if as a beginner in ai is
34:03
absolutely so if as a beginner in ai is it useless to start
34:04
it useless to start
34:04
it useless to start training things locally or should i
34:07
training things locally or should i
34:07
training things locally or should i always just start with something in the
34:08
always just start with something in the
34:08
always just start with something in the cloud
34:10
cloud
34:10
cloud that depends
34:15
so what i mean what does it depend on i
34:16
so what i mean what does it depend on i
34:16
so what i mean what does it depend on i mean help me out well it depends
34:18
mean help me out well it depends
34:18
mean help me out well it depends first of all the question you would like
34:20
first of all the question you would like
34:20
first of all the question you would like to solve i would say
34:22
to solve i would say
34:22
to solve i would say and the amount of data you have
34:23
and the amount of data you have
34:24
and the amount of data you have available so if you have
34:25
available so if you have
34:25
available so if you have a small data set well um then the
34:28
a small data set well um then the
34:28
a small data set well um then the question is is it enough
34:30
question is is it enough
34:30
question is is it enough or do you just want to classify an image
34:33
or do you just want to classify an image
34:33
or do you just want to classify an image and you can use for example the custom
34:34
and you can use for example the custom
34:34
and you can use for example the custom services directly so you don't need any
34:37
services directly so you don't need any
34:37
services directly so you don't need any compute yourself actually
34:39
compute yourself actually
34:39
compute yourself actually so i think first define your question
34:42
so i think first define your question
34:42
so i think first define your question and
34:42
and
34:42
and look at what you what kind of data you
34:44
look at what you what kind of data you
34:44
look at what you what kind of data you have and
34:46
have and
34:46
have and then see whether you can use already
34:48
then see whether you can use already
34:48
then see whether you can use already pre-built
34:50
pre-built
34:50
pre-built um models or services and if not then i
34:53
um models or services and if not then i
34:53
um models or services and if not then i would say build it in the cloud because
34:55
would say build it in the cloud because
34:55
would say build it in the cloud because then you can scale it
34:57
then you can scale it
34:57
then you can scale it so that's that's awesome because i think
35:00
so that's that's awesome because i think
35:00
so that's that's awesome because i think if i'm understanding you right you're
35:01
if i'm understanding you right you're
35:01
if i'm understanding you right you're basically saying
35:03
basically saying
35:03
basically saying yes and no it just depends on what
35:05
yes and no it just depends on what
35:05
yes and no it just depends on what you're doing and i think like like what
35:07
you're doing and i think like like what
35:07
you're doing and i think like like what you said i think the key insight for me
35:09
you said i think the key insight for me
35:09
you said i think the key insight for me is i don't think people recognize that
35:11
is i don't think people recognize that
35:11
is i don't think people recognize that ai
35:12
ai
35:12
ai is just an alternative way of
35:13
is just an alternative way of
35:13
is just an alternative way of programming and so you have
35:15
programming and so you have
35:16
programming and so you have to start on the question i love the
35:17
to start on the question i love the
35:17
to start on the question i love the answer you know it depends on what
35:19
answer you know it depends on what
35:19
answer you know it depends on what you're trying to solve so thank you for
35:21
you're trying to solve so thank you for
35:21
you're trying to solve so thank you for that uh marianne let's go to the next
35:23
that uh marianne let's go to the next
35:23
that uh marianne let's go to the next question
35:24
question
35:24
question uh what articles or books do you
35:27
uh what articles or books do you
35:27
uh what articles or books do you recommend for a person
35:29
recommend for a person
35:29
recommend for a person starting to use ai and ml veronica and
35:32
starting to use ai and ml veronica and
35:32
starting to use ai and ml veronica and miriam
35:32
miriam
35:32
miriam those questions are for both of you i
35:35
those questions are for both of you i
35:35
those questions are for both of you i can try to answer this question
35:37
can try to answer this question
35:37
can try to answer this question i actually just read microsoft
35:39
i actually just read microsoft
35:40
i actually just read microsoft documentation
35:41
documentation
35:41
documentation for all the tools because that
35:43
for all the tools because that
35:43
for all the tools because that documentation
35:44
documentation
35:44
documentation is amazing it has examples
35:48
is amazing it has examples
35:48
is amazing it has examples also some samples
35:51
also some samples
35:51
also some samples in github right with everything that you
35:54
in github right with everything that you
35:54
in github right with everything that you need with
35:55
need with
35:55
need with data with all connected tools um
35:58
data with all connected tools um
35:58
data with all connected tools um if it's a cognitive service then you
36:01
if it's a cognitive service then you
36:01
if it's a cognitive service then you just need to grab your keys and end
36:03
just need to grab your keys and end
36:03
just need to grab your keys and end point
36:03
point
36:03
point and just go for it and start playing
36:06
and just go for it and start playing
36:06
and just go for it and start playing with it
36:08
with it
36:08
with it i also read just a couple of books
36:11
i also read just a couple of books
36:11
i also read just a couple of books i don't remember authors from the top of
36:14
i don't remember authors from the top of
36:14
i don't remember authors from the top of my hand
36:15
my hand
36:15
my hand but it was just general understanding of
36:18
but it was just general understanding of
36:18
but it was just general understanding of pushing learning
36:19
pushing learning
36:19
pushing learning and deep learning but i think there are
36:21
and deep learning but i think there are
36:21
and deep learning but i think there are lots of
36:22
lots of
36:22
lots of good online resources what about you
36:26
good online resources what about you
36:26
good online resources what about you yeah yeah i i i think a little bit
36:29
yeah yeah i i i think a little bit
36:29
yeah yeah i i i think a little bit different background so i have
36:31
different background so i have
36:31
different background so i have lots of books on statistics and research
36:33
lots of books on statistics and research
36:33
lots of books on statistics and research methodology
36:34
methodology
36:34
methodology uh and there's a lot out there and um
36:38
uh and there's a lot out there and um
36:38
uh and there's a lot out there and um let's say a starter course or a really
36:40
let's say a starter course or a really
36:40
let's say a starter course or a really good course it's not specifically for
36:42
good course it's not specifically for
36:42
good course it's not specifically for starters i think everyone who wants to
36:43
starters i think everyone who wants to
36:43
starters i think everyone who wants to start with ml
36:44
start with ml
36:44
start with ml um was from andrew ang on coursera
36:48
um was from andrew ang on coursera
36:48
um was from andrew ang on coursera i think that professor is really uh
36:51
i think that professor is really uh
36:51
i think that professor is really uh really great and you learn the basics
36:54
really great and you learn the basics
36:54
really great and you learn the basics it's well told
36:55
it's well told
36:55
it's well told absolutely recommended and as far as
36:59
absolutely recommended and as far as
36:59
absolutely recommended and as far as it goes through the services i think you
37:01
it goes through the services i think you
37:01
it goes through the services i think you we can all explore microsoft learn
37:03
we can all explore microsoft learn
37:03
we can all explore microsoft learn and we have the whole microsoft docs
37:06
and we have the whole microsoft docs
37:06
and we have the whole microsoft docs where you find
37:06
where you find
37:06
where you find loads of information loads of also
37:09
loads of information loads of also
37:09
loads of information loads of also workshop material you can do exercises
37:12
workshop material you can do exercises
37:12
workshop material you can do exercises um so that's sort of the applied
37:14
um so that's sort of the applied
37:14
um so that's sort of the applied knowledge
37:15
knowledge
37:16
knowledge so here's another question i think this
37:17
so here's another question i think this
37:17
so here's another question i think this follows along from what you all were
37:19
follows along from what you all were
37:20
follows along from what you all were saying
37:20
saying
37:20
saying like um i get this sense
37:23
like um i get this sense
37:24
like um i get this sense and and maybe i'm wrong but i see two
37:26
and and maybe i'm wrong but i see two
37:26
and and maybe i'm wrong but i see two camps when we talk about
37:27
camps when we talk about
37:27
camps when we talk about like learning actual machine learning
37:29
like learning actual machine learning
37:29
like learning actual machine learning not just calling cognitive services
37:30
not just calling cognitive services
37:30
not just calling cognitive services which is you should totally do because
37:32
which is you should totally do because
37:32
which is you should totally do because it's awesome
37:33
it's awesome
37:33
it's awesome some people say like we should start
37:36
some people say like we should start
37:36
some people say like we should start with
37:36
with
37:36
with linear algebra and calculus and others
37:39
linear algebra and calculus and others
37:39
linear algebra and calculus and others say we should start with let's just
37:41
say we should start with let's just
37:41
say we should start with let's just build a model
37:42
build a model
37:42
build a model and to me it's a bottom-up versus a
37:44
and to me it's a bottom-up versus a
37:44
and to me it's a bottom-up versus a top-down approach
37:46
top-down approach
37:46
top-down approach which one do you both prefer and why
37:49
which one do you both prefer and why
37:49
which one do you both prefer and why do you see my question i'll start with
37:51
do you see my question i'll start with
37:51
do you see my question i'll start with you marion
37:52
you marion
37:52
you marion yeah well yeah i start with the linear
37:55
yeah well yeah i start with the linear
37:55
yeah well yeah i start with the linear regression
37:56
regression
37:56
regression sorry for that and calculus um and why
37:59
sorry for that and calculus um and why
38:00
sorry for that and calculus um and why well um i think well again it depends
38:03
well um i think well again it depends
38:04
well um i think well again it depends because it depends on the type of
38:05
because it depends on the type of
38:05
because it depends on the type of learner you are and
38:07
learner you are and
38:07
learner you are and um i think for some students it's really
38:09
um i think for some students it's really
38:09
um i think for some students it's really good to get
38:10
good to get
38:10
good to get those basics and one important thing
38:13
those basics and one important thing
38:13
those basics and one important thing is that for example if you do a linear
38:16
is that for example if you do a linear
38:16
is that for example if you do a linear regression
38:17
regression
38:17
regression it starts you have rules and if you
38:19
it starts you have rules and if you
38:19
it starts you have rules and if you don't know the rules
38:20
don't know the rules
38:20
don't know the rules the thing is that if you start using a
38:22
the thing is that if you start using a
38:22
the thing is that if you start using a service directly you don't know
38:24
service directly you don't know
38:24
service directly you don't know which rules you have to apply now you
38:26
which rules you have to apply now you
38:26
which rules you have to apply now you can assume
38:27
can assume
38:27
can assume that all the services already cover
38:29
that all the services already cover
38:29
that all the services already cover those assumptions you
38:31
those assumptions you
38:31
those assumptions you you have but you're not sure so in that
38:34
you have but you're not sure so in that
38:34
you have but you're not sure so in that way
38:34
way
38:34
way it helps if you have some background on
38:37
it helps if you have some background on
38:37
it helps if you have some background on the other hand
38:39
the other hand
38:39
the other hand maybe that's just a step too far for
38:41
maybe that's just a step too far for
38:41
maybe that's just a step too far for some people so then i would say just
38:42
some people so then i would say just
38:42
some people so then i would say just build a model
38:43
build a model
38:43
build a model get inspired and do your regression
38:47
get inspired and do your regression
38:47
get inspired and do your regression basics later on
38:51
what do you think i definitely agree
38:53
what do you think i definitely agree
38:53
what do you think i definitely agree with
38:54
with
38:54
with marianne that's a really good plan
38:58
marianne that's a really good plan
38:58
marianne that's a really good plan it depends on what kind of learner you
39:01
it depends on what kind of learner you
39:01
it depends on what kind of learner you are
39:02
are
39:02
are i can't completely distance myself from
39:06
i can't completely distance myself from
39:06
i can't completely distance myself from linear algebra and statistics
39:09
linear algebra and statistics
39:09
linear algebra and statistics because i had those courses um in
39:12
because i had those courses um in
39:12
because i had those courses um in my undergrad because underground isn't
39:15
my undergrad because underground isn't
39:15
my undergrad because underground isn't computer science so we had those courses
39:18
computer science so we had those courses
39:18
computer science so we had those courses um so i kind of had that base level
39:22
um so i kind of had that base level
39:22
um so i kind of had that base level of understanding um before actually i
39:25
of understanding um before actually i
39:25
of understanding um before actually i started building
39:26
started building
39:26
started building machine learning models but for
39:29
machine learning models but for
39:29
machine learning models but for some people it's it can be also valuable
39:33
some people it's it can be also valuable
39:33
some people it's it can be also valuable to start with um automl and start
39:37
to start with um automl and start
39:37
to start with um automl and start building the
39:38
building the
39:38
building the models and then actually start
39:41
models and then actually start
39:41
models and then actually start understanding
39:42
understanding
39:42
understanding how they built after the fact
39:45
how they built after the fact
39:45
how they built after the fact that's cool so another question from the
39:47
that's cool so another question from the
39:47
that's cool so another question from the audience from george from zero knowledge
39:49
audience from george from zero knowledge
39:49
audience from george from zero knowledge to being able to have a good grasp of ai
39:51
to being able to have a good grasp of ai
39:51
to being able to have a good grasp of ai how long does something like that
39:53
how long does something like that
39:53
how long does something like that take zero knowledge to a grasp of ai
39:56
take zero knowledge to a grasp of ai
39:56
take zero knowledge to a grasp of ai i mean i have an opinion but i want to
39:58
i mean i have an opinion but i want to
39:58
i mean i have an opinion but i want to hear your expert opinion
40:01
hear your expert opinion
40:01
hear your expert opinion experience what's your definition of a
40:04
experience what's your definition of a
40:04
experience what's your definition of a grasp
40:06
grasp
40:06
grasp then that depends i guess it depends i
40:09
then that depends i guess it depends i
40:09
then that depends i guess it depends i mean yeah it depends
40:10
mean yeah it depends
40:10
mean yeah it depends it depends on the type of uh energy you
40:13
it depends on the type of uh energy you
40:13
it depends on the type of uh energy you put into it
40:14
put into it
40:14
put into it um yeah it it it really depends
40:19
um yeah it it it really depends
40:19
um yeah it it it really depends i i think it it i think there's sort of
40:21
i i think it it i think there's sort of
40:21
i i think it it i think there's sort of a norm that you need
40:22
a norm that you need
40:22
a norm that you need 10 000 hours to do something to to
40:25
10 000 hours to do something to to
40:25
10 000 hours to do something to to master a subject
40:26
master a subject
40:26
master a subject so might also work for ai
40:31
veronica what do you think zero
40:35
veronica what do you think zero
40:35
veronica what do you think zero i think as social engineers we keep
40:38
i think as social engineers we keep
40:38
i think as social engineers we keep learning all
40:39
learning all
40:39
learning all our lives so i can just put
40:42
our lives so i can just put
40:42
our lives so i can just put the deadline okay you spend two weeks
40:46
the deadline okay you spend two weeks
40:46
the deadline okay you spend two weeks and
40:46
and
40:46
and you'll be a supermaster machine learning
40:50
you'll be a supermaster machine learning
40:50
you'll be a supermaster machine learning because technology technologies are
40:52
because technology technologies are
40:52
because technology technologies are changing and everything is changing
40:55
changing and everything is changing
40:55
changing and everything is changing so rapidly and we need to
40:58
so rapidly and we need to
40:58
so rapidly and we need to keep up with it so it's just a constant
41:01
keep up with it so it's just a constant
41:01
keep up with it so it's just a constant learning process
41:03
learning process
41:03
learning process yeah so for me like i think
41:05
yeah so for me like i think
41:05
yeah so for me like i think understanding what ai is
41:06
understanding what ai is
41:06
understanding what ai is isn't actually hard making ai work is
41:09
isn't actually hard making ai work is
41:09
isn't actually hard making ai work is hard just like like if you know what a
41:11
hard just like like if you know what a
41:11
hard just like like if you know what a function is
41:13
function is
41:13
function is that's easy like you have a grasp of
41:14
that's easy like you have a grasp of
41:14
that's easy like you have a grasp of what a function is but writing functions
41:16
what a function is but writing functions
41:16
what a function is but writing functions to do certain tasks
41:17
to do certain tasks
41:17
to do certain tasks is it's definitely hard to me machine
41:19
is it's definitely hard to me machine
41:20
is it's definitely hard to me machine learning is a lazy way of writing
41:21
learning is a lazy way of writing
41:21
learning is a lazy way of writing functions with data
41:24
functions with data
41:24
functions with data and i feel like and now now you
41:27
and i feel like and now now you
41:27
and i feel like and now now you understand ai
41:28
understand ai
41:28
understand ai it's a lazy way of writing functions
41:30
it's a lazy way of writing functions
41:30
it's a lazy way of writing functions with data instead of code
41:32
with data instead of code
41:32
with data instead of code well there is some code but that's the
41:34
well there is some code but that's the
41:34
well there is some code but that's the way i i see it
41:36
way i i see it
41:36
way i i see it so it looks like your audio is back is
41:37
so it looks like your audio is back is
41:37
so it looks like your audio is back is that right
42:09
is this for one of us specifically or
42:11
is this for one of us specifically or
42:11
is this for one of us specifically or you just want an answer
42:13
you just want an answer
42:13
you just want an answer yeah well as as mentioned prior
42:16
yeah well as as mentioned prior
42:16
yeah well as as mentioned prior uh on microsoft docs you have all kind
42:19
uh on microsoft docs you have all kind
42:19
uh on microsoft docs you have all kind of exercises you can start with
42:21
of exercises you can start with
42:21
of exercises you can start with and also microsoft learn you get
42:23
and also microsoft learn you get
42:23
and also microsoft learn you get actually also the explanation
42:25
actually also the explanation
42:25
actually also the explanation and the exercises so you can get videos
42:28
and the exercises so you can get videos
42:28
and the exercises so you can get videos you
42:29
you
42:29
you you get the date and of course it's well
42:31
you get the date and of course it's well
42:31
you get the date and of course it's well normally it's linked also to github so
42:33
normally it's linked also to github so
42:33
normally it's linked also to github so there are a lot of
42:33
there are a lot of
42:33
there are a lot of as veronica already mentioned there are
42:36
as veronica already mentioned there are
42:36
as veronica already mentioned there are a lot of tutorials out there
42:38
a lot of tutorials out there
42:38
a lot of tutorials out there and um yeah and then from really easy
42:41
and um yeah and then from really easy
42:41
and um yeah and then from really easy ones
42:42
ones
42:42
ones and out of the box copy paste to more
42:44
and out of the box copy paste to more
42:44
and out of the box copy paste to more mature models and things
42:47
mature models and things
42:47
mature models and things veronica you mentioned that you started
42:48
veronica you mentioned that you started
42:48
veronica you mentioned that you started with ml.net
42:50
with ml.net
42:50
with ml.net looks like willem they can't hear hear
42:51
looks like willem they can't hear hear
42:51
looks like willem they can't hear hear you right now so i'm just gonna i'm
42:53
you right now so i'm just gonna i'm
42:53
you right now so i'm just gonna i'm gonna talk over you because they can't
42:54
gonna talk over you because they can't
42:54
gonna talk over you because they can't hear you anyway
42:55
hear you anyway
42:55
hear you anyway so tell us about ml.net is that is that
42:57
so tell us about ml.net is that is that
42:57
so tell us about ml.net is that is that the main
42:58
the main
42:58
the main framework that you use or are there
43:00
framework that you use or are there
43:00
framework that you use or are there others yeah
43:01
others yeah
43:02
others yeah um i'm mostly working with ml.net since
43:04
um i'm mostly working with ml.net since
43:04
um i'm mostly working with ml.net since i'm a
43:05
i'm a
43:05
i'm a net developer mostly i'm um
43:09
net developer mostly i'm um
43:09
net developer mostly i'm um i've been using csharp for many many
43:12
i've been using csharp for many many
43:12
i've been using csharp for many many years
43:13
years
43:13
years and it was really easy for me to start
43:17
and it was really easy for me to start
43:17
and it was really easy for me to start with my.net there are other options
43:20
with my.net there are other options
43:20
with my.net there are other options there
43:21
there
43:21
there tensorflow on x europe so many different
43:25
tensorflow on x europe so many different
43:25
tensorflow on x europe so many different tools
43:26
tools
43:26
tools but for now i prefer that
43:31
but for now i prefer that
43:31
but for now i prefer that how hard was it to pick up and actually
43:33
how hard was it to pick up and actually
43:33
how hard was it to pick up and actually get some models
43:34
get some models
43:34
get some models built for me it was
43:38
built for me it was
43:38
built for me it was pretty easy because there is that model
43:40
pretty easy because there is that model
43:40
pretty easy because there is that model builder
43:41
builder
43:41
builder availability in the visual studio
43:45
availability in the visual studio
43:46
availability in the visual studio now before you actually had to install
43:48
now before you actually had to install
43:48
now before you actually had to install it separately
43:49
it separately
43:49
it separately as an add-on and now it is integrated
43:53
as an add-on and now it is integrated
43:53
as an add-on and now it is integrated with visual studio so you don't need to
43:55
with visual studio so you don't need to
43:55
with visual studio so you don't need to installing it in you just need to
43:58
installing it in you just need to
43:58
installing it in you just need to open the model builder it will use other
44:01
open the model builder it will use other
44:01
open the model builder it will use other metal capabilities
44:03
metal capabilities
44:03
metal capabilities and then it will it's just a
44:07
and then it will it's just a
44:07
and then it will it's just a ui tool where you can just pick and
44:10
ui tool where you can just pick and
44:10
ui tool where you can just pick and choose what you want
44:11
choose what you want
44:11
choose what you want and it will train custom model
44:15
and it will train custom model
44:15
and it will train custom model using your custom data and then you can
44:18
using your custom data and then you can
44:18
using your custom data and then you can just
44:19
just
44:19
just use it in all kinds of projects that's
44:22
use it in all kinds of projects that's
44:22
use it in all kinds of projects that's cool and you've been able to get stuff
44:23
cool and you've been able to get stuff
44:24
cool and you've been able to get stuff working
44:24
working
44:24
working just using the model builder for example
44:27
just using the model builder for example
44:27
just using the model builder for example yeah definitely
44:28
yeah definitely
44:28
yeah definitely that's cool so now because we're about
44:30
that's cool so now because we're about
44:30
that's cool so now because we're about five minutes out before the next
44:32
five minutes out before the next
44:32
five minutes out before the next break i want to make sure that i asked
44:35
break i want to make sure that i asked
44:35
break i want to make sure that i asked you about this because
44:36
you about this because
44:36
you about this because i always come at ai from like uh
44:39
i always come at ai from like uh
44:39
i always come at ai from like uh i like building tools perspective but
44:41
i like building tools perspective but
44:41
i like building tools perspective but you all make ai that people actually use
44:45
you all make ai that people actually use
44:45
you all make ai that people actually use you know and that's the difference
44:46
you know and that's the difference
44:46
you know and that's the difference between me like i make demos and you
44:48
between me like i make demos and you
44:48
between me like i make demos and you make software
44:49
make software
44:49
make software what ethical concerns do you have or
44:52
what ethical concerns do you have or
44:52
what ethical concerns do you have or should people have
44:53
should people have
44:53
should people have when they are creating ai based models
44:57
when they are creating ai based models
44:57
when they are creating ai based models and using them in deployment we'll start
44:59
and using them in deployment we'll start
44:59
and using them in deployment we'll start with you
44:59
with you
44:59
with you marion to get us a sense for what
45:01
marion to get us a sense for what
45:01
marion to get us a sense for what ethical things we should have in mind
45:03
ethical things we should have in mind
45:03
ethical things we should have in mind when we build these things
45:04
when we build these things
45:04
when we build these things um yeah a lot actually because
45:07
um yeah a lot actually because
45:07
um yeah a lot actually because you make a lot of steps along let's say
45:10
you make a lot of steps along let's say
45:10
you make a lot of steps along let's say building your model
45:11
building your model
45:11
building your model and first of all i would say be aware of
45:14
and first of all i would say be aware of
45:14
and first of all i would say be aware of your sample
45:15
your sample
45:15
your sample so is your sample really representing
45:18
so is your sample really representing
45:18
so is your sample really representing your population
45:19
your population
45:19
your population and if it comes to people well of course
45:21
and if it comes to people well of course
45:21
and if it comes to people well of course you you can't just only use
45:23
you you can't just only use
45:23
you you can't just only use white male 50 plus people perfect that's
45:27
white male 50 plus people perfect that's
45:27
white male 50 plus people perfect that's not
45:28
not
45:28
not representing um you know
45:31
representing um you know
45:31
representing um you know the population i would say uh so be sure
45:34
the population i would say uh so be sure
45:34
the population i would say uh so be sure you make sure that your sample is
45:36
you make sure that your sample is
45:36
you make sure that your sample is correct right to start with
45:38
correct right to start with
45:38
correct right to start with and um after that as a
45:42
and um after that as a
45:42
and um after that as a researcher you you make steps right you
45:44
researcher you you make steps right you
45:44
researcher you you make steps right you just you make decisions
45:46
just you make decisions
45:46
just you make decisions so make sure you document every step and
45:48
so make sure you document every step and
45:48
so make sure you document every step and share it with another researcher
45:49
share it with another researcher
45:49
share it with another researcher so he or she can also validate
45:52
so he or she can also validate
45:52
so he or she can also validate what you've done because i think we're
45:54
what you've done because i think we're
45:54
what you've done because i think we're all in a way we're all biased
45:56
all in a way we're all biased
45:56
all in a way we're all biased so in order to get that bias out of it i
45:59
so in order to get that bias out of it i
45:59
so in order to get that bias out of it i think it's very good
46:00
think it's very good
46:00
think it's very good to to collaborate with others and and
46:02
to to collaborate with others and and
46:02
to to collaborate with others and and share your decisions actually
46:05
share your decisions actually
46:05
share your decisions actually so oh sorry go ahead mary i don't want
46:08
so oh sorry go ahead mary i don't want
46:08
so oh sorry go ahead mary i don't want to interrupt you
46:08
to interrupt you
46:08
to interrupt you no no no please go ahead veronica oh no
46:11
no no no please go ahead veronica oh no
46:11
no no no please go ahead veronica oh no you had a show and i was like oh man
46:13
you had a show and i was like oh man
46:13
you had a show and i was like oh man she's going to say
46:13
she's going to say
46:14
she's going to say something so smart and i i wanted to
46:16
something so smart and i i wanted to
46:16
something so smart and i i wanted to like rain down
46:17
like rain down
46:17
like rain down the knowledge to rain down on me miriam
46:20
the knowledge to rain down on me miriam
46:20
the knowledge to rain down on me miriam go ahead
46:22
go ahead
46:22
go ahead so collaboration i think that's good
46:24
so collaboration i think that's good
46:24
so collaboration i think that's good it's a tourist principle right
46:25
it's a tourist principle right
46:25
it's a tourist principle right um two people see more than just one
46:28
um two people see more than just one
46:28
um two people see more than just one person and also make sure that
46:30
person and also make sure that
46:30
person and also make sure that because normally a solution is not a
46:32
because normally a solution is not a
46:32
because normally a solution is not a one-man show or one woman's show
46:34
one-man show or one woman's show
46:34
one-man show or one woman's show right so also make sure that your team
46:36
right so also make sure that your team
46:36
right so also make sure that your team itself is diverse
46:38
itself is diverse
46:38
itself is diverse right because if you have a developer's
46:41
right because if you have a developer's
46:41
right because if you have a developer's team of only let's say
46:43
team of only let's say
46:43
team of only let's say one type of person it you have a
46:46
one type of person it you have a
46:46
one type of person it you have a probability that your model is
46:48
probability that your model is
46:48
probability that your model is also biased again
46:51
also biased again
46:51
also biased again so and of course from an ethical
46:53
so and of course from an ethical
46:53
so and of course from an ethical perspective
46:54
perspective
46:54
perspective i would be really concerned um about
46:58
i would be really concerned um about
46:58
i would be really concerned um about what variables you're using um what are
47:00
what variables you're using um what are
47:00
what variables you're using um what are you doing with age what are you doing
47:02
you doing with age what are you doing
47:02
you doing with age what are you doing with gender
47:03
with gender
47:03
with gender uh what are you doing with education
47:05
uh what are you doing with education
47:05
uh what are you doing with education level those kind of things
47:07
level those kind of things
47:07
level those kind of things are you setting people behind for
47:09
are you setting people behind for
47:09
are you setting people behind for example
47:10
example
47:10
example so it's sometimes it's so easy to
47:13
so it's sometimes it's so easy to
47:13
so it's sometimes it's so easy to include
47:13
include
47:13
include a lot of data but you have to be aware
47:15
a lot of data but you have to be aware
47:15
a lot of data but you have to be aware of what you're including or
47:17
of what you're including or
47:17
of what you're including or or social economic status of people are
47:20
or social economic status of people are
47:20
or social economic status of people are you allowed to do that so you have to
47:21
you allowed to do that so you have to
47:21
you allowed to do that so you have to talk about it
47:22
talk about it
47:22
talk about it i think and and make sure you write down
47:25
i think and and make sure you write down
47:25
i think and and make sure you write down every decision i would say because then
47:27
every decision i would say because then
47:27
every decision i would say because then you can share it and you can talk about
47:28
you can share it and you can talk about
47:28
you can share it and you can talk about it and you can improve it
47:31
it and you can improve it
47:31
it and you can improve it uh your thoughts veronica on ethical use
47:33
uh your thoughts veronica on ethical use
47:33
uh your thoughts veronica on ethical use of ai
47:35
of ai
47:35
of ai uh yeah that topic is really popular
47:38
uh yeah that topic is really popular
47:38
uh yeah that topic is really popular right now
47:38
right now
47:38
right now and i'm actually super happy about it we
47:41
and i'm actually super happy about it we
47:41
and i'm actually super happy about it we had a table talk discussion
47:43
had a table talk discussion
47:43
had a table talk discussion that i i had a chance to be and see at
47:47
that i i had a chance to be and see at
47:47
that i i had a chance to be and see at i was at ignite so
47:50
i was at ignite so
47:50
i was at ignite so we were discussing responsible ai
47:53
we were discussing responsible ai
47:53
we were discussing responsible ai and ethics in machine learning
47:57
and ethics in machine learning
47:57
and ethics in machine learning we had great conversation i
48:00
we had great conversation i
48:00
we had great conversation i personally learned a lot and it's
48:03
personally learned a lot and it's
48:03
personally learned a lot and it's actually
48:04
actually
48:04
actually going back to marion's point we need to
48:07
going back to marion's point we need to
48:08
going back to marion's point we need to invite
48:08
invite
48:08
invite different opinions people with different
48:10
different opinions people with different
48:10
different opinions people with different backgrounds
48:12
backgrounds
48:12
backgrounds um different age um so
48:15
um different age um so
48:15
um different age um so we can have input from all those people
48:17
we can have input from all those people
48:17
we can have input from all those people in order to understand better
48:20
in order to understand better
48:20
in order to understand better how ai should be created
48:23
how ai should be created
48:23
how ai should be created and how it should be processed better
48:27
and how it should be processed better
48:27
and how it should be processed better and also we need to remember and already
48:30
and also we need to remember and already
48:30
and also we need to remember and already mentioned that
48:31
mentioned that
48:31
mentioned that that ai machine learning it is
48:35
that ai machine learning it is
48:35
that ai machine learning it is still a software so you need to
48:37
still a software so you need to
48:37
still a software so you need to understand the whole
48:39
understand the whole
48:39
understand the whole process you can use
48:42
process you can use
48:42
process you can use mlx to deploy
48:46
mlx to deploy
48:46
mlx to deploy your models you need to have that
48:48
your models you need to have that
48:48
your models you need to have that failure log
48:49
failure log
48:50
failure log so you understand what worked and what
48:52
so you understand what worked and what
48:52
so you understand what worked and what their work
48:53
their work
48:53
their work and then based on that log you can
48:56
and then based on that log you can
48:56
and then based on that log you can actually
48:56
actually
48:56
actually retrain your model to make it better
49:00
retrain your model to make it better
49:00
retrain your model to make it better it is really important to um pay
49:02
it is really important to um pay
49:02
it is really important to um pay attention to
49:03
attention to
49:03
attention to those as well that's really cool the way
49:06
those as well that's really cool the way
49:06
those as well that's really cool the way you both like wrap this up really nicely
49:09
you both like wrap this up really nicely
49:09
you both like wrap this up really nicely for everyone
49:10
for everyone
49:10
for everyone because marion you talked about like
49:13
because marion you talked about like
49:13
because marion you talked about like creating the model
49:14
creating the model
49:14
creating the model ethically and veronica you wrapped it up
49:17
ethically and veronica you wrapped it up
49:17
ethically and veronica you wrapped it up with
49:18
with
49:18
with you know just because you made it
49:20
you know just because you made it
49:20
you know just because you made it ethically doesn't mean it can remain
49:22
ethically doesn't mean it can remain
49:22
ethically doesn't mean it can remain ethical
49:23
ethical
49:23
ethical and so you have to continue to do
49:25
and so you have to continue to do
49:25
and so you have to continue to do operations
49:26
operations
49:26
operations over it even after the models out to
49:28
over it even after the models out to
49:28
over it even after the models out to continue to improve uh
49:30
continue to improve uh
49:30
continue to improve uh the ethicality this is not a real word i
49:32
the ethicality this is not a real word i
49:32
the ethicality this is not a real word i wonder how they're going to sign in
49:33
wonder how they're going to sign in
49:33
wonder how they're going to sign in the ethicality of this of a model
49:37
the ethicality of this of a model
49:37
the ethicality of this of a model and so i love how you you focus on the
49:39
and so i love how you you focus on the
49:39
and so i love how you you focus on the beginning the process
49:40
beginning the process
49:40
beginning the process and then veronica focused on the end the
49:42
and then veronica focused on the end the
49:42
and then veronica focused on the end the development and the ml ops of it
49:44
development and the ml ops of it
49:44
development and the ml ops of it so we are i think we are done with this
49:47
so we are i think we are done with this
49:47
so we are i think we are done with this particular segment am i am i right on
49:49
particular segment am i am i right on
49:49
particular segment am i am i right on thinking that well i'm
49:50
thinking that well i'm
49:50
thinking that well i'm yeah so um thank you very much both of
49:53
yeah so um thank you very much both of
49:53
yeah so um thank you very much both of you for
49:54
you for
49:54
you for for answering our questions any
49:55
for answering our questions any
49:55
for answering our questions any questions from the audience and you can
49:57
questions from the audience and you can
49:57
questions from the audience and you can hear my audios back again
49:59
hear my audios back again
49:59
hear my audios back again let's hope it sticks this time
50:02
let's hope it sticks this time
50:02
let's hope it sticks this time thank you very much again and we will
50:04
thank you very much again and we will
50:04
thank you very much again and we will see marion back
50:05
see marion back
50:05
see marion back later in show to talk a little bit more
50:08
later in show to talk a little bit more
50:08
later in show to talk a little bit more about
50:09
about
50:09
about building custom ai solutions using
50:12
building custom ai solutions using
50:12
building custom ai solutions using custom computer vision
50:13
custom computer vision
50:13
custom computer vision but first i want to ask veronica to
50:17
but first i want to ask veronica to
50:17
but first i want to ask veronica to yeah can you tell us a little bit about
50:19
yeah can you tell us a little bit about
50:19
yeah can you tell us a little bit about uh anomaly detection i believe
50:21
uh anomaly detection i believe
50:21
uh anomaly detection i believe you've prepared something for us we had
50:24
you've prepared something for us we had
50:24
you've prepared something for us we had a little bit of a chat about
50:25
a little bit of a chat about
50:26
a little bit of a chat about it earlier this week and i'm i'm really
50:27
it earlier this week and i'm i'm really
50:28
it earlier this week and i'm i'm really really curious um
50:29
really curious um
50:29
really curious um about anomaly detection so um
50:32
about anomaly detection so um
50:32
about anomaly detection so um yeah let's let's go over to that topic
50:34
yeah let's let's go over to that topic
50:34
yeah let's let's go over to that topic and and see
50:35
and and see
50:35
and and see what people actually can do if they
50:37
what people actually can do if they
50:37
what people actually can do if they start using cognitive services in
50:39
start using cognitive services in
50:39
start using cognitive services in practice in their own projects
50:42
practice in their own projects
50:42
practice in their own projects yeah so um good question um
50:46
yeah so um good question um
50:46
yeah so um good question um it's a brand new category of
50:49
it's a brand new category of
50:49
it's a brand new category of quality services i think
50:53
quality services i think
50:53
quality services i think microsoft started talking about a
50:55
microsoft started talking about a
50:55
microsoft started talking about a decision category
50:56
decision category
50:56
decision category only last year and lots of services
50:59
only last year and lots of services
50:59
only last year and lots of services there
51:00
there
51:00
there were on preview and this year they
51:03
were on preview and this year they
51:03
were on preview and this year they actually went
51:04
actually went
51:04
actually went ga so it's really cool to um
51:08
ga so it's really cool to um
51:08
ga so it's really cool to um play around with um decision services
51:11
play around with um decision services
51:12
play around with um decision services like anomaly detector personalizer
51:15
like anomaly detector personalizer
51:15
like anomaly detector personalizer and then the new one
51:20
the anomaly monitor uh it is still on
51:23
the anomaly monitor uh it is still on
51:23
the anomaly monitor uh it is still on preview
51:24
preview
51:24
preview but um so before that um
51:28
but um so before that um
51:28
but um so before that um the the new one the new service that um
51:31
the the new one the new service that um
51:32
the the new one the new service that um they actually showcased at um
51:35
they actually showcased at um
51:35
they actually showcased at um um the keynote just
51:38
um the keynote just
51:38
um the keynote just a couple of minutes ago there was that
51:41
a couple of minutes ago there was that
51:41
a couple of minutes ago there was that anomaly detector service
51:43
anomaly detector service
51:43
anomaly detector service and i think it's the base of that new
51:46
and i think it's the base of that new
51:46
and i think it's the base of that new service
51:47
service
51:47
service so it's important to understand how it
51:50
so it's important to understand how it
51:50
so it's important to understand how it works
51:51
works
51:51
works and how we can figure out anomalies
51:55
and how we can figure out anomalies
51:55
and how we can figure out anomalies in our data whether we upload
51:59
in our data whether we upload
51:59
in our data whether we upload the whole data at once like a batch of
52:02
the whole data at once like a batch of
52:02
the whole data at once like a batch of data
52:04
data
52:04
data or we just keep uploading um
52:07
or we just keep uploading um
52:07
or we just keep uploading um and have that constant flow of data
52:11
and have that constant flow of data
52:11
and have that constant flow of data and then we constantly analyze it and
52:13
and then we constantly analyze it and
52:13
and then we constantly analyze it and using
52:14
using
52:14
using the anomaly detector and um
52:17
the anomaly detector and um
52:18
the anomaly detector and um also we can connect this to different
52:20
also we can connect this to different
52:20
also we can connect this to different notifications
52:22
notifications
52:22
notifications so it is it is just really important
52:25
so it is it is just really important
52:25
so it is it is just really important to understand um what
52:28
to understand um what
52:28
to understand um what those anomalies are and how we can use
52:32
those anomalies are and how we can use
52:32
those anomalies are and how we can use that information
52:34
that information
52:34
that information to prevent them in in the future
52:37
to prevent them in in the future
52:37
to prevent them in in the future so we're talking here about data that's
52:40
so we're talking here about data that's
52:40
so we're talking here about data that's not normal actually
52:41
not normal actually
52:42
not normal actually so in what situations would i actually
52:45
so in what situations would i actually
52:45
so in what situations would i actually use an anomaly detector
52:48
there is so many applications for that i
52:51
there is so many applications for that i
52:51
there is so many applications for that i think
52:51
think
52:51
think um probably the most straightforward one
52:56
um probably the most straightforward one
52:56
um probably the most straightforward one is analyzing um your
53:00
is analyzing um your
53:00
is analyzing um your data from your website like how many
53:02
data from your website like how many
53:02
data from your website like how many people if you have
53:03
people if you have
53:03
people if you have for example you have a blog and um
53:06
for example you have a blog and um
53:06
for example you have a blog and um you have that statistics of how many
53:09
you have that statistics of how many
53:09
you have that statistics of how many people
53:10
people
53:10
people actually um go and with it every day
53:14
actually um go and with it every day
53:14
actually um go and with it every day um you can see that one day uh
53:17
um you can see that one day uh
53:17
um you can see that one day uh maybe a triple a
53:21
maybe a triple a
53:21
maybe a triple a beautiful amount of people decided to
53:23
beautiful amount of people decided to
53:23
beautiful amount of people decided to check
53:24
check
53:24
check your blog and actually that
53:27
your blog and actually that
53:27
your blog and actually that is the possibility of a ddos attack
53:32
is the possibility of a ddos attack
53:32
is the possibility of a ddos attack but maybe not maybe you just posted a
53:35
but maybe not maybe you just posted a
53:35
but maybe not maybe you just posted a great blog post and everyone decided to
53:38
great blog post and everyone decided to
53:38
great blog post and everyone decided to check it out
53:40
check it out
53:40
check it out so um the anomaly detector will detect
53:43
so um the anomaly detector will detect
53:43
so um the anomaly detector will detect um that anomaly and it's up to you
53:47
um that anomaly and it's up to you
53:47
um that anomaly and it's up to you to understand what was the cause
53:50
to understand what was the cause
53:50
to understand what was the cause um but it is just that the basic example
53:53
um but it is just that the basic example
53:53
um but it is just that the basic example uh but
53:54
uh but
53:54
uh but if you are out there and in um
53:57
if you are out there and in um
53:58
if you are out there and in um in like factories and you are working
54:01
in like factories and you are working
54:01
in like factories and you are working with
54:01
with
54:01
with more serious data like
54:05
more serious data like
54:05
more serious data like some kind of statistical information
54:08
some kind of statistical information
54:08
some kind of statistical information from
54:08
from
54:08
from devices then it's really important to
54:11
devices then it's really important to
54:11
devices then it's really important to monitor
54:12
monitor
54:12
monitor and make sure that um nothing is
54:16
and make sure that um nothing is
54:16
and make sure that um nothing is broken and everything is working and
54:18
broken and everything is working and
54:18
broken and everything is working and it's supposed to work
54:20
it's supposed to work
54:20
it's supposed to work otherwise um it can be dangerous
54:23
otherwise um it can be dangerous
54:23
otherwise um it can be dangerous or it can break lots of things related
54:26
or it can break lots of things related
54:26
or it can break lots of things related to it
54:29
so can you show us uh how to actually
54:31
so can you show us uh how to actually
54:31
so can you show us uh how to actually use the anomaly detection surface in
54:34
use the anomaly detection surface in
54:34
use the anomaly detection surface in in azure to build an anomaly detector
54:38
in azure to build an anomaly detector
54:38
in azure to build an anomaly detector sure yeah let me share my screen
54:53
sure yeah let me share my screen
54:53
sure yeah let me share my screen okay i hope you can see it
54:57
it is just um basic demo
55:00
it is just um basic demo
55:00
it is just um basic demo um the data that i got it's in
55:04
um the data that i got it's in
55:04
um the data that i got it's in csv format i
55:07
csv format i
55:07
csv format i i got it from one of the samples
55:10
i got it from one of the samples
55:10
i got it from one of the samples on github microsoft samples
55:14
on github microsoft samples
55:14
on github microsoft samples so it's one of good places for
55:18
so it's one of good places for
55:18
so it's one of good places for test data if you want to check um
55:21
test data if you want to check um
55:21
test data if you want to check um the models or you want to check how
55:23
the models or you want to check how
55:23
the models or you want to check how services work
55:25
services work
55:25
services work um so i've got to have i have that csv
55:27
um so i've got to have i have that csv
55:28
um so i've got to have i have that csv file here
55:29
file here
55:29
file here and then actually that's at three part
55:31
and then actually that's at three part
55:32
and then actually that's at three part demo
55:33
demo
55:33
demo the first part here is batch anomaly
55:37
the first part here is batch anomaly
55:37
the first part here is batch anomaly detection
55:38
detection
55:38
detection so we are detected anomaly in the whole
55:42
so we are detected anomaly in the whole
55:42
so we are detected anomaly in the whole data set and understanding
55:46
data set and understanding
55:46
data set and understanding how what happens in when
55:49
how what happens in when
55:49
how what happens in when what kind of anomaly happened there and
55:52
what kind of anomaly happened there and
55:52
what kind of anomaly happened there and let me go
55:52
let me go
55:52
let me go to actually that method
56:15
yeah um maybe if you don't want to have
56:19
yeah um maybe if you don't want to have
56:19
yeah um maybe if you don't want to have that
56:19
that
56:20
that real if it's not that important to have
56:23
real if it's not that important to have
56:23
real if it's not that important to have real
56:24
real
56:24
real time data and you are gonna do that
56:26
time data and you are gonna do that
56:26
time data and you are gonna do that batching process
56:27
batching process
56:28
batching process where you combine data for a certain
56:30
where you combine data for a certain
56:30
where you combine data for a certain amount of time
56:31
amount of time
56:31
amount of time and then you upload that badge to that
56:34
and then you upload that badge to that
56:34
and then you upload that badge to that service
56:35
service
56:35
service and figure out if anything happened
56:37
and figure out if anything happened
56:38
and figure out if anything happened there
56:41
yeah so that's them it's it's pretty
56:44
yeah so that's them it's it's pretty
56:44
yeah so that's them it's it's pretty straightforward
56:45
straightforward
56:45
straightforward it's um just a console application that
56:48
it's um just a console application that
56:48
it's um just a console application that um so that's the main method i'm using
56:50
um so that's the main method i'm using
56:50
um so that's the main method i'm using here uh detect entire series
56:53
here uh detect entire series
56:53
here uh detect entire series um it's running asynchronously i'm
56:57
um it's running asynchronously i'm
56:57
um it's running asynchronously i'm basically sharing the whole data set
57:01
basically sharing the whole data set
57:01
basically sharing the whole data set with the service and then i
57:05
with the service and then i
57:05
with the service and then i am just checking if it contains an
57:07
am just checking if it contains an
57:07
am just checking if it contains an anomaly
57:08
anomaly
57:08
anomaly then we are returning the number
57:12
then we are returning the number
57:12
then we are returning the number of that i id of that anomaly
57:16
of that i id of that anomaly
57:16
of that i id of that anomaly we can return other information about
57:20
we can return other information about
57:20
we can return other information about what i have in the data set
57:21
what i have in the data set
57:21
what i have in the data set i just wanted to share so it's easier to
57:24
i just wanted to share so it's easier to
57:24
i just wanted to share so it's easier to understand i have that
57:26
understand i have that
57:26
understand i have that um time and date here
57:29
um time and date here
57:29
um time and date here and then um some kind of
57:34
some kind of number so it might be
57:37
some kind of number so it might be
57:37
some kind of number so it might be number of requests to your website
57:42
number of requests to your website
57:42
number of requests to your website it can be a number of
57:45
it can be a number of
57:45
it can be a number of users on your website so
57:48
users on your website so
57:48
users on your website so it can be different information
57:51
it can be different information
57:51
it can be different information available
57:52
available
57:52
available and then
58:05
it's better
58:08
it's better
58:08
it's better okay perfect and yeah so we are
58:12
okay perfect and yeah so we are
58:12
okay perfect and yeah so we are i'm basically just passing it to
58:15
i'm basically just passing it to
58:16
i'm basically just passing it to the service and then if we are detecting
58:19
the service and then if we are detecting
58:19
the service and then if we are detecting anomaly
58:20
anomaly
58:20
anomaly uh then we return that information i can
58:23
uh then we return that information i can
58:23
uh then we return that information i can run it really quickly
58:25
run it really quickly
58:25
run it really quickly because the data set is not that huge
58:28
because the data set is not that huge
58:28
because the data set is not that huge so it's going to be pretty fast
59:12
so it's already finished and we can see
59:14
so it's already finished and we can see
59:14
so it's already finished and we can see that there was several anomalies
59:17
that there was several anomalies
59:17
that there was several anomalies and those are ids we can
59:20
and those are ids we can
59:20
and those are ids we can return other information but it's just
59:23
return other information but it's just
59:23
return other information but it's just for that specific example i decided to
59:26
for that specific example i decided to
59:26
for that specific example i decided to return ids
59:30
and yeah that's that's that
59:34
yeah so um um so this detects anomalies
59:38
yeah so um um so this detects anomalies
59:38
yeah so um um so this detects anomalies um i'm getting back ideas of of anything
59:41
um i'm getting back ideas of of anything
59:41
um i'm getting back ideas of of anything that are interesting to me um can i
59:44
that are interesting to me um can i
59:44
that are interesting to me um can i blindly forward to a user or do i have
59:47
blindly forward to a user or do i have
59:47
blindly forward to a user or do i have to do an additional perform an
59:49
to do an additional perform an
59:49
to do an additional perform an additional step
59:50
additional step
59:50
additional step before i can actually uh send it off to
59:52
before i can actually uh send it off to
59:52
before i can actually uh send it off to the user telling him or her
59:54
the user telling him or her
59:54
the user telling him or her hey you've got some interesting data in
59:56
hey you've got some interesting data in
59:56
hey you've got some interesting data in here that uh that's probably
59:59
here that uh that's probably
59:59
here that uh that's probably maybe a hacker trying to break in or
1:00:01
maybe a hacker trying to break in or
1:00:01
maybe a hacker trying to break in or something like that
1:00:03
something like that
1:00:03
something like that yeah it's sort of up to you if your
1:00:05
yeah it's sort of up to you if your
1:00:05
yeah it's sort of up to you if your users
1:00:06
users
1:00:06
users understand those ideas if it's something
1:00:09
understand those ideas if it's something
1:00:09
understand those ideas if it's something they have access to then that's totally
1:00:11
they have access to then that's totally
1:00:11
they have access to then that's totally fine to has just those
1:00:13
fine to has just those
1:00:13
fine to has just those ids or um you can pass the
1:00:17
ids or um you can pass the
1:00:18
ids or um you can pass the date and time and they can go back to
1:00:21
date and time and they can go back to
1:00:21
date and time and they can go back to the data set and figure out
1:00:23
the data set and figure out
1:00:23
the data set and figure out what exactly i mean when exactly that
1:00:26
what exactly i mean when exactly that
1:00:26
what exactly i mean when exactly that anomaly happened
1:00:29
anomaly happened
1:00:29
anomaly happened and what could we do so for example if i
1:00:32
and what could we do so for example if i
1:00:32
and what could we do so for example if i were to have get these anomalies can i
1:00:35
were to have get these anomalies can i
1:00:35
were to have get these anomalies can i use something like azure functions maybe
1:00:37
use something like azure functions maybe
1:00:37
use something like azure functions maybe to do something with the data and
1:00:39
to do something with the data and
1:00:39
to do something with the data and send it off to the user yeah that's a
1:00:42
send it off to the user yeah that's a
1:00:42
send it off to the user yeah that's a good question
1:00:43
good question
1:00:43
good question you definitely can have other functions
1:00:45
you definitely can have other functions
1:00:45
you definitely can have other functions i here i have an example
1:00:48
i here i have an example
1:00:48
i here i have an example um just
1:00:51
um just
1:00:51
um just for this case um
1:00:54
for this case um
1:00:54
for this case um so i am basically creating
1:00:58
so i am basically creating
1:00:58
so i am basically creating an azure function i mean i created it on
1:01:01
an azure function i mean i created it on
1:01:02
an azure function i mean i created it on azure
1:01:03
azure
1:01:03
azure and then i connected it to um
1:01:06
and then i connected it to um
1:01:06
and then i connected it to um a service email service
1:01:10
a service email service
1:01:10
a service email service and then i am basically
1:01:13
and then i am basically
1:01:13
and then i am basically passing that anomaly that i detected
1:01:17
passing that anomaly that i detected
1:01:17
passing that anomaly that i detected so now it's just id but again it can be
1:01:20
so now it's just id but again it can be
1:01:20
so now it's just id but again it can be whatever is more convenient i'm passing
1:01:23
whatever is more convenient i'm passing
1:01:23
whatever is more convenient i'm passing it to the queue
1:01:24
it to the queue
1:01:24
it to the queue and then the function will
1:01:28
and then the function will
1:01:28
and then the function will send um the message
1:01:31
send um the message
1:01:31
send um the message i mean the function will pass it to the
1:01:33
i mean the function will pass it to the
1:01:33
i mean the function will pass it to the email
1:01:34
email
1:01:34
email service and an email service will send
1:01:36
service and an email service will send
1:01:36
service and an email service will send an email
1:01:38
an email
1:01:38
an email you can send all anomalies or you can
1:01:41
you can send all anomalies or you can
1:01:41
you can send all anomalies or you can filter them out in the code and then
1:01:44
filter them out in the code and then
1:01:44
filter them out in the code and then just send
1:01:45
just send
1:01:45
just send something once a day or twice a day
1:01:49
something once a day or twice a day
1:01:49
something once a day or twice a day something happens i can
1:01:53
something happens i can
1:01:53
something happens i can show you the setup in the
1:01:56
show you the setup in the
1:01:56
show you the setup in the azure portal so it's pretty
1:01:58
azure portal so it's pretty
1:01:58
azure portal so it's pretty straightforward
1:02:00
straightforward
1:02:00
straightforward that azure function here
1:02:03
that azure function here
1:02:03
that azure function here and then um they
1:02:07
and then um they
1:02:07
and then um they have um out of the box
1:02:10
have um out of the box
1:02:10
have um out of the box connection to um send grid
1:02:14
connection to um send grid
1:02:14
connection to um send grid so when you go and create a new function
1:02:18
so when you go and create a new function
1:02:18
so when you go and create a new function here um it can be a send grid type of
1:02:22
here um it can be a send grid type of
1:02:22
here um it can be a send grid type of function and it also
1:02:26
function and it also
1:02:26
function and it also uses so centritis is i guess one of the
1:02:29
uses so centritis is i guess one of the
1:02:29
uses so centritis is i guess one of the services that is exposed in azure
1:02:32
services that is exposed in azure
1:02:32
services that is exposed in azure functions that you can use are
1:02:34
functions that you can use are
1:02:34
functions that you can use are any other other options for submitting
1:02:36
any other other options for submitting
1:02:36
any other other options for submitting anomalies to users
1:02:38
anomalies to users
1:02:38
anomalies to users yeah definitely there are several
1:02:40
yeah definitely there are several
1:02:40
yeah definitely there are several options
1:02:42
options
1:02:42
options you want you can use azure functions you
1:02:45
you want you can use azure functions you
1:02:45
you want you can use azure functions you can also use
1:02:48
can also use
1:02:48
can also use you can use them a bad app maybe you can
1:02:52
you can use them a bad app maybe you can
1:02:52
you can use them a bad app maybe you can have an api and you send in your data
1:02:54
have an api and you send in your data
1:02:54
have an api and you send in your data there
1:02:55
there
1:02:55
there and then it can be connected to anything
1:02:57
and then it can be connected to anything
1:02:57
and then it can be connected to anything you want
1:02:58
you want
1:02:58
you want um there are also i know signal r
1:03:02
um there are also i know signal r
1:03:02
um there are also i know signal r um is used for those pop-up
1:03:05
um is used for those pop-up
1:03:05
um is used for those pop-up notifications
1:03:06
notifications
1:03:06
notifications for um mobile solutions
1:03:10
for um mobile solutions
1:03:10
for um mobile solutions or um you can send a message using
1:03:13
or um you can send a message using
1:03:13
or um you can send a message using twilio
1:03:14
twilio
1:03:14
twilio actually centigrade grid is part of
1:03:17
actually centigrade grid is part of
1:03:17
actually centigrade grid is part of twilio too
1:03:18
twilio too
1:03:18
twilio too so some grid is for emails and then they
1:03:20
so some grid is for emails and then they
1:03:20
so some grid is for emails and then they have another option
1:03:21
have another option
1:03:21
have another option for um messages so you can get messages
1:03:25
for um messages so you can get messages
1:03:25
for um messages so you can get messages on your phone
1:03:26
on your phone
1:03:26
on your phone cool it sounds really really awesome so
1:03:28
cool it sounds really really awesome so
1:03:28
cool it sounds really really awesome so people
1:03:30
people
1:03:30
people can can just just grab this anomaly
1:03:33
can can just just grab this anomaly
1:03:33
can can just just grab this anomaly detection
1:03:33
detection
1:03:33
detection surface then connect it to
1:03:37
surface then connect it to
1:03:37
surface then connect it to up to azure functions and we you shortly
1:03:40
up to azure functions and we you shortly
1:03:40
up to azure functions and we you shortly mentioned the
1:03:41
mentioned the
1:03:41
mentioned the streaming bit where you can use
1:03:44
streaming bit where you can use
1:03:44
streaming bit where you can use just a single data point how does that
1:03:46
just a single data point how does that
1:03:46
just a single data point how does that actually work
1:03:48
actually work
1:03:48
actually work yeah so um here i have another
1:03:52
yeah so um here i have another
1:03:52
yeah so um here i have another function you can use last
1:03:56
function you can use last
1:03:56
function you can use last detected data points and to check
1:03:59
detected data points and to check
1:04:00
detected data points and to check if it was anomaly or not
1:04:03
if it was anomaly or not
1:04:03
if it was anomaly or not so basically you are comparing it
1:04:07
so basically you are comparing it
1:04:07
so basically you are comparing it to a previous data set that you have
1:04:11
to a previous data set that you have
1:04:11
to a previous data set that you have and see if it was anomaly or not and
1:04:13
and see if it was anomaly or not and
1:04:14
and see if it was anomaly or not and then return that information
1:04:16
then return that information
1:04:16
then return that information to users so you can either again
1:04:19
to users so you can either again
1:04:19
to users so you can either again use email or pop-up or
1:04:23
use email or pop-up or
1:04:23
use email or pop-up or message or anything you want
1:04:26
message or anything you want
1:04:26
message or anything you want that's really cool and i think it's
1:04:27
that's really cool and i think it's
1:04:27
that's really cool and i think it's especially useful in cases where we have
1:04:29
especially useful in cases where we have
1:04:29
especially useful in cases where we have iot data
1:04:30
iot data
1:04:30
iot data metrics coming from devices those kind
1:04:33
metrics coming from devices those kind
1:04:33
metrics coming from devices those kind of things that's that's the
1:04:34
of things that's that's the
1:04:34
of things that's that's the scenario here right yeah
1:04:38
scenario here right yeah
1:04:38
scenario here right yeah cool cool sounds really cool and what
1:04:41
cool cool sounds really cool and what
1:04:41
cool cool sounds really cool and what other options do i have once i have this
1:04:43
other options do i have once i have this
1:04:43
other options do i have once i have this anomaly detection
1:04:44
anomaly detection
1:04:44
anomaly detection um i've heard eric talk about the
1:04:47
um i've heard eric talk about the
1:04:47
um i've heard eric talk about the metrics advisor does that does that play
1:04:49
metrics advisor does that does that play
1:04:49
metrics advisor does that does that play any role in this
1:04:51
any role in this
1:04:51
any role in this anomaly detection scenario
1:04:54
anomaly detection scenario
1:04:54
anomaly detection scenario so metrics advisor i feel it's the next
1:04:58
so metrics advisor i feel it's the next
1:04:58
so metrics advisor i feel it's the next level
1:04:58
level
1:04:58
level of the anomaly detector
1:05:01
of the anomaly detector
1:05:02
of the anomaly detector because metrics advisor not only detects
1:05:04
because metrics advisor not only detects
1:05:04
because metrics advisor not only detects the anomaly
1:05:05
the anomaly
1:05:05
the anomaly it can also help you figure out
1:05:08
it can also help you figure out
1:05:08
it can also help you figure out what was the cause of this anomaly
1:05:12
what was the cause of this anomaly
1:05:12
what was the cause of this anomaly um so it's a little more advanced
1:05:15
um so it's a little more advanced
1:05:15
um so it's a little more advanced but if you just want to detect that
1:05:18
but if you just want to detect that
1:05:18
but if you just want to detect that anomaly
1:05:18
anomaly
1:05:18
anomaly and maybe it's something that you
1:05:22
and maybe it's something that you
1:05:22
and maybe it's something that you worked with you work with that iot
1:05:24
worked with you work with that iot
1:05:24
worked with you work with that iot solution or the data
1:05:27
solution or the data
1:05:27
solution or the data for quite some time and you kind of know
1:05:29
for quite some time and you kind of know
1:05:29
for quite some time and you kind of know what might be
1:05:30
what might be
1:05:30
what might be um the reason of failure
1:05:33
um the reason of failure
1:05:33
um the reason of failure and you just need to figure out when it
1:05:35
and you just need to figure out when it
1:05:35
and you just need to figure out when it happened in order to
1:05:36
happened in order to
1:05:36
happened in order to understand yourself then anomaly
1:05:38
understand yourself then anomaly
1:05:38
understand yourself then anomaly detector
1:05:39
detector
1:05:39
detector is also a good option to use
1:05:42
is also a good option to use
1:05:42
is also a good option to use um you you you have that
1:05:46
um you you you have that
1:05:46
um you you you have that functionality that you need and you
1:05:49
functionality that you need and you
1:05:49
functionality that you need and you don't
1:05:49
don't
1:05:50
don't need to use something that you don't
1:05:52
need to use something that you don't
1:05:52
need to use something that you don't need
1:05:54
need
1:05:54
need yeah exactly and do you know anything
1:05:56
yeah exactly and do you know anything
1:05:56
yeah exactly and do you know anything about the pricing
1:05:57
about the pricing
1:05:57
about the pricing of building a solution uh using
1:06:01
of building a solution uh using
1:06:01
of building a solution uh using this cognitive service this anomaly
1:06:03
this cognitive service this anomaly
1:06:03
this cognitive service this anomaly detector and
1:06:04
detector and
1:06:04
detector and maybe azure functions um
1:06:07
maybe azure functions um
1:06:07
maybe azure functions um that's a good question unfortunately i
1:06:10
that's a good question unfortunately i
1:06:10
that's a good question unfortunately i can tell you the
1:06:11
can tell you the
1:06:11
can tell you the end price because it depends on what
1:06:13
end price because it depends on what
1:06:13
end price because it depends on what kind of data you have
1:06:14
kind of data you have
1:06:14
kind of data you have the amount of data how often do you use
1:06:18
the amount of data how often do you use
1:06:18
the amount of data how often do you use it
1:06:19
it
1:06:19
it so it will vary
1:06:22
so it will vary
1:06:22
so it will vary depending on different factors but they
1:06:25
depending on different factors but they
1:06:25
depending on different factors but they have
1:06:26
have
1:06:26
have a really good calculus calculators
1:06:30
a really good calculus calculators
1:06:30
a really good calculus calculators on azure where they can
1:06:33
on azure where they can
1:06:33
on azure where they can help you to kind of understand
1:06:36
help you to kind of understand
1:06:36
help you to kind of understand how much on average you will be spending
1:06:39
how much on average you will be spending
1:06:39
how much on average you will be spending when you use
1:06:40
when you use
1:06:40
when you use all those services okay yeah that sounds
1:06:43
all those services okay yeah that sounds
1:06:43
all those services okay yeah that sounds like a good idea for people that are
1:06:45
like a good idea for people that are
1:06:45
like a good idea for people that are trying to build a solution with this
1:06:47
trying to build a solution with this
1:06:47
trying to build a solution with this actually i was also wondering
1:06:49
actually i was also wondering
1:06:49
actually i was also wondering is there any free tier available of this
1:06:51
is there any free tier available of this
1:06:51
is there any free tier available of this service i know there's a lot of free
1:06:53
service i know there's a lot of free
1:06:53
service i know there's a lot of free tier
1:06:53
tier
1:06:53
tier options for other services like computer
1:06:55
options for other services like computer
1:06:55
options for other services like computer vision text analysis
1:06:56
vision text analysis
1:06:56
vision text analysis is it also available for this one um
1:07:00
is it also available for this one um
1:07:00
is it also available for this one um i know they have um the trial version so
1:07:03
i know they have um the trial version so
1:07:03
i know they have um the trial version so even if you don't have azure
1:07:06
even if you don't have azure
1:07:06
even if you don't have azure subscription
1:07:07
subscription
1:07:07
subscription and you can have um seven day trial
1:07:10
and you can have um seven day trial
1:07:10
and you can have um seven day trial there
1:07:11
there
1:07:11
there just to check it out and see if it works
1:07:14
just to check it out and see if it works
1:07:14
just to check it out and see if it works for you
1:07:14
for you
1:07:14
for you or it doesn't um then for
1:07:18
or it doesn't um then for
1:07:18
or it doesn't um then for the metrics advisor they have
1:07:21
the metrics advisor they have
1:07:21
the metrics advisor they have a demo website where you can play with
1:07:23
a demo website where you can play with
1:07:24
a demo website where you can play with it and
1:07:24
it and
1:07:24
it and figure out if you want to use it or not
1:07:29
figure out if you want to use it or not
1:07:29
figure out if you want to use it or not there is almost
1:07:33
there is almost
1:07:33
there is almost almost free tier here so
1:07:36
almost free tier here so
1:07:36
almost free tier here so you don't pay that much if you decide to
1:07:39
you don't pay that much if you decide to
1:07:39
you don't pay that much if you decide to use that service
1:07:41
use that service
1:07:41
use that service but it is not completely free because
1:07:43
but it is not completely free because
1:07:43
but it is not completely free because there are
1:07:44
there are
1:07:44
there are um other posts associated
1:07:47
um other posts associated
1:07:47
um other posts associated with using that kind of resource
1:07:51
with using that kind of resource
1:07:51
with using that kind of resource okay yeah yeah that's good to know yeah
1:07:53
okay yeah yeah that's good to know yeah
1:07:53
okay yeah yeah that's good to know yeah so there's a question from the audience
1:07:55
so there's a question from the audience
1:07:55
so there's a question from the audience and i think you will like this one
1:07:57
and i think you will like this one
1:07:57
and i think you will like this one actually
1:07:58
actually
1:07:58
actually what if someone forces fake anomalies to
1:08:01
what if someone forces fake anomalies to
1:08:01
what if someone forces fake anomalies to confuse you
1:08:02
confuse you
1:08:02
confuse you or the ai what do you do then
1:08:06
yeah that's a good question so
1:08:09
yeah that's a good question so
1:08:10
yeah that's a good question so you can um set up your logic
1:08:13
you can um set up your logic
1:08:13
you can um set up your logic of your uh end product
1:08:16
of your uh end product
1:08:16
of your uh end product that it will figure out
1:08:20
that it will figure out
1:08:20
that it will figure out or you can set up actually the logic
1:08:23
or you can set up actually the logic
1:08:23
or you can set up actually the logic that way
1:08:24
that way
1:08:24
that way that if you know that okay maybe you
1:08:27
that if you know that okay maybe you
1:08:27
that if you know that okay maybe you wrote
1:08:27
wrote
1:08:27
wrote amazing blog posts and you know that
1:08:30
amazing blog posts and you know that
1:08:30
amazing blog posts and you know that lots of people
1:08:31
lots of people
1:08:31
lots of people will show up in your website to actually
1:08:36
will show up in your website to actually
1:08:36
will show up in your website to actually read it then you can um
1:08:40
read it then you can um
1:08:40
read it then you can um you can call you can cover
1:08:43
you can call you can cover
1:08:43
you can call you can cover that time frame um so
1:08:46
that time frame um so
1:08:46
that time frame um so you basically kind of suspend all the
1:08:49
you basically kind of suspend all the
1:08:50
you basically kind of suspend all the notifications
1:08:51
notifications
1:08:51
notifications or you are
1:08:55
or you are
1:08:55
or you are just skipping this period of time
1:08:58
just skipping this period of time
1:08:58
just skipping this period of time in your data and that way
1:09:01
in your data and that way
1:09:01
in your data and that way you won't get a
1:09:05
you won't get a
1:09:05
you won't get a put on these four comments that you
1:09:08
put on these four comments that you
1:09:08
put on these four comments that you probably know about
1:09:10
probably know about
1:09:10
probably know about so that's definitely possible yeah so so
1:09:13
so that's definitely possible yeah so so
1:09:13
so that's definitely possible yeah so so if seth
1:09:14
if seth
1:09:14
if seth decides to retweet your blog post and
1:09:17
decides to retweet your blog post and
1:09:17
decides to retweet your blog post and share it online
1:09:18
share it online
1:09:18
share it online and you get a ton of people on your
1:09:21
and you get a ton of people on your
1:09:21
and you get a ton of people on your website you can basically say okay
1:09:23
website you can basically say okay
1:09:23
website you can basically say okay hold on steph is doing something special
1:09:25
hold on steph is doing something special
1:09:25
hold on steph is doing something special here i'm disabling
1:09:27
here i'm disabling
1:09:27
here i'm disabling the detection temporary and i will
1:09:28
the detection temporary and i will
1:09:28
the detection temporary and i will continue after things have died down
1:09:30
continue after things have died down
1:09:30
continue after things have died down again
1:09:31
again
1:09:31
again and i know nothing special is happening
1:09:33
and i know nothing special is happening
1:09:33
and i know nothing special is happening that's basically the option that you
1:09:35
that's basically the option that you
1:09:35
that's basically the option that you have
1:09:36
have
1:09:36
have yeah yeah exactly well it sounds really
1:09:39
yeah yeah exactly well it sounds really
1:09:39
yeah yeah exactly well it sounds really useful
1:09:39
useful
1:09:39
useful i have to say um and so
1:09:43
i have to say um and so
1:09:43
i have to say um and so i i know we've talked about this before
1:09:46
i i know we've talked about this before
1:09:46
i i know we've talked about this before someone is asking
1:09:47
someone is asking
1:09:47
someone is asking um did you ever implement something like
1:09:50
um did you ever implement something like
1:09:50
um did you ever implement something like this anomaly detection
1:09:52
this anomaly detection
1:09:52
this anomaly detection in a case of cyber security in your
1:09:54
in a case of cyber security in your
1:09:54
in a case of cyber security in your daily work
1:09:57
daily work
1:09:57
daily work so unfortunately i
1:10:00
so unfortunately i
1:10:00
so unfortunately i never use it for my day job
1:10:05
never use it for my day job
1:10:06
never use it for my day job because we have lots of third-party
1:10:08
because we have lots of third-party
1:10:08
because we have lots of third-party tools um
1:10:09
tools um
1:10:09
tools um i am mostly doing back-and-forth
1:10:12
i am mostly doing back-and-forth
1:10:12
i am mostly doing back-and-forth development
1:10:12
development
1:10:12
development and there are lots of out of great out
1:10:15
and there are lots of out of great out
1:10:16
and there are lots of out of great out of the box tools
1:10:17
of the box tools
1:10:17
of the box tools um including app inside if
1:10:20
um including app inside if
1:10:20
um including app inside if um you are creating your map application
1:10:24
um you are creating your map application
1:10:24
um you are creating your map application using azure um or other tools
1:10:27
using azure um or other tools
1:10:28
using azure um or other tools that can monitor it so um it is not that
1:10:31
that can monitor it so um it is not that
1:10:31
that can monitor it so um it is not that critical to create a custom solution
1:10:34
critical to create a custom solution
1:10:34
critical to create a custom solution but for i think more serious
1:10:38
but for i think more serious
1:10:38
but for i think more serious uh applications i don't want to say that
1:10:42
uh applications i don't want to say that
1:10:42
uh applications i don't want to say that that is not serious
1:10:43
that is not serious
1:10:44
that is not serious i feel like factory
1:10:47
i feel like factory
1:10:47
i feel like factory machinery and iot can be
1:10:52
machinery and iot can be
1:10:52
machinery and iot can be can bring more damage if it doesn't work
1:10:54
can bring more damage if it doesn't work
1:10:54
can bring more damage if it doesn't work correctly
1:10:57
correctly
1:10:57
correctly yeah so so if i understand you correctly
1:10:59
yeah so so if i understand you correctly
1:10:59
yeah so so if i understand you correctly the the anomaly detector that's
1:11:01
the the anomaly detector that's
1:11:01
the the anomaly detector that's readily made in the shape of this
1:11:04
readily made in the shape of this
1:11:04
readily made in the shape of this cognitive surface is great to start
1:11:07
cognitive surface is great to start
1:11:07
cognitive surface is great to start but once you get in a very specific
1:11:09
but once you get in a very specific
1:11:09
but once you get in a very specific domain for example
1:11:10
domain for example
1:11:10
domain for example an iron smeltery or all
1:11:13
an iron smeltery or all
1:11:13
an iron smeltery or all oil refinery maybe those kind of
1:11:15
oil refinery maybe those kind of
1:11:16
oil refinery maybe those kind of applications
1:11:17
applications
1:11:17
applications it's better to start looking into the
1:11:19
it's better to start looking into the
1:11:19
it's better to start looking into the custom models that are
1:11:20
custom models that are
1:11:20
custom models that are available on azure as well in the azure
1:11:23
available on azure as well in the azure
1:11:23
available on azure as well in the azure ml service
1:11:25
ml service
1:11:25
ml service yeah yeah definitely okay
1:11:28
yeah yeah definitely okay
1:11:28
yeah yeah definitely okay yeah that's that's that's actually
1:11:30
yeah that's that's that's actually
1:11:30
yeah that's that's that's actually pretty interesting um
1:11:32
pretty interesting um
1:11:32
pretty interesting um so do you have any um additional tips or
1:11:36
so do you have any um additional tips or
1:11:36
so do you have any um additional tips or tricks for people who want to get
1:11:38
tricks for people who want to get
1:11:38
tricks for people who want to get started with this where can they find
1:11:39
started with this where can they find
1:11:39
started with this where can they find the resources
1:11:42
the resources
1:11:42
the resources yeah so they can actually
1:11:46
yeah so they can actually
1:11:46
yeah so they can actually check out them creative services portal
1:11:50
check out them creative services portal
1:11:50
check out them creative services portal you just google and bing
1:11:54
you just google and bing
1:11:54
you just google and bing and um then you will see um
1:11:57
and um then you will see um
1:11:57
and um then you will see um lots of articles lots of um useful
1:12:00
lots of articles lots of um useful
1:12:00
lots of articles lots of um useful examples and then also from the
1:12:03
examples and then also from the
1:12:03
examples and then also from the documentation
1:12:05
documentation
1:12:05
documentation they provide links to
1:12:08
they provide links to
1:12:08
they provide links to github samples and there they have
1:12:12
github samples and there they have
1:12:12
github samples and there they have the data they have implemented
1:12:16
the data they have implemented
1:12:16
the data they have implemented cognitive services so you can see how
1:12:18
cognitive services so you can see how
1:12:18
cognitive services so you can see how they work
1:12:19
they work
1:12:19
they work or even on that calling services portal
1:12:23
or even on that calling services portal
1:12:23
or even on that calling services portal you can have um samples
1:12:27
you can have um samples
1:12:27
you can have um samples right on the web page you can play with
1:12:29
right on the web page you can play with
1:12:29
right on the web page you can play with it with those samples to understand how
1:12:31
it with those samples to understand how
1:12:32
it with those samples to understand how they all work
1:12:33
they all work
1:12:34
they all work for matrix advisor i know there is
1:12:38
for matrix advisor i know there is
1:12:38
for matrix advisor i know there is a demo portal where you can
1:12:42
a demo portal where you can
1:12:42
a demo portal where you can test it and check it out so
1:12:45
test it and check it out so
1:12:45
test it and check it out so lots of good options to check it out and
1:12:48
lots of good options to check it out and
1:12:48
lots of good options to check it out and start playing with it
1:12:50
start playing with it
1:12:50
start playing with it cool cool thank you very much um so
1:12:54
cool cool thank you very much um so
1:12:54
cool cool thank you very much um so if there are any more questions um that
1:12:57
if there are any more questions um that
1:12:57
if there are any more questions um that you would like to ask
1:12:58
you would like to ask
1:12:58
you would like to ask veronica she will she will stick around
1:13:01
veronica she will she will stick around
1:13:01
veronica she will she will stick around i guess in the live chat
1:13:03
i guess in the live chat
1:13:03
i guess in the live chat and then try to answer a few more
1:13:04
and then try to answer a few more
1:13:04
and then try to answer a few more questions and just remember if you
1:13:07
questions and just remember if you
1:13:07
questions and just remember if you uh tweeted to us live using the hashtag
1:13:10
uh tweeted to us live using the hashtag
1:13:10
uh tweeted to us live using the hashtag g
1:13:10
g
1:13:10
g sharp corner and global ai community you
1:13:12
sharp corner and global ai community you
1:13:12
sharp corner and global ai community you get a chance to win a
1:13:13
get a chance to win a
1:13:13
get a chance to win a gift card 50 that's that's quite a large
1:13:16
gift card 50 that's that's quite a large
1:13:16
gift card 50 that's that's quite a large amount
1:13:17
amount
1:13:17
amount i don't know if you can get azure
1:13:18
i don't know if you can get azure
1:13:18
i don't know if you can get azure credits for that do you know seth
1:13:21
credits for that do you know seth
1:13:21
credits for that do you know seth do they accept gift cards for credits
1:13:24
do they accept gift cards for credits
1:13:24
do they accept gift cards for credits that's a really good question i don't
1:13:27
that's a really good question i don't
1:13:27
that's a really good question i don't know if
1:13:28
know if
1:13:28
know if we accept amazon gift cards
1:13:34
yeah but i mean that would be awesome if
1:13:36
yeah but i mean that would be awesome if
1:13:36
yeah but i mean that would be awesome if we did i gotta figure that out
1:13:38
we did i gotta figure that out
1:13:38
we did i gotta figure that out yeah yeah that would be actually great
1:13:40
yeah yeah that would be actually great
1:13:40
yeah yeah that would be actually great for people so if um
1:13:42
for people so if um
1:13:42
for people so if um also if you would like to try out any of
1:13:44
also if you would like to try out any of
1:13:44
also if you would like to try out any of the demos
1:13:45
the demos
1:13:45
the demos and you need some azure credits
1:13:47
and you need some azure credits
1:13:47
and you need some azure credits microsoft actually actually
1:13:48
microsoft actually actually
1:13:48
microsoft actually actually offers a free trial subscription on
1:13:51
offers a free trial subscription on
1:13:51
offers a free trial subscription on azure that gives you
1:13:52
azure that gives you
1:13:52
azure that gives you a small amount of money to get started
1:13:54
a small amount of money to get started
1:13:54
a small amount of money to get started and
1:13:55
and
1:13:55
and some of the services like for example
1:13:57
some of the services like for example
1:13:57
some of the services like for example azure functions
1:13:59
azure functions
1:13:59
azure functions are free up until a certain amount of
1:14:01
are free up until a certain amount of
1:14:01
are free up until a certain amount of calls so that's good to know as well
1:14:03
calls so that's good to know as well
1:14:03
calls so that's good to know as well if you want to get started with that um
1:14:05
if you want to get started with that um
1:14:05
if you want to get started with that um so
1:14:06
so
1:14:06
so um we're going off for a short break i
1:14:09
um we're going off for a short break i
1:14:09
um we're going off for a short break i guess
1:14:10
guess
1:14:10
guess to refresh our machines here in our
1:14:13
to refresh our machines here in our
1:14:13
to refresh our machines here in our drinks
1:14:14
drinks
1:14:14
drinks and we'll get back to you soon in a
1:14:16
and we'll get back to you soon in a
1:14:16
and we'll get back to you soon in a couple of minutes
1:15:12
[Music]
1:15:20
[Applause]
1:15:21
[Applause]
1:15:21
[Applause] [Music]
1:16:29
[Music]
1:16:46
[Music]
1:16:53
[Music]
1:17:52
[Music]
1:18:12
[Music]
1:18:19
[Music]
1:19:03
[Music]
1:19:19
[Music]
1:19:48
[Music]
1:19:48
[Music] so
1:19:50
so
1:19:50
so [Music]
1:20:05
[Music]
1:21:26
[Music]
1:21:26
[Music] do
1:21:36
[Music]
1:21:55
[Music]
1:21:55
[Music] [Applause]
1:21:56
[Applause]
1:21:56
[Applause] [Music]
1:22:36
do
1:22:43
because here william yeah
1:22:46
because here william yeah
1:22:46
because here william yeah [Music]
1:22:49
[Music]
1:22:49
[Music] yeah oh we're back
1:22:52
yeah oh we're back
1:22:52
yeah oh we're back think positive fog thoughts people and
1:22:54
think positive fog thoughts people and
1:22:54
think positive fog thoughts people and send us positive tweets
1:22:56
send us positive tweets
1:22:56
send us positive tweets because if you do and you add the c
1:22:58
because if you do and you add the c
1:22:58
because if you do and you add the c sharp corner and the global ai community
1:23:00
sharp corner and the global ai community
1:23:00
sharp corner and the global ai community hashtag to it
1:23:01
hashtag to it
1:23:01
hashtag to it you can win 50 gift card and um
1:23:04
you can win 50 gift card and um
1:23:04
you can win 50 gift card and um with that thank you marion for for
1:23:07
with that thank you marion for for
1:23:07
with that thank you marion for for joining us again
1:23:08
joining us again
1:23:08
joining us again and hello again yeah you never left did
1:23:11
and hello again yeah you never left did
1:23:12
and hello again yeah you never left did you
1:23:12
you
1:23:12
you no no no it was too interesting
1:23:16
no no no it was too interesting
1:23:16
no no no it was too interesting exactly yeah yeah so seth
1:23:20
exactly yeah yeah so seth
1:23:20
exactly yeah yeah so seth i'm i'm happy you're here as well um
1:23:23
i'm i'm happy you're here as well um
1:23:23
i'm i'm happy you're here as well um without you i don't know how this
1:23:25
without you i don't know how this
1:23:25
without you i don't know how this episode would ever happen
1:23:28
episode would ever happen
1:23:28
episode would ever happen probably could have been awesome
1:23:33
probably could have been awesome
1:23:33
probably could have been awesome but less goofy i am all about goofiness
1:23:35
but less goofy i am all about goofiness
1:23:35
but less goofy i am all about goofiness primarily because
1:23:38
primarily because
1:23:38
primarily because what other way is there to be and they
1:23:40
what other way is there to be and they
1:23:40
what other way is there to be and they told me marion
1:23:41
told me marion
1:23:41
told me marion that i have a half hour to talk to you
1:23:44
that i have a half hour to talk to you
1:23:44
that i have a half hour to talk to you oh wow
1:23:45
oh wow
1:23:45
oh wow building models with computer vision and
1:23:48
building models with computer vision and
1:23:48
building models with computer vision and as you know or may not know
1:23:50
as you know or may not know
1:23:50
as you know or may not know i am a huge fan of computer vision i
1:23:53
i am a huge fan of computer vision i
1:23:53
i am a huge fan of computer vision i started with nlp
1:23:53
started with nlp
1:23:54
started with nlp and then moved on to computer vision so
1:23:57
and then moved on to computer vision so
1:23:57
and then moved on to computer vision so for those that don't know what computer
1:23:58
for those that don't know what computer
1:23:58
for those that don't know what computer vision
1:23:59
vision
1:23:59
vision is why don't you describe the problem
1:24:02
is why don't you describe the problem
1:24:02
is why don't you describe the problem and then
1:24:02
and then
1:24:02
and then give us a sense for how you do it
1:24:06
give us a sense for how you do it
1:24:06
give us a sense for how you do it all right well um maybe if you're a
1:24:09
all right well um maybe if you're a
1:24:09
all right well um maybe if you're a parent right you have kids
1:24:10
parent right you have kids
1:24:10
parent right you have kids and they want a specific kind of two uh
1:24:14
and they want a specific kind of two uh
1:24:14
and they want a specific kind of two uh toys
1:24:15
toys
1:24:15
toys right and you cannot keep up with all
1:24:17
right and you cannot keep up with all
1:24:17
right and you cannot keep up with all their
1:24:18
their
1:24:18
their main characters and favorite characters
1:24:20
main characters and favorite characters
1:24:20
main characters and favorite characters they want to have so
1:24:22
they want to have so
1:24:22
they want to have so uh what if you could just use pictures
1:24:24
uh what if you could just use pictures
1:24:24
uh what if you could just use pictures and the model tells you all right
1:24:26
and the model tells you all right
1:24:26
and the model tells you all right this is character one or b or c or
1:24:28
this is character one or b or c or
1:24:28
this is character one or b or c or whatever um
1:24:30
whatever um
1:24:30
whatever um so basically it's to detect things
1:24:33
so basically it's to detect things
1:24:33
so basically it's to detect things and in our case uh thanks to hank
1:24:36
and in our case uh thanks to hank
1:24:36
and in our case uh thanks to hank bulman who made us a really nice data
1:24:39
bulman who made us a really nice data
1:24:39
bulman who made us a really nice data set with
1:24:40
set with
1:24:40
set with the simpsons lego actually i thought it
1:24:43
the simpsons lego actually i thought it
1:24:43
the simpsons lego actually i thought it was
1:24:44
was
1:24:44
was nice and also thinking about parents i'm
1:24:46
nice and also thinking about parents i'm
1:24:46
nice and also thinking about parents i'm not a parent myself but
1:24:47
not a parent myself but
1:24:47
not a parent myself but i do have two nieces and sometimes they
1:24:50
i do have two nieces and sometimes they
1:24:50
i do have two nieces and sometimes they like to
1:24:51
like to
1:24:51
like to get specific stuff for their birthdays
1:24:53
get specific stuff for their birthdays
1:24:54
get specific stuff for their birthdays and
1:24:54
and
1:24:54
and i thought well maybe this is a nice
1:24:56
i thought well maybe this is a nice
1:24:56
i thought well maybe this is a nice helping tool so you can recognize
1:24:59
helping tool so you can recognize
1:24:59
helping tool so you can recognize uh characters so you know when they ask
1:25:01
uh characters so you know when they ask
1:25:01
uh characters so you know when they ask for
1:25:02
for
1:25:02
for in this case bart simpson or lisa
1:25:04
in this case bart simpson or lisa
1:25:04
in this case bart simpson or lisa simpson that you will come home with the
1:25:06
simpson that you will come home with the
1:25:06
simpson that you will come home with the right stuff
1:25:07
right stuff
1:25:07
right stuff so and um yeah if you don't want to
1:25:11
so and um yeah if you don't want to
1:25:11
so and um yeah if you don't want to make you know a very complex model you
1:25:13
make you know a very complex model you
1:25:13
make you know a very complex model you can use custom vision which is basically
1:25:16
can use custom vision which is basically
1:25:16
can use custom vision which is basically using the pre-trained models from
1:25:18
using the pre-trained models from
1:25:18
using the pre-trained models from microsoft
1:25:20
microsoft
1:25:20
microsoft and just use it as an interface so you
1:25:21
and just use it as an interface so you
1:25:21
and just use it as an interface so you don't have to code
1:25:23
don't have to code
1:25:23
don't have to code you could just upload images we will see
1:25:26
you could just upload images we will see
1:25:26
you could just upload images we will see that later on
1:25:27
that later on
1:25:27
that later on and you will get an answer back so
1:25:30
and you will get an answer back so
1:25:30
and you will get an answer back so without programming you
1:25:31
without programming you
1:25:31
without programming you get actually the answer this is really
1:25:33
get actually the answer this is really
1:25:33
get actually the answer this is really cool i mean i literally had this problem
1:25:35
cool i mean i literally had this problem
1:25:35
cool i mean i literally had this problem the other day
1:25:36
the other day
1:25:36
the other day i wanted my i wanted to know which
1:25:38
i wanted my i wanted to know which
1:25:38
i wanted my i wanted to know which simpson
1:25:39
simpson
1:25:40
simpson my kid was playing no i'm just kidding
1:25:41
my kid was playing no i'm just kidding
1:25:41
my kid was playing no i'm just kidding but it is a useful problem and i love
1:25:43
but it is a useful problem and i love
1:25:43
but it is a useful problem and i love this example
1:25:44
this example
1:25:44
this example i'm a huge fan of it so should we dive
1:25:47
i'm a huge fan of it so should we dive
1:25:47
i'm a huge fan of it so should we dive in yeah i think that's a nice idea
1:25:50
in yeah i think that's a nice idea
1:25:50
in yeah i think that's a nice idea so i think that i should share my screen
1:25:54
so i think that i should share my screen
1:25:54
so i think that i should share my screen so let's see
1:25:55
so let's see
1:25:55
so let's see what i can share with you can you make
1:25:58
what i can share with you can you make
1:25:58
what i can share with you can you make your font
1:25:59
your font
1:25:59
your font a little bit bigger by like maybe twice
1:26:01
a little bit bigger by like maybe twice
1:26:02
a little bit bigger by like maybe twice the size well this is not
1:26:03
the size well this is not
1:26:03
the size well this is not my screen actually so oh yeah we need to
1:26:06
my screen actually so oh yeah we need to
1:26:06
my screen actually so oh yeah we need to go to the app other
1:26:07
go to the app other
1:26:07
go to the app other dishes so yes it's perfect
1:26:10
dishes so yes it's perfect
1:26:10
dishes so yes it's perfect can you see something i sure can at
1:26:13
can you see something i sure can at
1:26:13
can you see something i sure can at least
1:26:13
least
1:26:13
least i can it looks beautiful i see a bird
1:26:15
i can it looks beautiful i see a bird
1:26:16
i can it looks beautiful i see a bird and stuff okay
1:26:17
and stuff okay
1:26:17
and stuff okay good now you see the right screen i
1:26:19
good now you see the right screen i
1:26:19
good now you see the right screen i can't see the screen right now so
1:26:21
can't see the screen right now so
1:26:21
can't see the screen right now so please let me know if i'm showing weird
1:26:23
please let me know if i'm showing weird
1:26:23
please let me know if i'm showing weird things so i want to show you
1:26:25
things so i want to show you
1:26:25
things so i want to show you is basically the landing page for custom
1:26:28
is basically the landing page for custom
1:26:28
is basically the landing page for custom vision.ai
1:26:29
vision.ai
1:26:29
vision.ai it's um one of the services where
1:26:33
it's um one of the services where
1:26:33
it's um one of the services where you can actually use your own own images
1:26:37
you can actually use your own own images
1:26:37
you can actually use your own own images train your own model but based on
1:26:39
train your own model but based on
1:26:39
train your own model but based on already
1:26:40
already
1:26:40
already uh pre trained models from microsoft and
1:26:42
uh pre trained models from microsoft and
1:26:42
uh pre trained models from microsoft and then use it
1:26:43
then use it
1:26:43
then use it so when let's say that the standard
1:26:46
so when let's say that the standard
1:26:46
so when let's say that the standard cognitive services are not enough for
1:26:48
cognitive services are not enough for
1:26:48
cognitive services are not enough for you or you want something really
1:26:50
you or you want something really
1:26:50
you or you want something really specifically like for example i love
1:26:51
specifically like for example i love
1:26:51
specifically like for example i love dogs some people that know me maybe know
1:26:53
dogs some people that know me maybe know
1:26:53
dogs some people that know me maybe know the
1:26:54
the
1:26:54
the dog spotter i build and av boyd wrote a
1:26:56
dog spotter i build and av boyd wrote a
1:26:56
dog spotter i build and av boyd wrote a really nice tutorial for that
1:26:58
really nice tutorial for that
1:26:58
really nice tutorial for that too but now we're going to focus on
1:27:01
too but now we're going to focus on
1:27:02
too but now we're going to focus on simpsons and yeah of course maybe
1:27:03
simpsons and yeah of course maybe
1:27:03
simpsons and yeah of course maybe microsoft already trained their models
1:27:05
microsoft already trained their models
1:27:05
microsoft already trained their models also in the same sense but
1:27:07
also in the same sense but
1:27:07
also in the same sense but you know let's just assume they don't so
1:27:10
you know let's just assume they don't so
1:27:10
you know let's just assume they don't so um i'm gonna talk you through this if
1:27:13
um i'm gonna talk you through this if
1:27:13
um i'm gonna talk you through this if you
1:27:14
you
1:27:14
you uh do want to do it at home i wrote a
1:27:17
uh do want to do it at home i wrote a
1:27:17
uh do want to do it at home i wrote a blog about it
1:27:18
blog about it
1:27:18
blog about it i think villain will share the link to
1:27:20
i think villain will share the link to
1:27:20
i think villain will share the link to it and i also have a github repo so you
1:27:21
it and i also have a github repo so you
1:27:21
it and i also have a github repo so you can just
1:27:22
can just
1:27:22
can just fork it and play with it so let's just
1:27:25
fork it and play with it so let's just
1:27:25
fork it and play with it so let's just start with signing in
1:27:26
start with signing in
1:27:26
start with signing in so my main problem is if i see an image
1:27:29
so my main problem is if i see an image
1:27:30
so my main problem is if i see an image from the simpsons i don't know who's who
1:27:31
from the simpsons i don't know who's who
1:27:32
from the simpsons i don't know who's who right so we want to get an answer to
1:27:33
right so we want to get an answer to
1:27:33
right so we want to get an answer to that in order to do so
1:27:36
that in order to do so
1:27:36
that in order to do so we're going to build a model and then
1:27:38
we're going to build a model and then
1:27:38
we're going to build a model and then with some test images
1:27:40
with some test images
1:27:40
with some test images all coming from hank we're going to test
1:27:43
all coming from hank we're going to test
1:27:43
all coming from hank we're going to test if we did the right thing
1:27:45
if we did the right thing
1:27:46
if we did the right thing now um i will create a new project just
1:27:48
now um i will create a new project just
1:27:48
now um i will create a new project just to
1:27:49
to
1:27:49
to really talk you through everything so we
1:27:51
really talk you through everything so we
1:27:51
really talk you through everything so we have to give it a name
1:27:52
have to give it a name
1:27:52
have to give it a name we call it uh simpsons uh demo in this
1:27:55
we call it uh simpsons uh demo in this
1:27:55
we call it uh simpsons uh demo in this case
1:27:56
case
1:27:56
case um recognize
1:27:59
um recognize
1:28:00
um recognize the simpsons right and then we come to
1:28:03
the simpsons right and then we come to
1:28:03
the simpsons right and then we come to these things i'm not going to dive into
1:28:05
these things i'm not going to dive into
1:28:05
these things i'm not going to dive into the
1:28:06
the
1:28:06
the the specific variables and settings you
1:28:08
the specific variables and settings you
1:28:08
the specific variables and settings you can read that back
1:28:09
can read that back
1:28:09
can read that back but we need a resource of course we have
1:28:12
but we need a resource of course we have
1:28:12
but we need a resource of course we have to run this on azure so we need some
1:28:14
to run this on azure so we need some
1:28:14
to run this on azure so we need some um some resources i created already a
1:28:17
um some resources i created already a
1:28:17
um some resources i created already a simpsons one
1:28:18
simpsons one
1:28:18
simpsons one but you have the option to create
1:28:19
but you have the option to create
1:28:19
but you have the option to create everything yourself
1:28:21
everything yourself
1:28:21
everything yourself now we want to classify an image so
1:28:24
now we want to classify an image so
1:28:24
now we want to classify an image so the model comes back with which
1:28:27
the model comes back with which
1:28:27
the model comes back with which character it is so what class it is
1:28:29
character it is so what class it is
1:28:29
character it is so what class it is and we use images that have only
1:28:32
and we use images that have only
1:28:32
and we use images that have only one image per picture i will show you
1:28:35
one image per picture i will show you
1:28:35
one image per picture i will show you that later on
1:28:36
that later on
1:28:36
that later on and in this case it retails pretty good
1:28:39
and in this case it retails pretty good
1:28:39
and in this case it retails pretty good and we do something
1:28:41
and we do something
1:28:41
and we do something basic all right just quick there we go
1:28:44
basic all right just quick there we go
1:28:44
basic all right just quick there we go and now we have our basically our setup
1:28:48
and now we have our basically our setup
1:28:48
and now we have our basically our setup so you can see just in a few clicks
1:28:49
so you can see just in a few clicks
1:28:49
so you can see just in a few clicks we're done and hank
1:28:52
we're done and hank
1:28:52
we're done and hank has made a really nice website it's a
1:28:54
has made a really nice website it's a
1:28:54
has made a really nice website it's a developer's guide to azure ai
1:28:56
developer's guide to azure ai
1:28:56
developer's guide to azure ai and one of his uh labs can you all see
1:28:59
and one of his uh labs can you all see
1:28:59
and one of his uh labs can you all see that actually
1:29:00
that actually
1:29:00
that actually is my screen sure yeah
1:29:03
is my screen sure yeah
1:29:03
is my screen sure yeah good yeah oh sorry make it a little bit
1:29:05
good yeah oh sorry make it a little bit
1:29:05
good yeah oh sorry make it a little bit bigger
1:29:06
bigger
1:29:06
bigger they want you to zoom in just to hear if
1:29:08
they want you to zoom in just to hear if
1:29:08
they want you to zoom in just to hear if that's okay exactly
1:29:10
that's okay exactly
1:29:10
that's okay exactly so the second one is the custom vision
1:29:12
so the second one is the custom vision
1:29:12
so the second one is the custom vision one
1:29:13
one
1:29:13
one it's a lap you can also build this one
1:29:16
it's a lap you can also build this one
1:29:16
it's a lap you can also build this one yourself
1:29:17
yourself
1:29:17
yourself and you will find a nice link to
1:29:20
and you will find a nice link to
1:29:20
and you will find a nice link to the uh the data set so you will see this
1:29:23
the uh the data set so you will see this
1:29:23
the uh the data set so you will see this yourself you have a lego
1:29:25
yourself you have a lego
1:29:25
yourself you have a lego data set of course i downloaded it
1:29:27
data set of course i downloaded it
1:29:28
data set of course i downloaded it already
1:29:28
already
1:29:28
already so um it's there
1:29:32
and i can show you that actually i'm not
1:29:34
and i can show you that actually i'm not
1:29:34
and i can show you that actually i'm not sure whether
1:29:36
sure whether
1:29:36
sure whether that will be you can't see my screen i
1:29:40
that will be you can't see my screen i
1:29:40
that will be you can't see my screen i think
1:29:41
think
1:29:41
think well don't worry
1:29:44
okay so we go back we can see the screen
1:29:47
okay so we go back we can see the screen
1:29:47
okay so we go back we can see the screen right there yeah one of them
1:29:48
right there yeah one of them
1:29:48
right there yeah one of them so and okay and in order to train the
1:29:51
so and okay and in order to train the
1:29:51
so and okay and in order to train the model of course
1:29:52
model of course
1:29:52
model of course i have to tell the model what's what
1:29:54
i have to tell the model what's what
1:29:54
i have to tell the model what's what right because i cannot just say all
1:29:56
right because i cannot just say all
1:29:56
right because i cannot just say all right this is a picture
1:29:57
right this is a picture
1:29:57
right this is a picture just give the answer so you have to do
1:29:59
just give the answer so you have to do
1:29:59
just give the answer so you have to do something
1:30:00
something
1:30:00
something and this is where you can add images and
1:30:02
and this is where you can add images and
1:30:02
and this is where you can add images and you have to give it a name
1:30:04
you have to give it a name
1:30:04
you have to give it a name now i downloaded everything already i
1:30:06
now i downloaded everything already i
1:30:06
now i downloaded everything already i called it the simpsons we have the
1:30:08
called it the simpsons we have the
1:30:08
called it the simpsons we have the data set and hank already classified
1:30:11
data set and hank already classified
1:30:11
data set and hank already classified everything so in this case we're just
1:30:13
everything so in this case we're just
1:30:13
everything so in this case we're just going to do two and i'll tell you how
1:30:16
going to do two and i'll tell you how
1:30:16
going to do two and i'll tell you how you can finish it later on
1:30:17
you can finish it later on
1:30:17
you can finish it later on so from bart simpsons we will use um
1:30:21
so from bart simpsons we will use um
1:30:21
so from bart simpsons we will use um a few pictures now at least 20 that
1:30:23
a few pictures now at least 20 that
1:30:23
a few pictures now at least 20 that would be really great actually that's
1:30:24
would be really great actually that's
1:30:24
would be really great actually that's already enough
1:30:25
already enough
1:30:25
already enough um but you don't want to use everything
1:30:27
um but you don't want to use everything
1:30:28
um but you don't want to use everything because you want to test it later on
1:30:29
because you want to test it later on
1:30:30
because you want to test it later on and of course you it's not fair when you
1:30:32
and of course you it's not fair when you
1:30:32
and of course you it's not fair when you train a model with an
1:30:33
train a model with an
1:30:33
train a model with an image and you show the same image of
1:30:35
image and you show the same image of
1:30:35
image and you show the same image of course that's tricky so
1:30:37
course that's tricky so
1:30:37
course that's tricky so let's say we just take uh these
1:30:40
let's say we just take uh these
1:30:40
let's say we just take uh these and we save the last ones we open them
1:30:44
and we save the last ones we open them
1:30:44
and we save the last ones we open them and then we have to say all right this
1:30:46
and then we have to say all right this
1:30:46
and then we have to say all right this is bart simpson
1:30:47
is bart simpson
1:30:47
is bart simpson so i give it a tag okay i upload them
1:30:53
and there we go we have now 36 images
1:30:56
and there we go we have now 36 images
1:30:56
and there we go we have now 36 images to learn the model how board seems to
1:30:58
to learn the model how board seems to
1:30:58
to learn the model how board seems to look like
1:31:00
look like
1:31:00
look like and of course we want another image now
1:31:02
and of course we want another image now
1:31:02
and of course we want another image now seth which image do you want
1:31:04
seth which image do you want
1:31:04
seth which image do you want i think i'm going to go with of course
1:31:07
i think i'm going to go with of course
1:31:07
i think i'm going to go with of course krusty
1:31:07
krusty
1:31:08
krusty the clown all right good okay we go with
1:31:11
the clown all right good okay we go with
1:31:11
the clown all right good okay we go with crusty so the same
1:31:12
crusty so the same
1:31:12
crusty so the same we want to save some test images so we
1:31:14
we want to save some test images so we
1:31:14
we want to save some test images so we take the upper part of it and we say
1:31:17
take the upper part of it and we say
1:31:17
take the upper part of it and we say this is crusty
1:31:19
this is crusty
1:31:19
this is crusty okay there you go
1:31:27
okay good then we're done
1:31:30
okay good then we're done
1:31:30
okay good then we're done and that's actually it regarding the
1:31:33
and that's actually it regarding the
1:31:33
and that's actually it regarding the data
1:31:34
data
1:31:34
data now we're ready to train the model so
1:31:36
now we're ready to train the model so
1:31:36
now we're ready to train the model so we're going to train it we do it quickly
1:31:37
we're going to train it we do it quickly
1:31:37
we're going to train it we do it quickly of course
1:31:38
of course
1:31:38
of course uh because of the demo yeah here we have
1:31:40
uh because of the demo yeah here we have
1:31:40
uh because of the demo yeah here we have to wait a little while
1:31:43
to wait a little while
1:31:43
to wait a little while just be patient not too much hopefully
1:31:46
just be patient not too much hopefully
1:31:46
just be patient not too much hopefully so while it's going i'll ask you some
1:31:48
so while it's going i'll ask you some
1:31:48
so while it's going i'll ask you some questions yes so basically let me see if
1:31:50
questions yes so basically let me see if
1:31:50
questions yes so basically let me see if i understand
1:31:51
i understand
1:31:51
i understand correctly you are building a model that
1:31:54
correctly you are building a model that
1:31:54
correctly you are building a model that is going to
1:31:55
is going to
1:31:55
is going to tell us whether or not an
1:31:58
tell us whether or not an
1:31:58
tell us whether or not an image that you give it is krusty the
1:32:00
image that you give it is krusty the
1:32:00
image that you give it is krusty the clown or
1:32:01
clown or
1:32:01
clown or bart simpson is that right so the
1:32:04
bart simpson is that right so the
1:32:04
bart simpson is that right so the probability that's one of the two
1:32:05
probability that's one of the two
1:32:06
probability that's one of the two yeah fantastic or none hopefully
1:32:09
yeah fantastic or none hopefully
1:32:09
yeah fantastic or none hopefully or non-hobo okay then the other question
1:32:11
or non-hobo okay then the other question
1:32:11
or non-hobo okay then the other question i have
1:32:12
i have
1:32:12
i have is um once you've
1:32:16
is um once you've
1:32:16
is um once you've you've uploaded these images like i'm
1:32:18
you've uploaded these images like i'm
1:32:18
you've uploaded these images like i'm trying to understand is there code you
1:32:20
trying to understand is there code you
1:32:20
trying to understand is there code you have to write to get any of this to work
1:32:22
have to write to get any of this to work
1:32:22
have to write to get any of this to work no not at all microsoft have done it
1:32:25
no not at all microsoft have done it
1:32:25
no not at all microsoft have done it already for you
1:32:26
already for you
1:32:26
already for you because you're they already created a
1:32:29
because you're they already created a
1:32:29
because you're they already created a deep learning model
1:32:30
deep learning model
1:32:30
deep learning model and you don't have to bother about that
1:32:34
and you don't have to bother about that
1:32:34
and you don't have to bother about that i see and so that's the nice part can
1:32:36
i see and so that's the nice part can
1:32:36
i see and so that's the nice part can you give us a sense for what's actually
1:32:38
you give us a sense for what's actually
1:32:38
you give us a sense for what's actually going
1:32:38
going
1:32:38
going on in the background in the background
1:32:42
on in the background in the background
1:32:42
on in the background in the background well it's
1:32:43
well it's
1:32:43
well it's chunking up the image into little pixels
1:32:46
chunking up the image into little pixels
1:32:46
chunking up the image into little pixels and let's see it like a sort of library
1:32:50
and let's see it like a sort of library
1:32:50
and let's see it like a sort of library where microsoft has already recognized
1:32:53
where microsoft has already recognized
1:32:53
where microsoft has already recognized specific things in this case you're
1:32:55
specific things in this case you're
1:32:55
specific things in this case you're basically
1:32:56
basically
1:32:56
basically uh pixelize all your bart simpson's
1:32:59
uh pixelize all your bart simpson's
1:32:59
uh pixelize all your bart simpson's images
1:33:01
images
1:33:01
images and um then
1:33:04
and um then
1:33:04
and um then yeah then you're sort of comparing it if
1:33:07
yeah then you're sort of comparing it if
1:33:07
yeah then you're sort of comparing it if if that's a way to explain it
1:33:09
if that's a way to explain it
1:33:09
if that's a way to explain it and then the but of course in a more
1:33:12
and then the but of course in a more
1:33:12
and then the but of course in a more uh advanced way than just the human eye
1:33:14
uh advanced way than just the human eye
1:33:14
uh advanced way than just the human eye i would say
1:33:15
i would say
1:33:15
i would say got it basically you we cannot not you
1:33:18
got it basically you we cannot not you
1:33:18
got it basically you we cannot not you know we can see a lot of pictures and
1:33:19
know we can see a lot of pictures and
1:33:19
know we can see a lot of pictures and then
1:33:20
then
1:33:20
then you know remember it but of course a
1:33:22
you know remember it but of course a
1:33:22
you know remember it but of course a model can store it
1:33:24
model can store it
1:33:24
model can store it yeah okay so basically what it's doing
1:33:26
yeah okay so basically what it's doing
1:33:26
yeah okay so basically what it's doing is it's going through some
1:33:28
is it's going through some
1:33:28
is it's going through some process right now where it's looking at
1:33:31
process right now where it's looking at
1:33:31
process right now where it's looking at these pictures
1:33:32
these pictures
1:33:32
these pictures some looking i'm putting an air quote
1:33:34
some looking i'm putting an air quote
1:33:34
some looking i'm putting an air quote obviously computers don't have eyes
1:33:36
obviously computers don't have eyes
1:33:36
obviously computers don't have eyes look at looking at the picture and store
1:33:39
look at looking at the picture and store
1:33:39
look at looking at the picture and store it like hey this is bart and borrowed
1:33:40
it like hey this is bart and borrowed
1:33:40
it like hey this is bart and borrowed from all kind of angles and
1:33:42
from all kind of angles and
1:33:42
from all kind of angles and maybe different outfits or in this case
1:33:44
maybe different outfits or in this case
1:33:44
maybe different outfits or in this case with the same outfit
1:33:45
with the same outfit
1:33:45
with the same outfit but he just learns how bart looks like
1:33:49
but he just learns how bart looks like
1:33:49
but he just learns how bart looks like i see and then that way like if you want
1:33:51
i see and then that way like if you want
1:33:51
i see and then that way like if you want to and then if you want to project this
1:33:53
to and then if you want to project this
1:33:53
to and then if you want to project this to like
1:33:53
to like
1:33:53
to like a business use case you might have a
1:33:55
a business use case you might have a
1:33:55
a business use case you might have a situation like
1:33:56
situation like
1:33:56
situation like on a show floor or something where you
1:33:59
on a show floor or something where you
1:34:00
on a show floor or something where you want to be able to recognize where your
1:34:01
want to be able to recognize where your
1:34:01
want to be able to recognize where your products are for example
1:34:03
products are for example
1:34:03
products are for example would that be a i see what are you
1:34:06
would that be a i see what are you
1:34:06
would that be a i see what are you thinking about
1:34:09
a machine for example that has different
1:34:12
a machine for example that has different
1:34:12
a machine for example that has different spare parts and you want to know which
1:34:14
spare parts and you want to know which
1:34:14
spare parts and you want to know which part is broken so for example a person
1:34:17
part is broken so for example a person
1:34:17
part is broken so for example a person who's working on that can take a picture
1:34:19
who's working on that can take a picture
1:34:19
who's working on that can take a picture and he knows that i have to replace this
1:34:20
and he knows that i have to replace this
1:34:20
and he knows that i have to replace this part
1:34:21
part
1:34:21
part um also those practical things are very
1:34:24
um also those practical things are very
1:34:24
um also those practical things are very handy
1:34:25
handy
1:34:26
handy fantastic and so for for those that want
1:34:27
fantastic and so for for those that want
1:34:27
fantastic and so for for those that want to get started with this
1:34:29
to get started with this
1:34:29
to get started with this it's literally just create a custom bit
1:34:32
it's literally just create a custom bit
1:34:32
it's literally just create a custom bit have an account and then literally just
1:34:35
have an account and then literally just
1:34:35
have an account and then literally just go through the clicking that you did
1:34:36
go through the clicking that you did
1:34:36
go through the clicking that you did have you used this for anything
1:34:38
have you used this for anything
1:34:38
have you used this for anything other than bart simpson or have you seen
1:34:40
other than bart simpson or have you seen
1:34:40
other than bart simpson or have you seen or have any ideas on how people could
1:34:42
or have any ideas on how people could
1:34:42
or have any ideas on how people could use this
1:34:43
use this
1:34:43
use this yeah well i i really used it for dogs uh
1:34:46
yeah well i i really used it for dogs uh
1:34:46
yeah well i i really used it for dogs uh because the nice thing is that
1:34:47
because the nice thing is that
1:34:47
because the nice thing is that um you can wrap it up and build a build
1:34:50
um you can wrap it up and build a build
1:34:50
um you can wrap it up and build a build a small app around it so you have
1:34:52
a small app around it so you have
1:34:52
a small app around it so you have powerapps in your flow
1:34:53
powerapps in your flow
1:34:54
powerapps in your flow so basically what you do you build a
1:34:55
so basically what you do you build a
1:34:55
so basically what you do you build a simple app and you can take pictures
1:34:58
simple app and you can take pictures
1:34:58
simple app and you can take pictures and um what i did is just you know
1:35:00
and um what i did is just you know
1:35:00
and um what i did is just you know taking pictures of dogs
1:35:02
taking pictures of dogs
1:35:02
taking pictures of dogs and of course train them all but also
1:35:03
and of course train them all but also
1:35:03
and of course train them all but also taking real life pictures
1:35:05
taking real life pictures
1:35:05
taking real life pictures and using it on my own dogs and that was
1:35:07
and using it on my own dogs and that was
1:35:07
and using it on my own dogs and that was pretty fun
1:35:09
pretty fun
1:35:09
pretty fun but i know also there's also a business
1:35:10
but i know also there's also a business
1:35:10
but i know also there's also a business case i i can't
1:35:12
case i i can't
1:35:12
case i i can't tell you what but then they're using
1:35:14
tell you what but then they're using
1:35:14
tell you what but then they're using this as well
1:35:15
this as well
1:35:15
this as well got it so yes that's awesome all right
1:35:18
got it so yes that's awesome all right
1:35:18
got it so yes that's awesome all right so
1:35:18
so
1:35:18
so uh looks like it's done yeah we're done
1:35:21
uh looks like it's done yeah we're done
1:35:21
uh looks like it's done yeah we're done and and the good news yeah and of course
1:35:22
and and the good news yeah and of course
1:35:22
and and the good news yeah and of course this is sort of
1:35:23
this is sort of
1:35:23
this is sort of perfect right um the model is very
1:35:26
perfect right um the model is very
1:35:26
perfect right um the model is very precise so it can tell you
1:35:28
precise so it can tell you
1:35:28
precise so it can tell you very well whether um some image is
1:35:31
very well whether um some image is
1:35:31
very well whether um some image is crusty or not or bart simpson's or not
1:35:33
crusty or not or bart simpson's or not
1:35:33
crusty or not or bart simpson's or not so both krusty and bard are well
1:35:35
so both krusty and bard are well
1:35:36
so both krusty and bard are well represented
1:35:37
represented
1:35:37
represented you can see that under here right this
1:35:39
you can see that under here right this
1:35:39
you can see that under here right this part
1:35:41
part
1:35:41
part um so actually this is sort of your
1:35:43
um so actually this is sort of your
1:35:43
um so actually this is sort of your perfect model
1:35:44
perfect model
1:35:44
perfect model and of course you want to test the model
1:35:46
and of course you want to test the model
1:35:46
and of course you want to test the model right okay well actually now we can do a
1:35:49
right okay well actually now we can do a
1:35:49
right okay well actually now we can do a quick test but you have to close your
1:35:50
quick test but you have to close your
1:35:50
quick test but you have to close your eyes actually
1:35:51
eyes actually
1:35:51
eyes actually can you i mean no you will see the image
1:35:53
can you i mean no you will see the image
1:35:53
can you i mean no you will see the image so i have to get an image
1:35:56
so i have to get an image
1:35:56
so i have to get an image i can just browse a local file and uh
1:35:59
i can just browse a local file and uh
1:35:59
i can just browse a local file and uh yeah take a new one so this is your
1:36:01
yeah take a new one so this is your
1:36:01
yeah take a new one so this is your unseen model
1:36:02
unseen model
1:36:02
unseen model by unseen image by the model and it will
1:36:06
by unseen image by the model and it will
1:36:06
by unseen image by the model and it will come back
1:36:07
come back
1:36:07
come back actually in a flash second and it gives
1:36:10
actually in a flash second and it gives
1:36:10
actually in a flash second and it gives you the probability
1:36:11
you the probability
1:36:11
you the probability that this picture is crusty so in this
1:36:14
that this picture is crusty so in this
1:36:14
that this picture is crusty so in this case you've seen that
1:36:16
case you've seen that
1:36:16
case you've seen that um yeah basically in a few minutes right
1:36:19
um yeah basically in a few minutes right
1:36:19
um yeah basically in a few minutes right we're talking you've trained the model
1:36:20
we're talking you've trained the model
1:36:20
we're talking you've trained the model and you can recognize uh krusty or bart
1:36:23
and you can recognize uh krusty or bart
1:36:23
and you can recognize uh krusty or bart simpson of course we have to try bart
1:36:25
simpson of course we have to try bart
1:36:25
simpson of course we have to try bart too right because if not of course
1:36:28
too right because if not of course
1:36:28
too right because if not of course it wouldn't be okay so let's take the
1:36:31
it wouldn't be okay so let's take the
1:36:31
it wouldn't be okay so let's take the last one from bart
1:36:33
last one from bart
1:36:33
last one from bart and let's see and yay we're doing right
1:36:36
and let's see and yay we're doing right
1:36:36
and let's see and yay we're doing right okay
1:36:40
yeah there's a high profit so nearly 100
1:36:43
yeah there's a high profit so nearly 100
1:36:43
yeah there's a high profit so nearly 100 sure that it's
1:36:44
sure that it's
1:36:44
sure that it's smart um and now of course
1:36:47
smart um and now of course
1:36:48
smart um and now of course yeah this is just a small model and if
1:36:50
yeah this is just a small model and if
1:36:50
yeah this is just a small model and if you want to
1:36:51
you want to
1:36:51
you want to to elaborate on you want to
1:36:54
to elaborate on you want to
1:36:54
to elaborate on you want to to do more well it's actually quite
1:36:57
to do more well it's actually quite
1:36:57
to do more well it's actually quite easy it's just like adding images to
1:37:00
easy it's just like adding images to
1:37:00
easy it's just like adding images to your computer
1:37:00
your computer
1:37:00
your computer you say right i want to add more images
1:37:03
you say right i want to add more images
1:37:03
you say right i want to add more images i
1:37:04
i
1:37:04
i for example would like to include
1:37:07
for example would like to include
1:37:07
for example would like to include um i don't know let's do lisa simpson
1:37:10
um i don't know let's do lisa simpson
1:37:10
um i don't know let's do lisa simpson for example
1:37:12
for example
1:37:12
for example fantastic right oops
1:37:15
fantastic right oops
1:37:15
fantastic right oops i have to include her and you just add
1:37:17
i have to include her and you just add
1:37:17
i have to include her and you just add an image
1:37:18
an image
1:37:18
an image and you say alright this is lisa you
1:37:21
and you say alright this is lisa you
1:37:21
and you say alright this is lisa you upload it
1:37:24
there you go and you follow the steps
1:37:26
there you go and you follow the steps
1:37:26
there you go and you follow the steps you just train your model again
1:37:28
you just train your model again
1:37:28
you just train your model again and now we're not gonna wait to do that
1:37:30
and now we're not gonna wait to do that
1:37:30
and now we're not gonna wait to do that but just to show you that you can
1:37:32
but just to show you that you can
1:37:32
but just to show you that you can iterate so this is
1:37:33
iterate so this is
1:37:33
iterate so this is iteration too and as long as the siri
1:37:35
iteration too and as long as the siri
1:37:35
iteration too and as long as the siri grows you can add more images you can
1:37:38
grows you can add more images you can
1:37:38
grows you can add more images you can add more characters
1:37:39
add more characters
1:37:40
add more characters retrain your model and just use it
1:37:44
that's cool that's cool and so now for
1:37:47
that's cool that's cool and so now for
1:37:47
that's cool that's cool and so now for example
1:37:48
example
1:37:48
example you you can literally classify any kinds
1:37:51
you you can literally classify any kinds
1:37:51
you you can literally classify any kinds of images
1:37:52
of images
1:37:52
of images if you just upload them and and train
1:37:53
if you just upload them and and train
1:37:53
if you just upload them and and train the model yeah so you just
1:37:55
the model yeah so you just
1:37:55
the model yeah so you just need 20 at least 20 images and of course
1:37:59
need 20 at least 20 images and of course
1:37:59
need 20 at least 20 images and of course you have to
1:38:00
you have to
1:38:00
you have to think about it because i think we all
1:38:02
think about it because i think we all
1:38:02
think about it because i think we all know uh the funny picture with the
1:38:04
know uh the funny picture with the
1:38:04
know uh the funny picture with the chihuahua
1:38:05
chihuahua
1:38:05
chihuahua and uh like the muffin all right yes
1:38:08
and uh like the muffin all right yes
1:38:08
and uh like the muffin all right yes of course if you want to train them all
1:38:10
of course if you want to train them all
1:38:10
of course if you want to train them all on chihuahuas or muffins then make sure
1:38:13
on chihuahuas or muffins then make sure
1:38:13
on chihuahuas or muffins then make sure that
1:38:13
that
1:38:13
that it's as accurate as possible so take
1:38:16
it's as accurate as possible so take
1:38:16
it's as accurate as possible so take clear pictures
1:38:17
clear pictures
1:38:17
clear pictures um and of course not yeah you need good
1:38:20
um and of course not yeah you need good
1:38:20
um and of course not yeah you need good material it's it's still a little bit
1:38:22
material it's it's still a little bit
1:38:22
material it's it's still a little bit you know prepping crap out right so beat
1:38:25
you know prepping crap out right so beat
1:38:25
you know prepping crap out right so beat it with good model
1:38:26
it with good model
1:38:26
it with good model good pictures um you have the best
1:38:28
good pictures um you have the best
1:38:28
good pictures um you have the best chance of getting a good
1:38:29
chance of getting a good
1:38:30
chance of getting a good model i guess the last thing you would
1:38:32
model i guess the last thing you would
1:38:32
model i guess the last thing you would ever want
1:38:34
ever want
1:38:34
ever want would be to eat a chihuahua instead
1:38:38
would be to eat a chihuahua instead
1:38:38
would be to eat a chihuahua instead of a muffin yeah i think so
1:38:42
of a muffin yeah i think so
1:38:42
of a muffin yeah i think so here's some questions uh kira says what
1:38:44
here's some questions uh kira says what
1:38:44
here's some questions uh kira says what if you what if you test an image
1:38:46
if you what if you test an image
1:38:46
if you what if you test an image that isn't crusty nor bart ah
1:38:50
that isn't crusty nor bart ah
1:38:50
that isn't crusty nor bart ah okay okay let's do that let's see if i
1:38:52
okay okay let's do that let's see if i
1:38:52
okay okay let's do that let's see if i can interrupt the
1:38:53
can interrupt the
1:38:53
can interrupt the uh you can do a quick test on the
1:38:56
uh you can do a quick test on the
1:38:56
uh you can do a quick test on the iteration one
1:38:57
iteration one
1:38:57
iteration one exactly so we're going to do that so
1:38:59
exactly so we're going to do that so
1:38:59
exactly so we're going to do that so let's see it will come back with
1:39:01
let's see it will come back with
1:39:01
let's see it will come back with a probabilistic mr barnes
1:39:05
a probabilistic mr barnes
1:39:05
a probabilistic mr barnes let's see where he comes back with well
1:39:07
let's see where he comes back with well
1:39:07
let's see where he comes back with well still he thinks
1:39:08
still he thinks
1:39:08
still he thinks it's more than crusty right so here you
1:39:12
it's more than crusty right so here you
1:39:12
it's more than crusty right so here you can also see
1:39:13
can also see
1:39:13
can also see um possible problems right we have
1:39:17
um possible problems right we have
1:39:17
um possible problems right we have uh he said yeah it's bar but and that's
1:39:19
uh he said yeah it's bar but and that's
1:39:19
uh he said yeah it's bar but and that's the difference
1:39:20
the difference
1:39:20
the difference it's actually a low probability if it
1:39:22
it's actually a low probability if it
1:39:22
it's actually a low probability if it comes to the pictures that are more
1:39:24
comes to the pictures that are more
1:39:24
comes to the pictures that are more related to bart simpsons
1:39:26
related to bart simpsons
1:39:26
related to bart simpsons so this would be something like all
1:39:28
so this would be something like all
1:39:28
so this would be something like all right the model says it's barred
1:39:30
right the model says it's barred
1:39:30
right the model says it's barred but hey we're not as sure as normally
1:39:33
but hey we're not as sure as normally
1:39:33
but hey we're not as sure as normally so something is happening and what you
1:39:36
so something is happening and what you
1:39:36
so something is happening and what you can do
1:39:37
can do
1:39:38
can do and you just you know you're starting
1:39:39
and you just you know you're starting
1:39:39
and you just you know you're starting using your model and you can see your
1:39:40
using your model and you can see your
1:39:40
using your model and you can see your predictions
1:39:41
predictions
1:39:41
predictions and you can see all right i
1:39:45
and you can see all right i
1:39:45
and you can see all right i uh it would be nice to see my end ah i
1:39:47
uh it would be nice to see my end ah i
1:39:47
uh it would be nice to see my end ah i just got another one
1:39:50
just got another one
1:39:50
just got another one okay quick test sorry it's ratio one and
1:39:52
okay quick test sorry it's ratio one and
1:39:52
okay quick test sorry it's ratio one and here you can see
1:39:53
here you can see
1:39:53
here you can see what you've done and you can say all
1:39:55
what you've done and you can say all
1:39:55
what you've done and you can say all right it's classified
1:39:57
right it's classified
1:39:58
right it's classified as fart but hey it's not okay this
1:40:01
as fart but hey it's not okay this
1:40:01
as fart but hey it's not okay this is uh mr burns
1:40:04
is uh mr burns
1:40:04
is uh mr burns all right and
1:40:07
we will save it so it's now out of this
1:40:11
we will save it so it's now out of this
1:40:11
we will save it so it's now out of this it will be used to retrain the model so
1:40:13
it will be used to retrain the model so
1:40:13
it will be used to retrain the model so in this case you can
1:40:14
in this case you can
1:40:14
in this case you can do sort of a sort your own control and
1:40:18
do sort of a sort your own control and
1:40:18
do sort of a sort your own control and of course you can also
1:40:19
of course you can also
1:40:19
of course you can also i would write i think a program for that
1:40:22
i would write i think a program for that
1:40:22
i would write i think a program for that to say where everything below
1:40:24
to say where everything below
1:40:24
to say where everything below 95 in this case that's indicating bart
1:40:27
95 in this case that's indicating bart
1:40:27
95 in this case that's indicating bart simpson
1:40:28
simpson
1:40:28
simpson or crusty let's take a look at that and
1:40:30
or crusty let's take a look at that and
1:40:30
or crusty let's take a look at that and let's
1:40:31
let's
1:40:31
let's reclassify that that's cool so basically
1:40:34
reclassify that that's cool so basically
1:40:34
reclassify that that's cool so basically this model only knows about what you
1:40:37
this model only knows about what you
1:40:37
this model only knows about what you tell it
1:40:38
tell it
1:40:38
tell it and so if i were to get a picture of
1:40:39
and so if i were to get a picture of
1:40:39
and so if i were to get a picture of seth in this case i would either be
1:40:42
seth in this case i would either be
1:40:42
seth in this case i would either be bart krusty or mr burns if we retrain
1:40:47
bart krusty or mr burns if we retrain
1:40:47
bart krusty or mr burns if we retrain yeah probably it wouldn't be
1:40:50
yeah probably it wouldn't be
1:40:50
yeah probably it wouldn't be too difficult i think to get you but you
1:40:52
too difficult i think to get you but you
1:40:52
too difficult i think to get you but you can try that
1:40:55
can try that
1:40:55
can try that yeah i mean like and that's the thing i
1:40:56
yeah i mean like and that's the thing i
1:40:56
yeah i mean like and that's the thing i think it's important is that
1:40:58
think it's important is that
1:40:58
think it's important is that this particular model is not going to do
1:41:00
this particular model is not going to do
1:41:00
this particular model is not going to do anything outside of what you told it
1:41:02
anything outside of what you told it
1:41:02
anything outside of what you told it and i think that's that's one of the
1:41:04
and i think that's that's one of the
1:41:04
and i think that's that's one of the areas where i think
1:41:05
areas where i think
1:41:05
areas where i think folks get a little tripped up and
1:41:08
folks get a little tripped up and
1:41:08
folks get a little tripped up and thinking that ai
1:41:09
thinking that ai
1:41:09
thinking that ai does a lot more than what they
1:41:12
does a lot more than what they
1:41:12
does a lot more than what they think and so uh yeah all right
1:41:17
think and so uh yeah all right
1:41:17
think and so uh yeah all right let me go through some of the questions
1:41:20
let me go through some of the questions
1:41:20
let me go through some of the questions here uh someone asked about video
1:41:25
here uh someone asked about video
1:41:25
here uh someone asked about video and then uh amy answered she's really
1:41:28
and then uh amy answered she's really
1:41:28
and then uh amy answered she's really smart people
1:41:29
smart people
1:41:29
smart people uh she said there's something called
1:41:30
uh she said there's something called
1:41:30
uh she said there's something called video indexer and put that up
1:41:33
video indexer and put that up
1:41:33
video indexer and put that up vi.microsoft.com
1:41:34
vi.microsoft.com
1:41:34
vi.microsoft.com uh for video because it seems like it'd
1:41:35
uh for video because it seems like it'd
1:41:35
uh for video because it seems like it'd be similar and then they put up some
1:41:37
be similar and then they put up some
1:41:37
be similar and then they put up some links to the other one we also answered
1:41:39
links to the other one we also answered
1:41:39
links to the other one we also answered kira's question
1:41:40
kira's question
1:41:40
kira's question what if your test image isn't that of
1:41:42
what if your test image isn't that of
1:41:42
what if your test image isn't that of crusty nor bart
1:41:44
crusty nor bart
1:41:44
crusty nor bart and so we answered that so how hard how
1:41:46
and so we answered that so how hard how
1:41:46
and so we answered that so how hard how long did it take you to get set up with
1:41:48
long did it take you to get set up with
1:41:48
long did it take you to get set up with this service and
1:41:49
this service and
1:41:49
this service and and how hard is it to use because i
1:41:51
and how hard is it to use because i
1:41:51
and how hard is it to use because i imagine now if you want to include this
1:41:52
imagine now if you want to include this
1:41:52
imagine now if you want to include this into your application
1:41:54
into your application
1:41:54
into your application there's things you have to do to make it
1:41:56
there's things you have to do to make it
1:41:56
there's things you have to do to make it work now
1:41:57
work now
1:41:57
work now well not so many and to to set it up the
1:42:00
well not so many and to to set it up the
1:42:00
well not so many and to to set it up the only extra thing i did was creating the
1:42:02
only extra thing i did was creating the
1:42:02
only extra thing i did was creating the resource so basically you create a
1:42:03
resource so basically you create a
1:42:03
resource so basically you create a resource group and
1:42:05
resource group and
1:42:05
resource group and the resource itself it will take you
1:42:06
the resource itself it will take you
1:42:06
the resource itself it will take you another two to three minutes
1:42:08
another two to three minutes
1:42:08
another two to three minutes or maybe yeah i think two minutes to be
1:42:10
or maybe yeah i think two minutes to be
1:42:10
or maybe yeah i think two minutes to be honest and then when you want to use it
1:42:13
honest and then when you want to use it
1:42:13
honest and then when you want to use it um the good thing is that
1:42:16
um the good thing is that
1:42:16
um the good thing is that if it comes to your model it's
1:42:19
if it comes to your model it's
1:42:20
if it comes to your model it's just you get you get a i have to publish
1:42:23
just you get you get a i have to publish
1:42:23
just you get you get a i have to publish it so i will publish it that's the thing
1:42:25
it so i will publish it that's the thing
1:42:25
it so i will publish it that's the thing sorry if not i cannot use it but you
1:42:27
sorry if not i cannot use it but you
1:42:27
sorry if not i cannot use it but you will see that you will get a url
1:42:29
will see that you will get a url
1:42:30
will see that you will get a url that's your prediction url and for the
1:42:31
that's your prediction url and for the
1:42:32
that's your prediction url and for the developers here
1:42:33
developers here
1:42:33
developers here i think uh what you see here is enough
1:42:35
i think uh what you see here is enough
1:42:35
i think uh what you see here is enough so basically
1:42:36
so basically
1:42:36
so basically you have uh you can use the api
1:42:39
you have uh you can use the api
1:42:39
you have uh you can use the api so this is all you need and i think of
1:42:41
so this is all you need and i think of
1:42:41
so this is all you need and i think of veronica dropped out but
1:42:43
veronica dropped out but
1:42:43
veronica dropped out but any other developer against you right
1:42:46
any other developer against you right
1:42:46
any other developer against you right you have all the information you need
1:42:47
you have all the information you need
1:42:47
you have all the information you need you just have your image
1:42:49
you just have your image
1:42:49
you just have your image you send it to the endpoint and give it
1:42:51
you send it to the endpoint and give it
1:42:51
you send it to the endpoint and give it the right header information and the
1:42:52
the right header information and the
1:42:52
the right header information and the prediction key
1:42:54
prediction key
1:42:54
prediction key and you will get your answer back in
1:42:55
and you will get your answer back in
1:42:56
and you will get your answer back in your application
1:42:57
your application
1:42:57
your application or if you don't want to code uh it's
1:43:00
or if you don't want to code uh it's
1:43:00
or if you don't want to code uh it's also possible
1:43:01
also possible
1:43:01
also possible as mentioned you can use powerapps um
1:43:04
as mentioned you can use powerapps um
1:43:04
as mentioned you can use powerapps um and that's basically sort of canvas
1:43:06
and that's basically sort of canvas
1:43:06
and that's basically sort of canvas where you can drag on components you
1:43:08
where you can drag on components you
1:43:08
where you can drag on components you have to write a little bit of code but
1:43:10
have to write a little bit of code but
1:43:10
have to write a little bit of code but just
1:43:11
just
1:43:11
just some lines right um and you use
1:43:14
some lines right um and you use
1:43:14
some lines right um and you use flow and then the flow will take this
1:43:16
flow and then the flow will take this
1:43:16
flow and then the flow will take this logic so you
1:43:17
logic so you
1:43:17
logic so you you basically write out or you drag out
1:43:21
you basically write out or you drag out
1:43:21
you basically write out or you drag out all the components you need to access
1:43:23
all the components you need to access
1:43:23
all the components you need to access this
1:43:24
this
1:43:24
this service and you send the the value back
1:43:27
service and you send the the value back
1:43:27
service and you send the the value back to your application
1:43:29
to your application
1:43:29
to your application it's really cool and i will say i have
1:43:32
it's really cool and i will say i have
1:43:32
it's really cool and i will say i have used this service
1:43:34
used this service
1:43:34
used this service one of the cool things and uh you see
1:43:36
one of the cool things and uh you see
1:43:36
one of the cool things and uh you see once you click
1:43:37
once you click
1:43:37
once you click got it you can also export these models
1:43:41
got it you can also export these models
1:43:41
got it you can also export these models and i built one that looks at me that
1:43:44
and i built one that looks at me that
1:43:44
and i built one that looks at me that looks at me
1:43:45
looks at me
1:43:45
looks at me from a web page and tells me whether i'm
1:43:48
from a web page and tells me whether i'm
1:43:48
from a web page and tells me whether i'm having a rock paper or scissors
1:43:50
having a rock paper or scissors
1:43:50
having a rock paper or scissors symbol up so i can play rock paper
1:43:52
symbol up so i can play rock paper
1:43:52
symbol up so i can play rock paper scissors
1:43:54
scissors
1:43:54
scissors the computer i have that demo it's
1:43:56
the computer i have that demo it's
1:43:56
the computer i have that demo it's fantastic
1:43:57
fantastic
1:43:57
fantastic and it's all running in the browser
1:43:59
and it's all running in the browser
1:43:59
and it's all running in the browser because you can actually export these
1:44:01
because you can actually export these
1:44:01
because you can actually export these models
1:44:03
models
1:44:03
models and you can also put it onto a docker
1:44:05
and you can also put it onto a docker
1:44:05
and you can also put it onto a docker container
1:44:06
container
1:44:06
container so basically you can play with it uh as
1:44:09
so basically you can play with it uh as
1:44:09
so basically you can play with it uh as much as you like
1:44:10
much as you like
1:44:10
much as you like and and pack it in the way you like to
1:44:12
and and pack it in the way you like to
1:44:12
and and pack it in the way you like to work with it
1:44:13
work with it
1:44:13
work with it so from really zero well not zero code
1:44:16
so from really zero well not zero code
1:44:16
so from really zero well not zero code but let's say one percent coding with
1:44:18
but let's say one percent coding with
1:44:18
but let's say one percent coding with powerapps
1:44:19
powerapps
1:44:19
powerapps to everything you want to do with it
1:44:22
to everything you want to do with it
1:44:22
to everything you want to do with it that's really cool and like i said those
1:44:24
that's really cool and like i said those
1:44:24
that's really cool and like i said those can be downloaded and you can actually
1:44:26
can be downloaded and you can actually
1:44:26
can be downloaded and you can actually look
1:44:26
look
1:44:26
look inside of the models there's a really
1:44:28
inside of the models there's a really
1:44:28
inside of the models there's a really cool program called
1:44:29
cool program called
1:44:30
cool program called netron if you look that up and if you
1:44:32
netron if you look that up and if you
1:44:32
netron if you look that up and if you download the onyx model you can actually
1:44:34
download the onyx model you can actually
1:44:34
download the onyx model you can actually see what's inside so a couple of
1:44:37
see what's inside so a couple of
1:44:37
see what's inside so a couple of questions
1:44:38
questions
1:44:38
questions uh one matt uh his wife is doing
1:44:41
uh one matt uh his wife is doing
1:44:41
uh one matt uh his wife is doing research about coding
1:44:42
research about coding
1:44:42
research about coding infant and pair interactions uh reading
1:44:44
infant and pair interactions uh reading
1:44:44
infant and pair interactions uh reading babies minds
1:44:46
babies minds
1:44:46
babies minds i wish i could read baby's minds
1:44:49
i wish i could read baby's minds
1:44:49
i wish i could read baby's minds that would be awesome um another one use
1:44:52
that would be awesome um another one use
1:44:52
that would be awesome um another one use case at a bank to greet customers
1:44:54
case at a bank to greet customers
1:44:54
case at a bank to greet customers as they come in do i need that many
1:44:56
as they come in do i need that many
1:44:56
as they come in do i need that many pictures of customers to train the model
1:44:58
pictures of customers to train the model
1:44:58
pictures of customers to train the model and identify the customer
1:45:00
and identify the customer
1:45:00
and identify the customer customer recognition is this is this is
1:45:02
customer recognition is this is this is
1:45:02
customer recognition is this is this is this a good service to do that kind of
1:45:03
this a good service to do that kind of
1:45:03
this a good service to do that kind of thing
1:45:04
thing
1:45:04
thing well there we have the ethical
1:45:06
well there we have the ethical
1:45:06
well there we have the ethical discussion right um
1:45:08
discussion right um
1:45:08
discussion right um your customers uh would need to give
1:45:11
your customers uh would need to give
1:45:11
your customers uh would need to give their permission for you to store the
1:45:13
their permission for you to store the
1:45:14
their permission for you to store the information to use the information
1:45:16
information to use the information
1:45:16
information to use the information you have the probability to take the
1:45:18
you have the probability to take the
1:45:18
you have the probability to take the data away to retrain your model
1:45:22
data away to retrain your model
1:45:22
data away to retrain your model now all that is possible of course but
1:45:24
now all that is possible of course but
1:45:24
now all that is possible of course but you have to think about it
1:45:25
you have to think about it
1:45:25
you have to think about it because you cannot just take pictures of
1:45:28
because you cannot just take pictures of
1:45:28
because you cannot just take pictures of your customers so in theory you can
1:45:29
your customers so in theory you can
1:45:29
your customers so in theory you can right because if all the cctv you can
1:45:31
right because if all the cctv you can
1:45:31
right because if all the cctv you can just classify
1:45:32
just classify
1:45:32
just classify your customers train it so you can
1:45:35
your customers train it so you can
1:45:35
your customers train it so you can recognize the customer
1:45:37
recognize the customer
1:45:37
recognize the customer but i think this is a nice ethical
1:45:39
but i think this is a nice ethical
1:45:40
but i think this is a nice ethical question
1:45:41
question
1:45:41
question yeah would it be correct to do so
1:45:45
yeah would it be correct to do so
1:45:45
yeah would it be correct to do so i guess it depends right like for
1:45:46
i guess it depends right like for
1:45:46
i guess it depends right like for example like when i when i talk to my
1:45:48
example like when i when i talk to my
1:45:48
example like when i when i talk to my phone it looks at my face and i'm okay
1:45:50
phone it looks at my face and i'm okay
1:45:50
phone it looks at my face and i'm okay with it because i told it
1:45:52
with it because i told it
1:45:52
with it because i told it uh so that's a really good question
1:45:53
uh so that's a really good question
1:45:53
uh so that's a really good question muhammad looks like it is indeed
1:45:55
muhammad looks like it is indeed
1:45:55
muhammad looks like it is indeed possible i know there's services that
1:45:56
possible i know there's services that
1:45:56
possible i know there's services that will do that uh but again you have to
1:45:58
will do that uh but again you have to
1:45:58
will do that uh but again you have to think about the ethical
1:45:59
think about the ethical
1:46:00
think about the ethical concerns there yeah but think about i
1:46:02
concerns there yeah but think about i
1:46:02
concerns there yeah but think about i said maybe not a bank but
1:46:03
said maybe not a bank but
1:46:03
said maybe not a bank but what for example people that suffer
1:46:05
what for example people that suffer
1:46:05
what for example people that suffer dementia right maybe they don't
1:46:07
dementia right maybe they don't
1:46:07
dementia right maybe they don't recognize their own family members now
1:46:09
recognize their own family members now
1:46:10
recognize their own family members now in that case
1:46:10
in that case
1:46:10
in that case you could build a very simple app to
1:46:14
you could build a very simple app to
1:46:14
you could build a very simple app to help a person recognize his own family
1:46:17
help a person recognize his own family
1:46:17
help a person recognize his own family members
1:46:18
members
1:46:18
members and of course you need all the consent
1:46:19
and of course you need all the consent
1:46:20
and of course you need all the consent from everyone but um
1:46:22
from everyone but um
1:46:22
from everyone but um if you get a consent then then this
1:46:25
if you get a consent then then this
1:46:25
if you get a consent then then this might be really helpful for people with
1:46:27
might be really helpful for people with
1:46:27
might be really helpful for people with memory problems correct correct uh
1:46:30
memory problems correct correct uh
1:46:30
memory problems correct correct uh someone's asking about the
1:46:32
someone's asking about the
1:46:32
someone's asking about the uh resource i'm gonna put it there in
1:46:34
uh resource i'm gonna put it there in
1:46:34
uh resource i'm gonna put it there in the chat again i think that's the right
1:46:35
the chat again i think that's the right
1:46:36
the chat again i think that's the right one let me double check it here
1:46:38
one let me double check it here
1:46:38
one let me double check it here uh yes indeed that is indeed the right
1:46:41
uh yes indeed that is indeed the right
1:46:41
uh yes indeed that is indeed the right uh
1:46:42
uh
1:46:42
uh one but it's a different that's an
1:46:44
one but it's a different that's an
1:46:44
one but it's a different that's an object detection one apologies
1:46:45
object detection one apologies
1:46:45
object detection one apologies but amy posted some really good links
1:46:48
but amy posted some really good links
1:46:48
but amy posted some really good links there for
1:46:49
there for
1:46:49
there for all of them uh make sure you do that
1:46:51
all of them uh make sure you do that
1:46:51
all of them uh make sure you do that look at those
1:46:52
look at those
1:46:52
look at those mary anything else to finish up before
1:46:54
mary anything else to finish up before
1:46:54
mary anything else to finish up before we start
1:46:55
we start
1:46:55
we start transitioning over to the next session
1:46:58
transitioning over to the next session
1:46:58
transitioning over to the next session that i'm super excited about
1:47:00
that i'm super excited about
1:47:00
that i'm super excited about i would just like to invite everyone
1:47:03
i would just like to invite everyone
1:47:03
i would just like to invite everyone that's really new to ai
1:47:04
that's really new to ai
1:47:04
that's really new to ai i mean just just start and and take your
1:47:07
i mean just just start and and take your
1:47:07
i mean just just start and and take your starting path but
1:47:09
starting path but
1:47:09
starting path but by building these kind of models you've
1:47:11
by building these kind of models you've
1:47:11
by building these kind of models you've seen it it's not fake we build it live
1:47:14
seen it it's not fake we build it live
1:47:14
seen it it's not fake we build it live you can do that yourself and um so don't
1:47:17
you can do that yourself and um so don't
1:47:17
you can do that yourself and um so don't be afraid to start
1:47:19
be afraid to start
1:47:19
be afraid to start don't be afraid of all the math or
1:47:20
don't be afraid of all the math or
1:47:20
don't be afraid of all the math or whatever you can see that
1:47:22
whatever you can see that
1:47:22
whatever you can see that basically in a few minutes you built
1:47:24
basically in a few minutes you built
1:47:24
basically in a few minutes you built your own models and go play with it
1:47:26
your own models and go play with it
1:47:26
your own models and go play with it i would say awesome well this has been
1:47:29
i would say awesome well this has been
1:47:29
i would say awesome well this has been fantastical willem we're going to turn
1:47:31
fantastical willem we're going to turn
1:47:31
fantastical willem we're going to turn it back do you have any other questions
1:47:33
it back do you have any other questions
1:47:33
it back do you have any other questions for
1:47:33
for
1:47:33
for marion before she goes off and does more
1:47:36
marion before she goes off and does more
1:47:36
marion before she goes off and does more amazing things
1:47:55
actually i'm just
1:47:58
actually i'm just
1:47:58
actually i'm just just ah that's a good question i don't
1:48:01
just ah that's a good question i don't
1:48:01
just ah that's a good question i don't know the answer
1:48:05
upload images via an api um so if i
1:48:08
upload images via an api um so if i
1:48:08
upload images via an api um so if i were doing devops in a regular
1:48:10
were doing devops in a regular
1:48:10
were doing devops in a regular application
1:48:11
application
1:48:12
application is there any api that i can call to
1:48:14
is there any api that i can call to
1:48:14
is there any api that i can call to trigger the training cycle on this
1:48:16
trigger the training cycle on this
1:48:16
trigger the training cycle on this custom computer mission service
1:48:26
quite amazing i have to take classes
1:48:29
quite amazing i have to take classes
1:48:29
quite amazing i have to take classes from you sir
1:48:30
from you sir
1:48:30
from you sir well maybe i should take classes from
1:48:32
well maybe i should take classes from
1:48:32
well maybe i should take classes from you
1:48:33
you
1:48:33
you and then we take classes from each other
1:48:36
and then we take classes from each other
1:48:36
and then we take classes from each other it's like learning it's so cool
1:48:41
what why don't we do something we should
1:48:44
what why don't we do something we should
1:48:44
what why don't we do something we should do something where we like all get
1:48:45
do something where we like all get
1:48:46
do something where we like all get together and then maybe stream it out
1:48:48
together and then maybe stream it out
1:48:48
together and then maybe stream it out and maybe talk about the stuff that we
1:48:49
and maybe talk about the stuff that we
1:48:50
and maybe talk about the stuff that we know
1:48:50
know
1:48:50
know we should do something just like that i
1:48:53
we should do something just like that i
1:48:53
we should do something just like that i mean
1:48:58
okay
1:49:04
i mean yeah
1:49:14
i mean yeah
1:49:14
i mean yeah yeah i think well we've got some plans
1:49:17
yeah i think well we've got some plans
1:49:17
yeah i think well we've got some plans so we'll get back to that
1:49:18
so we'll get back to that
1:49:18
so we'll get back to that um in the next next few months yeah
1:49:22
um in the next next few months yeah
1:49:22
um in the next next few months yeah cool oh yeah uh
1:49:26
cool oh yeah uh
1:49:26
cool oh yeah uh let's well maybe maybe let's wait for a
1:49:28
let's well maybe maybe let's wait for a
1:49:28
let's well maybe maybe let's wait for a little bit because we have uh our next
1:49:30
little bit because we have uh our next
1:49:30
little bit because we have uh our next uh
1:49:31
uh
1:49:31
uh guest coming up uh richard campbell is
1:49:33
guest coming up uh richard campbell is
1:49:33
guest coming up uh richard campbell is going to join us
1:49:34
going to join us
1:49:34
going to join us and and he has a pretty interesting talk
1:49:37
and and he has a pretty interesting talk
1:49:37
and and he has a pretty interesting talk about the history of ai
1:49:38
about the history of ai
1:49:38
about the history of ai actually uh we're going to look back at
1:49:41
actually uh we're going to look back at
1:49:41
actually uh we're going to look back at how it all began and how we had these
1:49:43
how it all began and how we had these
1:49:43
how it all began and how we had these winters and it's like the ice age in
1:49:46
winters and it's like the ice age in
1:49:46
winters and it's like the ice age in here
1:49:46
here
1:49:46
here um so i'm really happy to have you
1:49:48
um so i'm really happy to have you
1:49:48
um so i'm really happy to have you marion thank you very much for
1:49:50
marion thank you very much for
1:49:50
marion thank you very much for the cool demo uh and then showing us all
1:49:53
the cool demo uh and then showing us all
1:49:53
the cool demo uh and then showing us all the stuff and how you got started
1:49:56
the stuff and how you got started
1:49:56
the stuff and how you got started so do tweet us your questions
1:50:00
so do tweet us your questions
1:50:00
so do tweet us your questions uh remember to include the c-sharp
1:50:03
uh remember to include the c-sharp
1:50:03
uh remember to include the c-sharp corner and the global ai community
1:50:05
corner and the global ai community
1:50:05
corner and the global ai community hashtag in there
1:50:07
hashtag in there
1:50:07
hashtag in there because we're giving away free stuff
1:50:09
because we're giving away free stuff
1:50:09
because we're giving away free stuff fifty dollars gift card
1:50:11
fifty dollars gift card
1:50:11
fifty dollars gift card uh from amazon and i'm here i'm well sir
1:50:14
uh from amazon and i'm here i'm well sir
1:50:14
uh from amazon and i'm here i'm well sir how are you
1:50:14
how are you
1:50:14
how are you we're choosing the best the most fun
1:50:16
we're choosing the best the most fun
1:50:16
we're choosing the best the most fun tweeting i love my audio work and
1:50:18
tweeting i love my audio work and
1:50:18
tweeting i love my audio work and everything
1:50:19
everything
1:50:19
everything we have a good technology moment so so
1:50:21
we have a good technology moment so so
1:50:21
we have a good technology moment so so far things are working well richard
1:50:24
far things are working well richard
1:50:24
far things are working well richard always works across all technologies uh
1:50:27
always works across all technologies uh
1:50:27
always works across all technologies uh yeah i wish that was true friend but
1:50:29
yeah i wish that was true friend but
1:50:30
yeah i wish that was true friend but uh the pile of gadgets involved in
1:50:32
uh the pile of gadgets involved in
1:50:32
uh the pile of gadgets involved in making all this look work
1:50:33
making all this look work
1:50:33
making all this look work and work well is not trivial yeah pretty
1:50:36
and work well is not trivial yeah pretty
1:50:36
and work well is not trivial yeah pretty good right we're having that exciting
1:50:38
good right we're having that exciting
1:50:38
good right we're having that exciting evening here
1:50:47
that that's
1:50:56
not at all you should see the set that
1:50:59
not at all you should see the set that
1:50:59
not at all you should see the set that we have here i find myself
1:51:00
we have here i find myself
1:51:00
we have here i find myself doing history work a lot more these days
1:51:04
doing history work a lot more these days
1:51:04
doing history work a lot more these days uh
1:51:18
artificial intelligence now i know this
1:51:20
artificial intelligence now i know this
1:51:20
artificial intelligence now i know this is a very different crowd i mean
1:51:22
is a very different crowd i mean
1:51:22
is a very different crowd i mean mostly when you're talking to cardinals
1:51:24
mostly when you're talking to cardinals
1:51:24
mostly when you're talking to cardinals you're talking to quite elderly men
1:51:26
you're talking to quite elderly men
1:51:26
you're talking to quite elderly men who are not particularly technically
1:51:27
who are not particularly technically
1:51:28
who are not particularly technically savvy so there's a lot more explaining
1:51:29
savvy so there's a lot more explaining
1:51:29
savvy so there's a lot more explaining but i think
1:51:30
but i think
1:51:30
but i think well looking at the class the quality of
1:51:32
well looking at the class the quality of
1:51:32
well looking at the class the quality of questions coming from this audience
1:51:34
questions coming from this audience
1:51:34
questions coming from this audience they're a lot more experienced and a lot
1:51:36
they're a lot more experienced and a lot
1:51:36
they're a lot more experienced and a lot more technical so this is tuned
1:51:37
more technical so this is tuned
1:51:37
more technical so this is tuned differently i definitely want to hit
1:51:39
differently i definitely want to hit
1:51:39
differently i definitely want to hit some of the highlight points that have
1:51:41
some of the highlight points that have
1:51:41
some of the highlight points that have happened i think over the past
1:51:42
happened i think over the past
1:51:42
happened i think over the past 70 years that have led us to
1:51:45
70 years that have led us to
1:51:45
70 years that have led us to this moment in 2020
1:51:54
it's true uh and and me too i mean i
1:51:56
it's true uh and and me too i mean i
1:51:56
it's true uh and and me too i mean i know i
1:51:57
know i
1:51:57
know i look older but i'm not that old but uh
1:52:00
look older but i'm not that old but uh
1:52:00
look older but i'm not that old but uh you know the interplay between the
1:52:02
you know the interplay between the
1:52:02
you know the interplay between the hardware the software and the various
1:52:05
hardware the software and the various
1:52:05
hardware the software and the various philosophies that got labeled with
1:52:08
philosophies that got labeled with
1:52:08
philosophies that got labeled with artificial intelligence
1:52:09
artificial intelligence
1:52:09
artificial intelligence is interesting it's it has twisted and
1:52:12
is interesting it's it has twisted and
1:52:12
is interesting it's it has twisted and turned a number of times
1:52:15
turned a number of times
1:52:15
turned a number of times oh wow
1:52:22
all right well should we put this let's
1:52:24
all right well should we put this let's
1:52:24
all right well should we put this let's get the slide up
1:52:25
get the slide up
1:52:26
get the slide up i'll get the screen sharing piece
1:52:27
i'll get the screen sharing piece
1:52:27
i'll get the screen sharing piece working and make sure i share the right
1:52:30
working and make sure i share the right
1:52:30
working and make sure i share the right screen because there are many
1:52:32
screen because there are many
1:52:32
screen because there are many i know you recommended we have two
1:52:34
i know you recommended we have two
1:52:34
i know you recommended we have two monitors but i'm not a two monitor kind
1:52:36
monitors but i'm not a two monitor kind
1:52:36
monitors but i'm not a two monitor kind of guy
1:52:37
of guy
1:52:37
of guy really so there's many monitors for me
1:52:40
really so there's many monitors for me
1:52:40
really so there's many monitors for me uh but that's not a bad thing either
1:52:42
uh but that's not a bad thing either
1:52:42
uh but that's not a bad thing either uh i thanks so much for letting me
1:52:44
uh i thanks so much for letting me
1:52:44
uh i thanks so much for letting me participate in this uh it's a great
1:52:46
participate in this uh it's a great
1:52:46
participate in this uh it's a great conversation it's something i think is
1:52:47
conversation it's something i think is
1:52:47
conversation it's something i think is really important
1:52:48
really important
1:52:48
really important because it's easy to look at the
1:52:50
because it's easy to look at the
1:52:50
because it's easy to look at the extraordinary set of technology we have
1:52:52
extraordinary set of technology we have
1:52:52
extraordinary set of technology we have in front of us today just say
1:52:54
in front of us today just say
1:52:54
in front of us today just say and not be clear that there were
1:52:56
and not be clear that there were
1:52:56
and not be clear that there were definitely we are building on the
1:52:58
definitely we are building on the
1:52:58
definitely we are building on the shoulders of giants many many giants and
1:53:00
shoulders of giants many many giants and
1:53:00
shoulders of giants many many giants and i want to start the story
1:53:02
i want to start the story
1:53:02
i want to start the story with one of the giants of computing of
1:53:04
with one of the giants of computing of
1:53:04
with one of the giants of computing of all time ever
1:53:05
all time ever
1:53:05
all time ever alan turing so alan turing
1:53:09
alan turing so alan turing
1:53:09
alan turing so alan turing is not the only person behind modern
1:53:11
is not the only person behind modern
1:53:12
is not the only person behind modern computing but he's one of the principal
1:53:13
computing but he's one of the principal
1:53:13
computing but he's one of the principal people
1:53:14
people
1:53:14
people and he did not going to turn artificial
1:53:16
and he did not going to turn artificial
1:53:16
and he did not going to turn artificial intelligence but he did
1:53:18
intelligence but he did
1:53:18
intelligence but he did ask the question in 1950 in a paper
1:53:21
ask the question in 1950 in a paper
1:53:21
ask the question in 1950 in a paper he said quite simply can machines think
1:53:24
he said quite simply can machines think
1:53:24
he said quite simply can machines think and he described this concept of what he
1:53:26
and he described this concept of what he
1:53:26
and he described this concept of what he called
1:53:27
called
1:53:27
called the imitation game we would later name
1:53:30
the imitation game we would later name
1:53:30
the imitation game we would later name it
1:53:30
it
1:53:30
it the turing test but he would certainly
1:53:33
the turing test but he would certainly
1:53:33
the turing test but he would certainly have never named it that way
1:53:35
have never named it that way
1:53:35
have never named it that way and so you think about in 1950 the the
1:53:38
and so you think about in 1950 the the
1:53:38
and so you think about in 1950 the the world war ii is not long past they're
1:53:41
world war ii is not long past they're
1:53:41
world war ii is not long past they're still doing recovery
1:53:42
still doing recovery
1:53:42
still doing recovery and the modern computer is not really
1:53:45
and the modern computer is not really
1:53:45
and the modern computer is not really born yet we have
1:53:46
born yet we have
1:53:46
born yet we have computers but they are pretty bespoke
1:53:49
computers but they are pretty bespoke
1:53:49
computers but they are pretty bespoke they're all one-of-a-kinds
1:53:51
they're all one-of-a-kinds
1:53:51
they're all one-of-a-kinds programming them is quite complex we're
1:53:52
programming them is quite complex we're
1:53:52
programming them is quite complex we're only just beginning to get to the idea
1:53:54
only just beginning to get to the idea
1:53:54
only just beginning to get to the idea of programming languages and already the
1:53:57
of programming languages and already the
1:53:57
of programming languages and already the question of
1:53:58
question of
1:53:58
question of can machines think is starting to
1:54:00
can machines think is starting to
1:54:00
can machines think is starting to surface and
1:54:01
surface and
1:54:01
surface and that paper is referred to
1:54:05
that paper is referred to
1:54:05
that paper is referred to as sort of a seminal work that leads in
1:54:08
as sort of a seminal work that leads in
1:54:08
as sort of a seminal work that leads in the 50s to broader conversations
1:54:10
the 50s to broader conversations
1:54:10
the 50s to broader conversations so the first time you really hear the
1:54:13
so the first time you really hear the
1:54:13
so the first time you really hear the phrase artificial intelligence
1:54:15
phrase artificial intelligence
1:54:15
phrase artificial intelligence is in the summer of 1956 when a
1:54:17
is in the summer of 1956 when a
1:54:17
is in the summer of 1956 when a gathering of these
1:54:19
gathering of these
1:54:19
gathering of these computing scientists mathematicians
1:54:22
computing scientists mathematicians
1:54:22
computing scientists mathematicians and engineers started to think through
1:54:25
and engineers started to think through
1:54:25
and engineers started to think through can we
1:54:25
can we
1:54:25
can we create an artificial intelligence now
1:54:29
create an artificial intelligence now
1:54:29
create an artificial intelligence now this was in dartmouth marvin minsky and
1:54:32
this was in dartmouth marvin minsky and
1:54:32
this was in dartmouth marvin minsky and many others
1:54:33
many others
1:54:33
many others together they've certainly celebrated
1:54:34
together they've certainly celebrated
1:54:34
together they've certainly celebrated all of this now that's only in
1:54:36
all of this now that's only in
1:54:36
all of this now that's only in academic circles that we really see that
1:54:39
academic circles that we really see that
1:54:39
academic circles that we really see that concept and they start with
1:54:40
concept and they start with
1:54:40
concept and they start with experimentations
1:54:42
experimentations
1:54:42
experimentations in again early machines and a lot of the
1:54:45
in again early machines and a lot of the
1:54:45
in again early machines and a lot of the funding for this
1:54:46
funding for this
1:54:46
funding for this comes from the military and so they've
1:54:49
comes from the military and so they've
1:54:49
comes from the military and so they've been sold this idea that we can build
1:54:51
been sold this idea that we can build
1:54:51
been sold this idea that we can build this remarkable machine they're already
1:54:52
this remarkable machine they're already
1:54:52
this remarkable machine they're already predicting by 1956
1:54:54
predicting by 1956
1:54:54
predicting by 1956 that in the 1980s artificial
1:54:56
that in the 1980s artificial
1:54:56
that in the 1980s artificial intelligence will be solved that will
1:54:58
intelligence will be solved that will
1:54:58
intelligence will be solved that will have an intelligence this has always
1:54:59
have an intelligence this has always
1:55:00
have an intelligence this has always been
1:55:00
been
1:55:00
been this play of trying to
1:55:04
this play of trying to
1:55:04
this play of trying to get funding to create something profound
1:55:08
get funding to create something profound
1:55:08
get funding to create something profound and along the way various technologies
1:55:10
and along the way various technologies
1:55:10
and along the way various technologies emerge and in that time
1:55:13
emerge and in that time
1:55:13
emerge and in that time a lot of the most important things that
1:55:14
a lot of the most important things that
1:55:14
a lot of the most important things that were invented had more to do with
1:55:16
were invented had more to do with
1:55:16
were invented had more to do with logistical planning which the military
1:55:18
logistical planning which the military
1:55:18
logistical planning which the military needed a lot of and continued to use
1:55:20
needed a lot of and continued to use
1:55:20
needed a lot of and continued to use the intelligence part that's a little
1:55:22
the intelligence part that's a little
1:55:22
the intelligence part that's a little more complex and in the 1960s
1:55:24
more complex and in the 1960s
1:55:24
more complex and in the 1960s funding is starting to wind down but
1:55:26
funding is starting to wind down but
1:55:26
funding is starting to wind down but computing is evolving
1:55:28
computing is evolving
1:55:28
computing is evolving and i think one of the cornerstones of
1:55:30
and i think one of the cornerstones of
1:55:30
and i think one of the cornerstones of that was a important man by the name of
1:55:32
that was a important man by the name of
1:55:32
that was a important man by the name of gordon moore
1:55:33
gordon moore
1:55:33
gordon moore and in 65 he wrote a paper as well
1:55:37
and in 65 he wrote a paper as well
1:55:37
and in 65 he wrote a paper as well and in that paper he talked about the
1:55:38
and in that paper he talked about the
1:55:38
and in that paper he talked about the beginning of the microprocessor which he
1:55:41
beginning of the microprocessor which he
1:55:41
beginning of the microprocessor which he he'd been helping to invent and he said
1:55:43
he'd been helping to invent and he said
1:55:43
he'd been helping to invent and he said he noticed a
1:55:44
he noticed a
1:55:44
he noticed a pattern of behavior that for a given
1:55:47
pattern of behavior that for a given
1:55:47
pattern of behavior that for a given amount of
1:55:48
amount of
1:55:48
amount of money the amount of transitions you
1:55:51
money the amount of transitions you
1:55:51
money the amount of transitions you could put on a piece of silicon
1:55:52
could put on a piece of silicon
1:55:52
could put on a piece of silicon doubled roughly every 18 months to two
1:55:56
doubled roughly every 18 months to two
1:55:56
doubled roughly every 18 months to two years
1:55:57
years
1:55:57
years now he didn't call it a law he called it
1:55:59
now he didn't call it a law he called it
1:55:59
now he didn't call it a law he called it an observation
1:56:00
an observation
1:56:00
an observation he certainly didn't call it moore's law
1:56:02
he certainly didn't call it moore's law
1:56:02
he certainly didn't call it moore's law other people called it that
1:56:04
other people called it that
1:56:04
other people called it that but it was this observation of this
1:56:05
but it was this observation of this
1:56:05
but it was this observation of this doubling of computing
1:56:08
doubling of computing
1:56:08
doubling of computing potential not necessarily speed
1:56:11
potential not necessarily speed
1:56:11
potential not necessarily speed but really for a given cost the amount
1:56:13
but really for a given cost the amount
1:56:14
but really for a given cost the amount of technology to be packed onto a chip
1:56:16
of technology to be packed onto a chip
1:56:16
of technology to be packed onto a chip expanded and i bring this up for a very
1:56:18
expanded and i bring this up for a very
1:56:18
expanded and i bring this up for a very important reason
1:56:19
important reason
1:56:19
important reason the evolution of artificial intelligence
1:56:22
the evolution of artificial intelligence
1:56:22
the evolution of artificial intelligence is bound to its hardware
1:56:24
is bound to its hardware
1:56:24
is bound to its hardware and so the iterations of hardware
1:56:26
and so the iterations of hardware
1:56:26
and so the iterations of hardware ultimately open new
1:56:27
ultimately open new
1:56:28
ultimately open new doors to the potential of computing now
1:56:30
doors to the potential of computing now
1:56:30
doors to the potential of computing now this is still a largely academic
1:56:32
this is still a largely academic
1:56:32
this is still a largely academic exercise and it's taking place mostly in
1:56:34
exercise and it's taking place mostly in
1:56:34
exercise and it's taking place mostly in universities with
1:56:35
universities with
1:56:35
universities with very large very expensive computers
1:56:38
very large very expensive computers
1:56:38
very large very expensive computers where every minute of compute time
1:56:40
where every minute of compute time
1:56:40
where every minute of compute time costs money and so you're seeing
1:56:43
costs money and so you're seeing
1:56:43
costs money and so you're seeing scientists
1:56:44
scientists
1:56:44
scientists pursue grants to do these experiments
1:56:47
pursue grants to do these experiments
1:56:47
pursue grants to do these experiments now by the time you get into the 60s
1:56:50
now by the time you get into the 60s
1:56:50
now by the time you get into the 60s we're getting some pretty interesting
1:56:51
we're getting some pretty interesting
1:56:51
we're getting some pretty interesting pieces of software
1:56:53
pieces of software
1:56:53
pieces of software and joseph weisenbaum in 1964 built an
1:56:56
and joseph weisenbaum in 1964 built an
1:56:56
and joseph weisenbaum in 1964 built an app he called
1:56:57
app he called
1:56:57
app he called eliza you could call this the first chat
1:57:01
eliza you could call this the first chat
1:57:01
eliza you could call this the first chat bot and it was invented even before i
1:57:03
bot and it was invented even before i
1:57:03
bot and it was invented even before i was born
1:57:04
was born
1:57:04
was born what it was actually doing was
1:57:05
what it was actually doing was
1:57:05
what it was actually doing was simulating what was called rogerian
1:57:07
simulating what was called rogerian
1:57:07
simulating what was called rogerian psychotherapy this is after a
1:57:09
psychotherapy this is after a
1:57:09
psychotherapy this is after a psychotherapist with the name of
1:57:10
psychotherapist with the name of
1:57:10
psychotherapist with the name of carl rogers who was famous for sort of
1:57:12
carl rogers who was famous for sort of
1:57:12
carl rogers who was famous for sort of parroting back
1:57:13
parroting back
1:57:13
parroting back what the patients said to get them to
1:57:15
what the patients said to get them to
1:57:15
what the patients said to get them to explore it further
1:57:17
explore it further
1:57:17
explore it further and this has been a sample application
1:57:19
and this has been a sample application
1:57:19
and this has been a sample application that has followed us around for many
1:57:20
that has followed us around for many
1:57:20
that has followed us around for many many many years i've seen this on pcs
1:57:22
many many years i've seen this on pcs
1:57:22
many many years i've seen this on pcs and so forth
1:57:23
and so forth
1:57:23
and so forth even this screenshot is from a much more
1:57:25
even this screenshot is from a much more
1:57:25
even this screenshot is from a much more modern implementation
1:57:26
modern implementation
1:57:26
modern implementation of the one that it once existed in the
1:57:28
of the one that it once existed in the
1:57:28
of the one that it once existed in the 60s
1:57:29
60s
1:57:30
60s what's interesting about these early
1:57:32
what's interesting about these early
1:57:32
what's interesting about these early days is already we had these
1:57:35
days is already we had these
1:57:35
days is already we had these pictures of entities in artificial
1:57:38
pictures of entities in artificial
1:57:38
pictures of entities in artificial intelligence and it pressed against the
1:57:41
intelligence and it pressed against the
1:57:41
intelligence and it pressed against the human's tendency to anthropomorphize
1:57:44
human's tendency to anthropomorphize
1:57:44
human's tendency to anthropomorphize fundamentally humans sort of look for
1:57:47
fundamentally humans sort of look for
1:57:47
fundamentally humans sort of look for humanity elsewhere
1:57:49
humanity elsewhere
1:57:49
humanity elsewhere we look for faces we see them everywhere
1:57:51
we look for faces we see them everywhere
1:57:51
we look for faces we see them everywhere i mean folks see faces
1:57:52
i mean folks see faces
1:57:52
i mean folks see faces in trees they see them in sporting
1:57:56
in trees they see them in sporting
1:57:56
in trees they see them in sporting equipment
1:57:57
equipment
1:57:57
equipment they even see them in mountains on mars
1:58:01
they even see them in mountains on mars
1:58:01
they even see them in mountains on mars so we are prone to
1:58:06
so we are prone to
1:58:06
so we are prone to giving entity or agency to these tools
1:58:09
giving entity or agency to these tools
1:58:09
giving entity or agency to these tools that we've created
1:58:10
that we've created
1:58:10
that we've created uh even though they're not really there
1:58:13
uh even though they're not really there
1:58:13
uh even though they're not really there and the ultimate manifestation of this
1:58:14
and the ultimate manifestation of this
1:58:14
and the ultimate manifestation of this and really the first time that the
1:58:16
and really the first time that the
1:58:16
and really the first time that the phrase artificial intelligence appears
1:58:18
phrase artificial intelligence appears
1:58:18
phrase artificial intelligence appears in the social conscious and the
1:58:19
in the social conscious and the
1:58:20
in the social conscious and the broader population is in 1968
1:58:23
broader population is in 1968
1:58:23
broader population is in 1968 in a movie called 2001 a space odyssey
1:58:26
in a movie called 2001 a space odyssey
1:58:26
in a movie called 2001 a space odyssey now this movie is important for a bunch
1:58:28
now this movie is important for a bunch
1:58:28
now this movie is important for a bunch of reasons in fact
1:58:29
of reasons in fact
1:58:29
of reasons in fact most directors that make space
1:58:33
most directors that make space
1:58:33
most directors that make space movies still refer to this movie as sort
1:58:35
movies still refer to this movie as sort
1:58:35
movies still refer to this movie as sort of the definitive work
1:58:37
of the definitive work
1:58:37
of the definitive work kubrick as much whatever you may have
1:58:39
kubrick as much whatever you may have
1:58:39
kubrick as much whatever you may have thought of him
1:58:40
thought of him
1:58:40
thought of him he pushed the envelope before man had
1:58:42
he pushed the envelope before man had
1:58:42
he pushed the envelope before man had landed on the moon
1:58:43
landed on the moon
1:58:43
landed on the moon he had already made a movie about
1:58:45
he had already made a movie about
1:58:45
he had already made a movie about mankind living in space and
1:58:48
mankind living in space and
1:58:48
mankind living in space and he had commissioned arthur c clarke as
1:58:51
he had commissioned arthur c clarke as
1:58:51
he had commissioned arthur c clarke as his
1:58:51
his
1:58:51
his futurist they actually collaborated over
1:58:53
futurist they actually collaborated over
1:58:53
futurist they actually collaborated over making the movie
1:58:54
making the movie
1:58:54
making the movie the book would come later so this was
1:58:57
the book would come later so this was
1:58:57
the book would come later so this was not a book
1:58:58
not a book
1:58:58
not a book adaptation in a movie rather was a
1:59:00
adaptation in a movie rather was a
1:59:00
adaptation in a movie rather was a vision of a movie that became a book but
1:59:02
vision of a movie that became a book but
1:59:02
vision of a movie that became a book but arthur c clarke
1:59:03
arthur c clarke
1:59:03
arthur c clarke is a futurist he's passed now but
1:59:06
is a futurist he's passed now but
1:59:06
is a futurist he's passed now but in by that time he had already written
1:59:08
in by that time he had already written
1:59:08
in by that time he had already written about geostationary satellites
1:59:10
about geostationary satellites
1:59:10
about geostationary satellites a decade before they could be built and
1:59:13
a decade before they could be built and
1:59:13
a decade before they could be built and they talked about
1:59:14
they talked about
1:59:14
they talked about machine intelligence and they used the
1:59:16
machine intelligence and they used the
1:59:16
machine intelligence and they used the phrase artificial intelligence and said
1:59:18
phrase artificial intelligence and said
1:59:18
phrase artificial intelligence and said that
1:59:19
that
1:59:19
that by the year 2001 we will have machines
1:59:21
by the year 2001 we will have machines
1:59:21
by the year 2001 we will have machines with intelligence that
1:59:22
with intelligence that
1:59:22
with intelligence that have matched or exceeded humans now the
1:59:25
have matched or exceeded humans now the
1:59:25
have matched or exceeded humans now the downside to this as cool as it may be is
1:59:27
downside to this as cool as it may be is
1:59:27
downside to this as cool as it may be is of course that
1:59:28
of course that
1:59:28
of course that that artificial intelligence that first
1:59:30
that artificial intelligence that first
1:59:30
that artificial intelligence that first time that it's seen in the public
1:59:32
time that it's seen in the public
1:59:32
time that it's seen in the public the machine tries to kill everybody and
1:59:35
the machine tries to kill everybody and
1:59:35
the machine tries to kill everybody and so we set a tone
1:59:36
so we set a tone
1:59:36
so we set a tone right from the very beginning to the
1:59:38
right from the very beginning to the
1:59:38
right from the very beginning to the rest of the world
1:59:40
rest of the world
1:59:40
rest of the world that yes we're going to make this
1:59:42
that yes we're going to make this
1:59:42
that yes we're going to make this artificial intelligence and yeah it's
1:59:44
artificial intelligence and yeah it's
1:59:44
artificial intelligence and yeah it's going to try and kill everybody you know
1:59:45
going to try and kill everybody you know
1:59:45
going to try and kill everybody you know terminators all those things will come
1:59:47
terminators all those things will come
1:59:47
terminators all those things will come later
1:59:48
later
1:59:48
later but for better for worse with for us as
1:59:50
but for better for worse with for us as
1:59:50
but for better for worse with for us as technologists
1:59:52
technologists
1:59:52
technologists we've kind of set ourselves up into this
1:59:54
we've kind of set ourselves up into this
1:59:54
we've kind of set ourselves up into this current
1:59:55
current
1:59:55
current challenge i'm not saying the ethical
1:59:56
challenge i'm not saying the ethical
1:59:56
challenge i'm not saying the ethical debate isn't important
1:59:58
debate isn't important
1:59:58
debate isn't important and and something we absolutely need to
2:00:00
and and something we absolutely need to
2:00:00
and and something we absolutely need to focus on but understand
2:00:02
focus on but understand
2:00:02
focus on but understand that popular culture has also grabbed
2:00:05
that popular culture has also grabbed
2:00:05
that popular culture has also grabbed onto this anthropomorphization
2:00:07
onto this anthropomorphization
2:00:07
onto this anthropomorphization and created perceptions that are
2:00:10
and created perceptions that are
2:00:10
and created perceptions that are incorrect
2:00:11
incorrect
2:00:11
incorrect if you're working in this space you know
2:00:13
if you're working in this space you know
2:00:13
if you're working in this space you know these things aren't
2:00:14
these things aren't
2:00:14
these things aren't true and that we don't have artificial
2:00:17
true and that we don't have artificial
2:00:17
true and that we don't have artificial generalized intelligence we're still
2:00:19
generalized intelligence we're still
2:00:19
generalized intelligence we're still exploring what that would even mean for
2:00:20
exploring what that would even mean for
2:00:20
exploring what that would even mean for us we build specialized intelligences
2:00:22
us we build specialized intelligences
2:00:22
us we build specialized intelligences and so more often than not when you end
2:00:25
and so more often than not when you end
2:00:25
and so more often than not when you end up talking
2:00:26
up talking
2:00:26
up talking in public about office diligence comes
2:00:28
in public about office diligence comes
2:00:28
in public about office diligence comes across more as artificial stupidity
2:00:30
across more as artificial stupidity
2:00:30
across more as artificial stupidity and by the end of the 60s the promises
2:00:33
and by the end of the 60s the promises
2:00:33
and by the end of the 60s the promises that have been made in the 50s
2:00:35
that have been made in the 50s
2:00:36
that have been made in the 50s didn't measure up and again i think
2:00:38
didn't measure up and again i think
2:00:38
didn't measure up and again i think movies didn't help that
2:00:40
movies didn't help that
2:00:40
movies didn't help that here we were able to create these
2:00:41
here we were able to create these
2:00:41
here we were able to create these simulations of intelligences
2:00:43
simulations of intelligences
2:00:43
simulations of intelligences we just couldn't actually build them and
2:00:45
we just couldn't actually build them and
2:00:45
we just couldn't actually build them and so a lot of military funding and general
2:00:46
so a lot of military funding and general
2:00:46
so a lot of military funding and general funding dried up and this is the first
2:00:48
funding dried up and this is the first
2:00:48
funding dried up and this is the first time we hear this phrase
2:00:49
time we hear this phrase
2:00:49
time we hear this phrase ai winter so the the decrease in funding
2:00:53
ai winter so the the decrease in funding
2:00:53
ai winter so the the decrease in funding that didn't mean that the technology
2:00:55
that didn't mean that the technology
2:00:55
that didn't mean that the technology never worked they did build important
2:00:57
never worked they did build important
2:00:57
never worked they did build important and powerful things
2:00:59
and powerful things
2:00:59
and powerful things that benefited the folks that were using
2:01:01
that benefited the folks that were using
2:01:01
that benefited the folks that were using them and continue to this day
2:01:02
them and continue to this day
2:01:02
them and continue to this day the logistical engine of the us military
2:01:05
the logistical engine of the us military
2:01:05
the logistical engine of the us military is a profound
2:01:06
is a profound
2:01:06
is a profound capability that exceeds every other
2:01:09
capability that exceeds every other
2:01:09
capability that exceeds every other nation in the world
2:01:11
nation in the world
2:01:12
nation in the world and the technology still continued but
2:01:14
and the technology still continued but
2:01:14
and the technology still continued but the money that was available in
2:01:15
the money that was available in
2:01:15
the money that was available in for research and inside of universities
2:01:18
for research and inside of universities
2:01:18
for research and inside of universities not as much
2:01:19
not as much
2:01:19
not as much and so when you get to the late 70s
2:01:21
and so when you get to the late 70s
2:01:22
and so when you get to the late 70s after a quieter period you do see a new
2:01:24
after a quieter period you do see a new
2:01:24
after a quieter period you do see a new generation
2:01:25
generation
2:01:25
generation of what was then wrapped in the ai
2:01:27
of what was then wrapped in the ai
2:01:27
of what was then wrapped in the ai banner again so this is the
2:01:29
banner again so this is the
2:01:29
banner again so this is the expert system model and this is in
2:01:32
expert system model and this is in
2:01:32
expert system model and this is in the late 70s early 80s which is also the
2:01:34
the late 70s early 80s which is also the
2:01:34
the late 70s early 80s which is also the emergence of the personal computer
2:01:37
emergence of the personal computer
2:01:37
emergence of the personal computer so when you think in terms of the apple
2:01:39
so when you think in terms of the apple
2:01:39
so when you think in terms of the apple ii and the ibm pc
2:01:42
ii and the ibm pc
2:01:42
ii and the ibm pc that gordon moore has driven down the
2:01:45
that gordon moore has driven down the
2:01:45
that gordon moore has driven down the price of compute to the point where you
2:01:46
price of compute to the point where you
2:01:46
price of compute to the point where you can have a machine on your desk rather
2:01:48
can have a machine on your desk rather
2:01:48
can have a machine on your desk rather than require
2:01:49
than require
2:01:50
than require university scale machines and so
2:01:53
university scale machines and so
2:01:53
university scale machines and so this generation of expert systems where
2:01:55
this generation of expert systems where
2:01:55
this generation of expert systems where they're capturing expert knowledge
2:01:57
they're capturing expert knowledge
2:01:57
they're capturing expert knowledge largely in decision tree models so
2:02:00
largely in decision tree models so
2:02:00
largely in decision tree models so called almost like a super eliza where
2:02:02
called almost like a super eliza where
2:02:02
called almost like a super eliza where it's just
2:02:03
it's just
2:02:03
it's just able to answer questions but in order to
2:02:05
able to answer questions but in order to
2:02:05
able to answer questions but in order to have them run
2:02:06
have them run
2:02:06
have them run fast enough in those early computers
2:02:08
fast enough in those early computers
2:02:08
fast enough in those early computers they're building custom
2:02:10
they're building custom
2:02:10
they're building custom super pcs and so by the late 80s
2:02:14
super pcs and so by the late 80s
2:02:14
super pcs and so by the late 80s there are a set of machines they called
2:02:16
there are a set of machines they called
2:02:16
there are a set of machines they called lisp machines because they were largely
2:02:17
lisp machines because they were largely
2:02:17
lisp machines because they were largely programmed in the lisp
2:02:19
programmed in the lisp
2:02:19
programmed in the lisp and they where the typical ibm pc of the
2:02:21
and they where the typical ibm pc of the
2:02:21
and they where the typical ibm pc of the day was
2:02:22
day was
2:02:22
day was three to four thousand dollars u.s these
2:02:25
three to four thousand dollars u.s these
2:02:25
three to four thousand dollars u.s these are fifty to a hundred thousand dollars
2:02:27
are fifty to a hundred thousand dollars
2:02:28
are fifty to a hundred thousand dollars a piece
2:02:28
a piece
2:02:28
a piece and there's about a dozen vendors
2:02:30
and there's about a dozen vendors
2:02:30
and there's about a dozen vendors selling them and
2:02:32
selling them and
2:02:32
selling them and they're not delivering the value for the
2:02:34
they're not delivering the value for the
2:02:34
they're not delivering the value for the money that cost at the end by the 1987
2:02:37
money that cost at the end by the 1987
2:02:37
money that cost at the end by the 1987 sort of the end of this generation of
2:02:39
sort of the end of this generation of
2:02:39
sort of the end of this generation of competing about less than 10 000 of
2:02:41
competing about less than 10 000 of
2:02:41
competing about less than 10 000 of these machines have been sold
2:02:42
these machines have been sold
2:02:42
these machines have been sold and most of those companies go out of
2:02:44
and most of those companies go out of
2:02:44
and most of those companies go out of business and so you get
2:02:46
business and so you get
2:02:46
business and so you get a second ai winter towards the late 80s
2:02:49
a second ai winter towards the late 80s
2:02:49
a second ai winter towards the late 80s now all the models i've talked about so
2:02:51
now all the models i've talked about so
2:02:51
now all the models i've talked about so far were very much decision tree models
2:02:53
far were very much decision tree models
2:02:53
far were very much decision tree models eliza certainly a decision tree model
2:02:55
eliza certainly a decision tree model
2:02:56
eliza certainly a decision tree model and and all of these expert systems fall
2:02:58
and and all of these expert systems fall
2:02:58
and and all of these expert systems fall in that class and it
2:02:59
in that class and it
2:02:59
in that class and it brings this conversation forward that
2:03:01
brings this conversation forward that
2:03:01
brings this conversation forward that emerged at that time
2:03:03
emerged at that time
2:03:03
emerged at that time around this concept called the meets
2:03:04
around this concept called the meets
2:03:04
around this concept called the meets versus the scruffy so the needs
2:03:07
versus the scruffy so the needs
2:03:07
versus the scruffy so the needs and let's face it we're talking about
2:03:08
and let's face it we're talking about
2:03:08
and let's face it we're talking about seth juarez here he's very neat
2:03:10
seth juarez here he's very neat
2:03:10
seth juarez here he's very neat uh are about logical and symbolic
2:03:12
uh are about logical and symbolic
2:03:12
uh are about logical and symbolic reasoning
2:03:13
reasoning
2:03:13
reasoning so creating complex algorithms that have
2:03:16
so creating complex algorithms that have
2:03:16
so creating complex algorithms that have repeatable steps that are logical and
2:03:18
repeatable steps that are logical and
2:03:18
repeatable steps that are logical and validatable
2:03:19
validatable
2:03:19
validatable uh lisp and prolog being good at that
2:03:22
uh lisp and prolog being good at that
2:03:22
uh lisp and prolog being good at that this is
2:03:23
this is
2:03:23
this is most of the activity that happened up
2:03:24
most of the activity that happened up
2:03:24
most of the activity that happened up until now but there was an emerging
2:03:27
until now but there was an emerging
2:03:27
until now but there was an emerging group
2:03:27
group
2:03:27
group the scruffies that said that
2:03:29
the scruffies that said that
2:03:29
the scruffies that said that intelligence
2:03:30
intelligence
2:03:30
intelligence is just too complex to adhere to neat
2:03:33
is just too complex to adhere to neat
2:03:33
is just too complex to adhere to neat methodologies
2:03:34
methodologies
2:03:34
methodologies and so now they're getting more into the
2:03:36
and so now they're getting more into the
2:03:36
and so now they're getting more into the neural net models
2:03:37
neural net models
2:03:38
neural net models object inherited models that
2:03:41
object inherited models that
2:03:41
object inherited models that created their own complexities in the
2:03:42
created their own complexities in the
2:03:42
created their own complexities in the sense that they were harder to measure
2:03:44
sense that they were harder to measure
2:03:44
sense that they were harder to measure harder to validate
2:03:45
harder to validate
2:03:45
harder to validate but they created remarkable results we
2:03:48
but they created remarkable results we
2:03:48
but they created remarkable results we know today with the benefit of hindsight
2:03:51
know today with the benefit of hindsight
2:03:51
know today with the benefit of hindsight that in the late 80s early 90s when they
2:03:54
that in the late 80s early 90s when they
2:03:54
that in the late 80s early 90s when they were talking about this that the
2:03:56
were talking about this that the
2:03:56
were talking about this that the scruffies were going to have a good day
2:03:57
scruffies were going to have a good day
2:03:58
scruffies were going to have a good day but it didn't mean that needs didn't
2:03:59
but it didn't mean that needs didn't
2:03:59
but it didn't mean that needs didn't continue to progress and by 1997
2:04:02
continue to progress and by 1997
2:04:02
continue to progress and by 1997 you get i would argue one of the
2:04:04
you get i would argue one of the
2:04:04
you get i would argue one of the greatest wins for the needs of all time
2:04:05
greatest wins for the needs of all time
2:04:06
greatest wins for the needs of all time in a computer built by ibm so notice
2:04:08
in a computer built by ibm so notice
2:04:08
in a computer built by ibm so notice we've moved out of the university now
2:04:09
we've moved out of the university now
2:04:09
we've moved out of the university now we're talking
2:04:10
we're talking
2:04:10
we're talking about research groups and large
2:04:11
about research groups and large
2:04:11
about research groups and large corporations the project was called deep
2:04:14
corporations the project was called deep
2:04:14
corporations the project was called deep blue
2:04:15
blue
2:04:15
blue and it was to make a map a grand master
2:04:18
and it was to make a map a grand master
2:04:18
and it was to make a map a grand master class
2:04:19
class
2:04:19
class chess player now the performance of this
2:04:21
chess player now the performance of this
2:04:21
chess player now the performance of this machine you can see the size of it it is
2:04:23
machine you can see the size of it it is
2:04:24
machine you can see the size of it it is taller than you are it's about an 11
2:04:27
taller than you are it's about an 11
2:04:27
taller than you are it's about an 11 gigaflop
2:04:28
gigaflop
2:04:28
gigaflop mini computer and i would point out in
2:04:31
mini computer and i would point out in
2:04:31
mini computer and i would point out in 1997 11 gigaflops is a lot but that's
2:04:33
1997 11 gigaflops is a lot but that's
2:04:33
1997 11 gigaflops is a lot but that's about the same horsepower as an
2:04:34
about the same horsepower as an
2:04:34
about the same horsepower as an ipad 2 circuit 2011.
2:04:39
ipad 2 circuit 2011.
2:04:39
ipad 2 circuit 2011. but it did beat gary kasparov it was
2:04:41
but it did beat gary kasparov it was
2:04:41
but it did beat gary kasparov it was became the best chips player in the
2:04:43
became the best chips player in the
2:04:43
became the best chips player in the world and considered a remarkable
2:04:44
world and considered a remarkable
2:04:44
world and considered a remarkable victory now again ibm continued on
2:04:46
victory now again ibm continued on
2:04:46
victory now again ibm continued on this technology is the origins for
2:04:49
this technology is the origins for
2:04:49
this technology is the origins for watson and watson hits its peak in 2011
2:04:54
watson and watson hits its peak in 2011
2:04:54
watson and watson hits its peak in 2011 with its ability to win at jeopardy
2:04:58
with its ability to win at jeopardy
2:04:58
with its ability to win at jeopardy and so uh what the watson technology
2:05:00
and so uh what the watson technology
2:05:00
and so uh what the watson technology communities today you can participate in
2:05:01
communities today you can participate in
2:05:01
communities today you can participate in if you want they have a lot of
2:05:03
if you want they have a lot of
2:05:03
if you want they have a lot of licensing opportunities and they heavily
2:05:05
licensing opportunities and they heavily
2:05:05
licensing opportunities and they heavily into the medical spaces and so forth
2:05:07
into the medical spaces and so forth
2:05:07
into the medical spaces and so forth but it is important to remember that it
2:05:09
but it is important to remember that it
2:05:09
but it is important to remember that it falls into the needs category very much
2:05:10
falls into the needs category very much
2:05:10
falls into the needs category very much the decision tree models and they've
2:05:12
the decision tree models and they've
2:05:12
the decision tree models and they've really
2:05:12
really
2:05:12
really stretched that model to the lengths they
2:05:15
stretched that model to the lengths they
2:05:15
stretched that model to the lengths they possibly can
2:05:17
possibly can
2:05:17
possibly can let's change gears and jump to the
2:05:19
let's change gears and jump to the
2:05:19
let's change gears and jump to the scruffy conversation
2:05:20
scruffy conversation
2:05:20
scruffy conversation and i want to talk about one particular
2:05:23
and i want to talk about one particular
2:05:23
and i want to talk about one particular scientist
2:05:24
scientist
2:05:24
scientist jeffrey hinton now i have my personal
2:05:25
jeffrey hinton now i have my personal
2:05:25
jeffrey hinton now i have my personal biases toward jeffrey hinden
2:05:27
biases toward jeffrey hinden
2:05:27
biases toward jeffrey hinden he happens to live in canada has a do i
2:05:29
he happens to live in canada has a do i
2:05:29
he happens to live in canada has a do i although he's originally from england
2:05:30
although he's originally from england
2:05:30
although he's originally from england uh he got his phd in ai in 1977
2:05:34
uh he got his phd in ai in 1977
2:05:34
uh he got his phd in ai in 1977 not a great time to be an ai researcher
2:05:36
not a great time to be an ai researcher
2:05:36
not a great time to be an ai researcher it was very challenging time
2:05:38
it was very challenging time
2:05:38
it was very challenging time but he was one of the folks deeply
2:05:40
but he was one of the folks deeply
2:05:40
but he was one of the folks deeply entered
2:05:41
entered
2:05:41
entered in the neural net model that the
2:05:44
in the neural net model that the
2:05:44
in the neural net model that the scruffies loved
2:05:45
scruffies loved
2:05:45
scruffies loved so much and by the 80s
2:05:48
so much and by the 80s
2:05:48
so much and by the 80s in that time of the list systems and so
2:05:50
in that time of the list systems and so
2:05:50
in that time of the list systems and so forth starting to end
2:05:52
forth starting to end
2:05:52
forth starting to end he was talking about the fact that
2:05:54
he was talking about the fact that
2:05:54
he was talking about the fact that neural nets were in the 80s were just
2:05:56
neural nets were in the 80s were just
2:05:56
neural nets were in the 80s were just not sophisticated enough to do the kind
2:05:57
not sophisticated enough to do the kind
2:05:57
not sophisticated enough to do the kind of work that needed that he
2:05:59
of work that needed that he
2:05:59
of work that needed that he come up with a mathematical construct
2:06:00
come up with a mathematical construct
2:06:00
come up with a mathematical construct called back propagation
2:06:02
called back propagation
2:06:02
called back propagation that open the possibility for much
2:06:04
that open the possibility for much
2:06:04
that open the possibility for much deeper neural nets but sort of ends that
2:06:06
deeper neural nets but sort of ends that
2:06:06
deeper neural nets but sort of ends that paper with
2:06:07
paper with
2:06:08
paper with these computers aren't going to do it
2:06:09
these computers aren't going to do it
2:06:09
these computers aren't going to do it they're not strong enough and so
2:06:12
they're not strong enough and so
2:06:12
they're not strong enough and so that quiet period
2:06:15
that quiet period
2:06:15
that quiet period for neural nets and that's in that
2:06:17
for neural nets and that's in that
2:06:17
for neural nets and that's in that second ai winter these deep neural nets
2:06:20
second ai winter these deep neural nets
2:06:20
second ai winter these deep neural nets they don't go anywhere for a few years
2:06:22
they don't go anywhere for a few years
2:06:22
they don't go anywhere for a few years it's actually students of jeffrey
2:06:24
it's actually students of jeffrey
2:06:24
it's actually students of jeffrey hintons who sort of dust off that old
2:06:25
hintons who sort of dust off that old
2:06:25
hintons who sort of dust off that old paper from the 80s
2:06:27
paper from the 80s
2:06:27
paper from the 80s and start working on a new problem uh
2:06:29
and start working on a new problem uh
2:06:29
and start working on a new problem uh thinking the compute is now
2:06:30
thinking the compute is now
2:06:30
thinking the compute is now there so we've got a new generation of
2:06:32
there so we've got a new generation of
2:06:32
there so we've got a new generation of computational power
2:06:33
computational power
2:06:34
computational power and so by the in the 2000s
2:06:37
and so by the in the 2000s
2:06:37
and so by the in the 2000s there was a new competition so this is
2:06:39
there was a new competition so this is
2:06:39
there was a new competition so this is the imagenet large scale visual
2:06:41
the imagenet large scale visual
2:06:41
the imagenet large scale visual recognition challenge
2:06:42
recognition challenge
2:06:42
recognition challenge started in 2010 uh taking advantage of
2:06:46
started in 2010 uh taking advantage of
2:06:46
started in 2010 uh taking advantage of the cloud and available or the
2:06:48
the cloud and available or the
2:06:48
the cloud and available or the the internet and availability of data it
2:06:50
the internet and availability of data it
2:06:50
the internet and availability of data it is a collection of 14 million
2:06:53
is a collection of 14 million
2:06:53
is a collection of 14 million hand annotated images so it's a set of
2:06:55
hand annotated images so it's a set of
2:06:56
hand annotated images so it's a set of images
2:06:56
images
2:06:56
images that are i already identified tools for
2:06:59
that are i already identified tools for
2:06:59
that are i already identified tools for training with
2:07:00
training with
2:07:00
training with and the competition was to build
2:07:02
and the competition was to build
2:07:02
and the competition was to build software that could recognize those
2:07:04
software that could recognize those
2:07:04
software that could recognize those images
2:07:05
images
2:07:05
images reliably and in the first couple of
2:07:06
reliably and in the first couple of
2:07:06
reliably and in the first couple of years the competition 2010 2011
2:07:09
years the competition 2010 2011
2:07:09
years the competition 2010 2011 virtual software were decision tree
2:07:11
virtual software were decision tree
2:07:11
virtual software were decision tree solutions and the best they ever did
2:07:13
solutions and the best they ever did
2:07:13
solutions and the best they ever did was 74 recognition accuracy not bad
2:07:17
was 74 recognition accuracy not bad
2:07:17
was 74 recognition accuracy not bad humans do better not great but
2:07:20
humans do better not great but
2:07:20
humans do better not great but hinton's group in 2012 implemented a
2:07:23
hinton's group in 2012 implemented a
2:07:24
hinton's group in 2012 implemented a deep convoluted neural network
2:07:26
deep convoluted neural network
2:07:26
deep convoluted neural network and they ran in the competition on
2:07:28
and they ran in the competition on
2:07:28
and they ran in the competition on september 30th
2:07:29
september 30th
2:07:29
september 30th 2012 and their first attempt
2:07:32
2012 and their first attempt
2:07:32
2012 and their first attempt in during the contest got an 84 result
2:07:36
in during the contest got an 84 result
2:07:36
in during the contest got an 84 result by 2015 it was a hundred percent and so
2:07:40
by 2015 it was a hundred percent and so
2:07:40
by 2015 it was a hundred percent and so in that sense the scruffy's had their
2:07:42
in that sense the scruffy's had their
2:07:42
in that sense the scruffy's had their day
2:07:42
day
2:07:42
day the best the decision trees could do on
2:07:45
the best the decision trees could do on
2:07:45
the best the decision trees could do on this
2:07:45
this
2:07:45
this was quickly superseded by this modern
2:07:48
was quickly superseded by this modern
2:07:48
was quickly superseded by this modern neural net model
2:07:49
neural net model
2:07:49
neural net model and uh it opened the door to a huge
2:07:52
and uh it opened the door to a huge
2:07:52
and uh it opened the door to a huge explosion of technology
2:07:54
explosion of technology
2:07:54
explosion of technology when i think about artificial
2:07:56
when i think about artificial
2:07:56
when i think about artificial intelligence today in this
2:07:57
intelligence today in this
2:07:58
intelligence today in this new spring that we've been living in
2:08:01
new spring that we've been living in
2:08:01
new spring that we've been living in it starts here with this image
2:08:03
it starts here with this image
2:08:04
it starts here with this image recognition now quickly moved over to
2:08:05
recognition now quickly moved over to
2:08:05
recognition now quickly moved over to voice recognition because
2:08:07
voice recognition because
2:08:07
voice recognition because you know one of the beautiful things
2:08:09
you know one of the beautiful things
2:08:09
you know one of the beautiful things about the academic model
2:08:11
about the academic model
2:08:11
about the academic model is the sharing of the papers the sharing
2:08:13
is the sharing of the papers the sharing
2:08:13
is the sharing of the papers the sharing of the core concepts
2:08:15
of the core concepts
2:08:15
of the core concepts and so very quickly other organizations
2:08:17
and so very quickly other organizations
2:08:17
and so very quickly other organizations pick this up
2:08:18
pick this up
2:08:18
pick this up and so looking at voice recognition the
2:08:21
and so looking at voice recognition the
2:08:21
and so looking at voice recognition the emergence
2:08:21
emergence
2:08:22
emergence of siri roughly in that same time so sri
2:08:25
of siri roughly in that same time so sri
2:08:25
of siri roughly in that same time so sri was the company largely an offshoot of a
2:08:27
was the company largely an offshoot of a
2:08:27
was the company largely an offshoot of a darpa project that was acquired by apple
2:08:30
darpa project that was acquired by apple
2:08:30
darpa project that was acquired by apple using the convoluted neural net to do
2:08:32
using the convoluted neural net to do
2:08:32
using the convoluted neural net to do voice recognition and voice recognition
2:08:34
voice recognition and voice recognition
2:08:34
voice recognition and voice recognition had existed four years before this
2:08:37
had existed four years before this
2:08:37
had existed four years before this but it was much more that decision tree
2:08:39
but it was much more that decision tree
2:08:39
but it was much more that decision tree model and it required a lot of training
2:08:40
model and it required a lot of training
2:08:40
model and it required a lot of training it was much more difficult to make voice
2:08:42
it was much more difficult to make voice
2:08:42
it was much more difficult to make voice recognition work
2:08:43
recognition work
2:08:43
recognition work and then along came siri and it just
2:08:45
and then along came siri and it just
2:08:45
and then along came siri and it just worked
2:08:47
worked
2:08:47
worked unless you were australian because some
2:08:49
unless you were australian because some
2:08:50
unless you were australian because some accidents are harder than other accents
2:08:51
accidents are harder than other accents
2:08:51
accidents are harder than other accents but that that siri google voice even
2:08:55
but that that siri google voice even
2:08:55
but that that siri google voice even cortana
2:08:56
cortana
2:08:56
cortana all roughly along the same lines and
2:08:58
all roughly along the same lines and
2:08:58
all roughly along the same lines and what had happened was voice recognition
2:09:00
what had happened was voice recognition
2:09:00
what had happened was voice recognition and the old models that were 93 94
2:09:03
and the old models that were 93 94
2:09:03
and the old models that were 93 94 accurate
2:09:04
accurate
2:09:04
accurate suddenly jumped up to 99 percent
2:09:05
suddenly jumped up to 99 percent
2:09:05
suddenly jumped up to 99 percent accurate by switching over to that
2:09:07
accurate by switching over to that
2:09:07
accurate by switching over to that neural network
2:09:09
neural network
2:09:09
neural network there's an interesting pattern that's
2:09:11
there's an interesting pattern that's
2:09:11
there's an interesting pattern that's happening here over and over and over
2:09:12
happening here over and over and over
2:09:12
happening here over and over and over again
2:09:13
again
2:09:13
again in the evolution of artificial
2:09:14
in the evolution of artificial
2:09:14
in the evolution of artificial intelligence and i've got this
2:09:16
intelligence and i've got this
2:09:16
intelligence and i've got this diagram i found that i found that sort
2:09:18
diagram i found that i found that sort
2:09:18
diagram i found that i found that sort of breaks it all down
2:09:20
of breaks it all down
2:09:20
of breaks it all down artificial intelligence seems to be the
2:09:22
artificial intelligence seems to be the
2:09:22
artificial intelligence seems to be the name of a technology that when it
2:09:24
name of a technology that when it
2:09:24
name of a technology that when it is in its immature state or rather when
2:09:27
is in its immature state or rather when
2:09:27
is in its immature state or rather when it doesn't work
2:09:28
it doesn't work
2:09:28
it doesn't work the moment it does work it gets a new
2:09:30
the moment it does work it gets a new
2:09:30
the moment it does work it gets a new name
2:09:31
name
2:09:31
name so if you look carefully at this
2:09:33
so if you look carefully at this
2:09:33
so if you look carefully at this visualization you notice sort of in the
2:09:34
visualization you notice sort of in the
2:09:34
visualization you notice sort of in the middle there the planning scheduling
2:09:36
middle there the planning scheduling
2:09:36
middle there the planning scheduling optimization
2:09:36
optimization
2:09:36
optimization the technologies that were worked on in
2:09:38
the technologies that were worked on in
2:09:38
the technologies that were worked on in the 50s and the 60s by minsky and their
2:09:40
the 50s and the 60s by minsky and their
2:09:40
the 50s and the 60s by minsky and their folks
2:09:40
folks
2:09:40
folks so the shortest line there and then the
2:09:42
so the shortest line there and then the
2:09:42
so the shortest line there and then the expert systems from the 70 and i haven't
2:09:44
expert systems from the 70 and i haven't
2:09:44
expert systems from the 70 and i haven't really talked about robotics
2:09:46
really talked about robotics
2:09:46
really talked about robotics but that's also from the 70s when robots
2:09:48
but that's also from the 70s when robots
2:09:48
but that's also from the 70s when robots really appear in manufacturing
2:09:50
really appear in manufacturing
2:09:50
really appear in manufacturing you know there's a real robots existed
2:09:52
you know there's a real robots existed
2:09:52
you know there's a real robots existed before the concept of artificial
2:09:54
before the concept of artificial
2:09:54
before the concept of artificial intelligence
2:09:55
intelligence
2:09:55
intelligence there are some integrations there but
2:09:56
there are some integrations there but
2:09:56
there are some integrations there but they're kind of their own thing
2:09:58
they're kind of their own thing
2:09:58
they're kind of their own thing speech systems have been around a very
2:10:00
speech systems have been around a very
2:10:00
speech systems have been around a very long time but they were transformed
2:10:02
long time but they were transformed
2:10:02
long time but they were transformed by the convoluted net and now we have
2:10:05
by the convoluted net and now we have
2:10:05
by the convoluted net and now we have these
2:10:06
these
2:10:06
these as these technologies are spun out we
2:10:07
as these technologies are spun out we
2:10:08
as these technologies are spun out we give them new names our image
2:10:09
give them new names our image
2:10:09
give them new names our image recognition machine vision systems
2:10:11
recognition machine vision systems
2:10:11
recognition machine vision systems our new ability to do natural language
2:10:13
our new ability to do natural language
2:10:13
our new ability to do natural language processing even to the point of
2:10:14
processing even to the point of
2:10:14
processing even to the point of translation
2:10:15
translation
2:10:15
translation and doing classification sentiment
2:10:17
and doing classification sentiment
2:10:17
and doing classification sentiment analysis those sorts of things
2:10:18
analysis those sorts of things
2:10:18
analysis those sorts of things and ultimately in our modern version of
2:10:21
and ultimately in our modern version of
2:10:21
and ultimately in our modern version of machine learning the deep learning
2:10:22
machine learning the deep learning
2:10:22
machine learning the deep learning predictive analytics models
2:10:24
predictive analytics models
2:10:24
predictive analytics models and it's the hardware that's made the
2:10:25
and it's the hardware that's made the
2:10:26
and it's the hardware that's made the difference here but not just the
2:10:27
difference here but not just the
2:10:27
difference here but not just the hardware
2:10:27
hardware
2:10:27
hardware how we utilize it the importance of the
2:10:30
how we utilize it the importance of the
2:10:30
how we utilize it the importance of the cloud
2:10:31
cloud
2:10:31
cloud and it's in expanding the opportunities
2:10:33
and it's in expanding the opportunities
2:10:33
and it's in expanding the opportunities for artificial intelligence
2:10:35
for artificial intelligence
2:10:35
for artificial intelligence because often our training models
2:10:36
because often our training models
2:10:36
because often our training models require huge amounts of compute
2:10:38
require huge amounts of compute
2:10:38
require huge amounts of compute for a relatively short amount of time
2:10:40
for a relatively short amount of time
2:10:40
for a relatively short amount of time and so
2:10:41
and so
2:10:41
and so we would still be confined to a few
2:10:44
we would still be confined to a few
2:10:44
we would still be confined to a few small areas in the world with that kind
2:10:46
small areas in the world with that kind
2:10:46
small areas in the world with that kind of compute available
2:10:47
of compute available
2:10:47
of compute available and sort of restricted access to it that
2:10:49
and sort of restricted access to it that
2:10:49
and sort of restricted access to it that would limit the expansion of artificial
2:10:51
would limit the expansion of artificial
2:10:52
would limit the expansion of artificial intelligence
2:10:52
intelligence
2:10:52
intelligence the cloud leveled the field that anyone
2:10:56
the cloud leveled the field that anyone
2:10:56
the cloud leveled the field that anyone could build a model and rent that
2:10:58
could build a model and rent that
2:10:58
could build a model and rent that computing power for the short amount of
2:11:00
computing power for the short amount of
2:11:00
computing power for the short amount of time they need it for a reasonable price
2:11:02
time they need it for a reasonable price
2:11:02
time they need it for a reasonable price and then execute on their models and
2:11:04
and then execute on their models and
2:11:04
and then execute on their models and there's only been a few occasions where
2:11:05
there's only been a few occasions where
2:11:05
there's only been a few occasions where we've seen
2:11:06
we've seen
2:11:06
we've seen major artificial intelligence events
2:11:09
major artificial intelligence events
2:11:09
major artificial intelligence events occurring
2:11:10
occurring
2:11:10
occurring that got us to a point where even the
2:11:12
that got us to a point where even the
2:11:12
that got us to a point where even the cloud was training but computers
2:11:13
cloud was training but computers
2:11:13
cloud was training but computers continue to expand the evolution of the
2:11:15
continue to expand the evolution of the
2:11:15
continue to expand the evolution of the gpu and scalar computing
2:11:17
gpu and scalar computing
2:11:17
gpu and scalar computing optimized machines for these kinds of
2:11:20
optimized machines for these kinds of
2:11:20
optimized machines for these kinds of neural net models
2:11:21
neural net models
2:11:21
neural net models brings us closer and closer to the
2:11:23
brings us closer and closer to the
2:11:23
brings us closer and closer to the diversity of being able to do this in
2:11:25
diversity of being able to do this in
2:11:25
diversity of being able to do this in almost any location at any time
2:11:27
almost any location at any time
2:11:27
almost any location at any time i also think the availability of data
2:11:29
i also think the availability of data
2:11:29
i also think the availability of data has a huge impact and for better or
2:11:30
has a huge impact and for better or
2:11:30
has a huge impact and for better or worse we have to point to social media
2:11:32
worse we have to point to social media
2:11:32
worse we have to point to social media as the impacts around that the tendency
2:11:35
as the impacts around that the tendency
2:11:35
as the impacts around that the tendency for more people to put more data online
2:11:37
for more people to put more data online
2:11:37
for more people to put more data online where it's readily available has
2:11:40
where it's readily available has
2:11:40
where it's readily available has provided mechanisms for us to train more
2:11:42
provided mechanisms for us to train more
2:11:42
provided mechanisms for us to train more models in more ways
2:11:45
models in more ways
2:11:45
models in more ways all of these things have worked together
2:11:47
all of these things have worked together
2:11:47
all of these things have worked together to get us now to
2:11:48
to get us now to
2:11:48
to get us now to just a few years ago i started to talk
2:11:51
just a few years ago i started to talk
2:11:51
just a few years ago i started to talk about this next
2:11:51
about this next
2:11:52
about this next generation so 2011 2012 2013
2:11:55
generation so 2011 2012 2013
2:11:55
generation so 2011 2012 2013 very ascendant models of the of those
2:11:57
very ascendant models of the of those
2:11:57
very ascendant models of the of those original concepts of hinton and i don't
2:11:59
original concepts of hinton and i don't
2:11:59
original concepts of hinton and i don't want to make
2:12:00
want to make
2:12:00
want to make put us all onto jeffrey hinden i think
2:12:01
put us all onto jeffrey hinden i think
2:12:01
put us all onto jeffrey hinden i think he's an extraordinary man but there are
2:12:03
he's an extraordinary man but there are
2:12:03
he's an extraordinary man but there are a
2:12:03
a
2:12:03
a number of extraordinary people in this
2:12:05
number of extraordinary people in this
2:12:05
number of extraordinary people in this field that has sent it all at the same
2:12:07
field that has sent it all at the same
2:12:07
field that has sent it all at the same time
2:12:08
time
2:12:08
time i do want to talk about go because i
2:12:10
i do want to talk about go because i
2:12:10
i do want to talk about go because i think it's got a great corollary to the
2:12:12
think it's got a great corollary to the
2:12:12
think it's got a great corollary to the success in the 1970s by ibm with deep
2:12:15
success in the 1970s by ibm with deep
2:12:15
success in the 1970s by ibm with deep blue
2:12:16
blue
2:12:16
blue go was considered an impossible game to
2:12:18
go was considered an impossible game to
2:12:18
go was considered an impossible game to model and it is certainly in the
2:12:19
model and it is certainly in the
2:12:19
model and it is certainly in the decision tree model
2:12:21
decision tree model
2:12:21
decision tree model the 19 by 19 grid games uh
2:12:24
the 19 by 19 grid games uh
2:12:24
the 19 by 19 grid games uh are deeply cerebral very complex games
2:12:28
are deeply cerebral very complex games
2:12:28
are deeply cerebral very complex games you can't brute force your way through a
2:12:30
you can't brute force your way through a
2:12:30
you can't brute force your way through a computational model on go
2:12:32
computational model on go
2:12:32
computational model on go it's too many possibilities uh the
2:12:35
it's too many possibilities uh the
2:12:35
it's too many possibilities uh the google folks took it on with alphago in
2:12:37
google folks took it on with alphago in
2:12:37
google folks took it on with alphago in the neural net model they spent two
2:12:39
the neural net model they spent two
2:12:39
the neural net model they spent two years training that model
2:12:41
years training that model
2:12:41
years training that model teaching it games and so by 2016 it
2:12:44
teaching it games and so by 2016 it
2:12:44
teaching it games and so by 2016 it beats a world champion class and of
2:12:46
beats a world champion class and of
2:12:46
beats a world champion class and of course the
2:12:47
course the
2:12:47
course the famous event is in may of 2017
2:12:50
famous event is in may of 2017
2:12:50
famous event is in may of 2017 when kg is beat three to nothing
2:12:53
when kg is beat three to nothing
2:12:53
when kg is beat three to nothing by alphaco uh
2:12:57
by alphaco uh
2:12:57
by alphaco uh another ascendant moment but they didn't
2:12:59
another ascendant moment but they didn't
2:12:59
another ascendant moment but they didn't stop there i think what's more
2:13:00
stop there i think what's more
2:13:00
stop there i think what's more interesting still is alpha zero
2:13:03
interesting still is alpha zero
2:13:03
interesting still is alpha zero that they took the the approach they did
2:13:05
that they took the the approach they did
2:13:05
that they took the the approach they did with alphago which took a couple of
2:13:07
with alphago which took a couple of
2:13:07
with alphago which took a couple of years to build with sort of very
2:13:08
years to build with sort of very
2:13:08
years to build with sort of very traditional training model
2:13:10
traditional training model
2:13:10
traditional training model and here with alpha zero they used a
2:13:12
and here with alpha zero they used a
2:13:12
and here with alpha zero they used a much more sophisticated approach in
2:13:14
much more sophisticated approach in
2:13:14
much more sophisticated approach in neural nets the adversarial approach
2:13:16
neural nets the adversarial approach
2:13:16
neural nets the adversarial approach where it actually played itself for 40
2:13:17
where it actually played itself for 40
2:13:17
where it actually played itself for 40 days knowing only the rules of the game
2:13:19
days knowing only the rules of the game
2:13:19
days knowing only the rules of the game rather than having looked at many many
2:13:21
rather than having looked at many many
2:13:21
rather than having looked at many many games
2:13:21
games
2:13:21
games the way you would typically train a
2:13:23
the way you would typically train a
2:13:23
the way you would typically train a model and so that in that short amount
2:13:25
model and so that in that short amount
2:13:25
model and so that in that short amount of time
2:13:26
of time
2:13:26
of time it was able to beat alphago against
2:13:29
it was able to beat alphago against
2:13:30
it was able to beat alphago against uh play and go they also did a
2:13:32
uh play and go they also did a
2:13:32
uh play and go they also did a demonstration and this was
2:13:33
demonstration and this was
2:13:33
demonstration and this was uh just a couple of years ago where they
2:13:35
uh just a couple of years ago where they
2:13:35
uh just a couple of years ago where they put it up against the chess program
2:13:37
put it up against the chess program
2:13:37
put it up against the chess program stockfish
2:13:38
stockfish
2:13:38
stockfish and uh not only did they show that it
2:13:40
and uh not only did they show that it
2:13:40
and uh not only did they show that it could beat stock fish in a relatively
2:13:42
could beat stock fish in a relatively
2:13:42
could beat stock fish in a relatively short amount of time
2:13:43
short amount of time
2:13:43
short amount of time but the computing resources required
2:13:46
but the computing resources required
2:13:46
but the computing resources required to run the model to be beat the best
2:13:49
to run the model to be beat the best
2:13:49
to run the model to be beat the best digital chess players was hugely lower
2:13:53
digital chess players was hugely lower
2:13:53
digital chess players was hugely lower that it was a much more efficient way
2:13:55
that it was a much more efficient way
2:13:56
that it was a much more efficient way when you have a good model
2:13:57
when you have a good model
2:13:57
when you have a good model to actually execute these kinds of
2:13:58
to actually execute these kinds of
2:13:58
to actually execute these kinds of complex problems now there's still some
2:14:00
complex problems now there's still some
2:14:00
complex problems now there's still some conversations around what this all looks
2:14:02
conversations around what this all looks
2:14:02
conversations around what this all looks like
2:14:02
like
2:14:02
like but it speaks to this ongoing evolution
2:14:06
but it speaks to this ongoing evolution
2:14:06
but it speaks to this ongoing evolution of ai
2:14:07
of ai
2:14:07
of ai in more and more sophisticated models
2:14:09
in more and more sophisticated models
2:14:09
in more and more sophisticated models now
2:14:10
now
2:14:10
now i was going to steer largely clear of
2:14:11
i was going to steer largely clear of
2:14:11
i was going to steer largely clear of robotics but really
2:14:13
robotics but really
2:14:13
robotics but really getting down into sort of like closer
2:14:15
getting down into sort of like closer
2:14:15
getting down into sort of like closer and closer to the current times
2:14:17
and closer to the current times
2:14:17
and closer to the current times i want to talk about mit dactyl
2:14:20
i want to talk about mit dactyl
2:14:20
i want to talk about mit dactyl so this is a robotic hand for
2:14:22
so this is a robotic hand for
2:14:22
so this is a robotic hand for manipulating
2:14:24
manipulating
2:14:24
manipulating objects and we know that training models
2:14:27
objects and we know that training models
2:14:27
objects and we know that training models takes time you have to do it over and
2:14:28
takes time you have to do it over and
2:14:28
takes time you have to do it over and over and over again and so as soon as
2:14:29
over and over again and so as soon as
2:14:30
over and over again and so as soon as you start talking about training
2:14:31
you start talking about training
2:14:31
you start talking about training articulation like this
2:14:33
articulation like this
2:14:33
articulation like this it can take impossibly long what the mit
2:14:36
it can take impossibly long what the mit
2:14:36
it can take impossibly long what the mit group succeeded to do with dactyl was to
2:14:38
group succeeded to do with dactyl was to
2:14:38
group succeeded to do with dactyl was to simulate the hand
2:14:40
simulate the hand
2:14:40
simulate the hand in a virtual reality model and so to
2:14:42
in a virtual reality model and so to
2:14:42
in a virtual reality model and so to allow it to
2:14:44
allow it to
2:14:44
allow it to practice the manipulation of an object
2:14:46
practice the manipulation of an object
2:14:46
practice the manipulation of an object in a digital realm
2:14:47
in a digital realm
2:14:47
in a digital realm at much higher speeds the equivalent of
2:14:50
at much higher speeds the equivalent of
2:14:50
at much higher speeds the equivalent of hundreds of years of practice
2:14:52
hundreds of years of practice
2:14:52
hundreds of years of practice and then to take that model implement it
2:14:54
and then to take that model implement it
2:14:54
and then to take that model implement it physically
2:14:55
physically
2:14:55
physically and have it have the same articulations
2:14:58
and have it have the same articulations
2:14:58
and have it have the same articulations so this is a i think a transitionary
2:15:00
so this is a i think a transitionary
2:15:00
so this is a i think a transitionary model we're starting to get into
2:15:02
model we're starting to get into
2:15:02
model we're starting to get into where the fact that we can create
2:15:03
where the fact that we can create
2:15:03
where the fact that we can create physical representations of those
2:15:05
physical representations of those
2:15:05
physical representations of those learnings not just identifying images
2:15:07
learnings not just identifying images
2:15:07
learnings not just identifying images with probabilities
2:15:08
with probabilities
2:15:08
with probabilities or simulating voice or recognizing voice
2:15:10
or simulating voice or recognizing voice
2:15:10
or simulating voice or recognizing voice or any of these far more intangibles
2:15:12
or any of these far more intangibles
2:15:12
or any of these far more intangibles but to actually have physical activities
2:15:14
but to actually have physical activities
2:15:14
but to actually have physical activities now be manifest
2:15:15
now be manifest
2:15:15
now be manifest with that same kind of training in a
2:15:17
with that same kind of training in a
2:15:17
with that same kind of training in a virtual realm
2:15:19
virtual realm
2:15:19
virtual realm that to me is barely history this is a
2:15:23
that to me is barely history this is a
2:15:23
that to me is barely history this is a year or so ago
2:15:24
year or so ago
2:15:24
year or so ago that we're getting to this level of
2:15:26
that we're getting to this level of
2:15:26
that we're getting to this level of experimentation and it opens the door to
2:15:28
experimentation and it opens the door to
2:15:28
experimentation and it opens the door to new potential
2:15:30
new potential
2:15:30
new potential but it opens the question which is are
2:15:32
but it opens the question which is are
2:15:32
but it opens the question which is are we continuing to evolve these are mostly
2:15:34
we continuing to evolve these are mostly
2:15:34
we continuing to evolve these are mostly engineering exercises fundamentally we
2:15:36
engineering exercises fundamentally we
2:15:36
engineering exercises fundamentally we look at gains in the
2:15:37
look at gains in the
2:15:37
look at gains in the in the modern neural net we're
2:15:40
in the modern neural net we're
2:15:40
in the modern neural net we're engineering with those now we're doing
2:15:42
engineering with those now we're doing
2:15:42
engineering with those now we're doing bigger and bigger projects
2:15:43
bigger and bigger projects
2:15:43
bigger and bigger projects organizations like microsoft and google
2:15:45
organizations like microsoft and google
2:15:45
organizations like microsoft and google and so forth providing us tools
2:15:47
and so forth providing us tools
2:15:47
and so forth providing us tools so that we as engineers can take these
2:15:50
so that we as engineers can take these
2:15:50
so that we as engineers can take these technologies and implement to provide
2:15:51
technologies and implement to provide
2:15:51
technologies and implement to provide benefits of companies
2:15:53
benefits of companies
2:15:53
benefits of companies but there's a conversation of are we
2:15:55
but there's a conversation of are we
2:15:55
but there's a conversation of are we expecting another ai winter
2:15:57
expecting another ai winter
2:15:57
expecting another ai winter and folks like jeff hinton have talked
2:15:59
and folks like jeff hinton have talked
2:15:59
and folks like jeff hinton have talked about the fact that there's another ai
2:16:00
about the fact that there's another ai
2:16:00
about the fact that there's another ai winter coming but what they're really
2:16:01
winter coming but what they're really
2:16:01
winter coming but what they're really talking about is that
2:16:03
talking about is that
2:16:03
talking about is that the money being spent on research is
2:16:05
the money being spent on research is
2:16:05
the money being spent on research is starting to
2:16:07
starting to
2:16:07
starting to diminish now i would point out two
2:16:11
diminish now i would point out two
2:16:11
diminish now i would point out two things one is
2:16:12
things one is
2:16:12
things one is in all the winters that i've talked
2:16:13
in all the winters that i've talked
2:16:13
in all the winters that i've talked about it's not like the work disappeared
2:16:15
about it's not like the work disappeared
2:16:15
about it's not like the work disappeared and it's not like work didn't continue
2:16:18
and it's not like work didn't continue
2:16:18
and it's not like work didn't continue the rates changed i think we've hit a
2:16:20
the rates changed i think we've hit a
2:16:20
the rates changed i think we've hit a period
2:16:21
period
2:16:21
period now where we're getting such good
2:16:23
now where we're getting such good
2:16:23
now where we're getting such good results with the technology have that
2:16:25
results with the technology have that
2:16:25
results with the technology have that there's
2:16:25
there's
2:16:25
there's the energy to do research may be
2:16:26
the energy to do research may be
2:16:26
the energy to do research may be diminishing somewhat so that perhaps
2:16:28
diminishing somewhat so that perhaps
2:16:28
diminishing somewhat so that perhaps there's fewer new innovations
2:16:30
there's fewer new innovations
2:16:30
there's fewer new innovations and notice really not a whole lot of
2:16:32
and notice really not a whole lot of
2:16:32
and notice really not a whole lot of conversation about building an
2:16:34
conversation about building an
2:16:34
conversation about building an intelligence either we're building we're
2:16:36
intelligence either we're building we're
2:16:36
intelligence either we're building we're still using these technologies to build
2:16:39
still using these technologies to build
2:16:39
still using these technologies to build tools
2:16:39
tools
2:16:39
tools to do useful things for us but not
2:16:42
to do useful things for us but not
2:16:42
to do useful things for us but not necessarily to be
2:16:43
necessarily to be
2:16:43
necessarily to be that tremendously intelligent i would
2:16:45
that tremendously intelligent i would
2:16:45
that tremendously intelligent i would argue that we are still
2:16:47
argue that we are still
2:16:47
argue that we are still in a springtime or in a summer
2:16:50
in a springtime or in a summer
2:16:50
in a springtime or in a summer where we are providing value to
2:16:52
where we are providing value to
2:16:52
where we are providing value to organizations
2:16:53
organizations
2:16:54
organizations using a technology and as long as we're
2:16:57
using a technology and as long as we're
2:16:57
using a technology and as long as we're doing that as long as we're taking that
2:16:58
doing that as long as we're taking that
2:16:58
doing that as long as we're taking that technology
2:16:59
technology
2:16:59
technology creating unique value from it our work
2:17:01
creating unique value from it our work
2:17:01
creating unique value from it our work is going to continue
2:17:03
is going to continue
2:17:03
is going to continue the rate of innovation waxes and wanes
2:17:05
the rate of innovation waxes and wanes
2:17:05
the rate of innovation waxes and wanes with need
2:17:06
with need
2:17:06
with need i would argue the pandemics put a lot of
2:17:08
i would argue the pandemics put a lot of
2:17:08
i would argue the pandemics put a lot of pressure on us
2:17:09
pressure on us
2:17:10
pressure on us as experimenters to show value rapidly
2:17:13
as experimenters to show value rapidly
2:17:13
as experimenters to show value rapidly and that
2:17:13
and that
2:17:14
and that tends to diminish broader research
2:17:17
tends to diminish broader research
2:17:17
tends to diminish broader research but that's okay this is a special time
2:17:20
but that's okay this is a special time
2:17:20
but that's okay this is a special time providing value is important it's good
2:17:22
providing value is important it's good
2:17:22
providing value is important it's good for us to have a retrenchment
2:17:23
for us to have a retrenchment
2:17:23
for us to have a retrenchment think about the value and at the same
2:17:26
think about the value and at the same
2:17:26
think about the value and at the same time when that value starts to level off
2:17:28
time when that value starts to level off
2:17:28
time when that value starts to level off we can do some more exploration thanks
2:17:30
we can do some more exploration thanks
2:17:30
we can do some more exploration thanks so much for your time on this
2:17:32
so much for your time on this
2:17:32
so much for your time on this i've really enjoyed talking about it's
2:17:33
i've really enjoyed talking about it's
2:17:34
i've really enjoyed talking about it's one of my favorite subjects
2:17:35
one of my favorite subjects
2:17:35
one of my favorite subjects and i'm hoping to stick around for the
2:17:36
and i'm hoping to stick around for the
2:17:36
and i'm hoping to stick around for the next half hour or so and be part of the
2:17:38
next half hour or so and be part of the
2:17:38
next half hour or so and be part of the panel coming up
2:17:42
panel coming up
2:17:42
panel coming up that was awesome i'm glad i'm one of the
2:17:45
that was awesome i'm glad i'm one of the
2:17:46
that was awesome i'm glad i'm one of the needs
2:17:47
needs
2:17:47
needs and the schools set well you know folks
2:17:50
and the schools set well you know folks
2:17:50
and the schools set well you know folks i don't know the folks know but i
2:17:51
i don't know the folks know but i
2:17:51
i don't know the folks know but i certainly know you have a degree in
2:17:52
certainly know you have a degree in
2:17:52
certainly know you have a degree in mathematics
2:17:53
mathematics
2:17:54
mathematics all right i did get a master's degree in
2:17:55
all right i did get a master's degree in
2:17:56
all right i did get a master's degree in computer science and
2:17:57
computer science and
2:17:57
computer science and people don't know i was a doctoral
2:17:58
people don't know i was a doctoral
2:17:58
people don't know i was a doctoral student and i studied machine learning
2:18:02
student and i studied machine learning
2:18:02
student and i studied machine learning when the neets were kind of still in
2:18:04
when the neets were kind of still in
2:18:04
when the neets were kind of still in charge and my models were very
2:18:06
charge and my models were very
2:18:06
charge and my models were very mathematically beautiful
2:18:08
mathematically beautiful
2:18:08
mathematically beautiful and i have always struggled with neural
2:18:10
and i have always struggled with neural
2:18:10
and i have always struggled with neural networks because
2:18:12
networks because
2:18:12
networks because mathematically they're just not
2:18:16
mathematically they're just not
2:18:16
mathematically they're just not like anybody i know that has a strong
2:18:19
like anybody i know that has a strong
2:18:19
like anybody i know that has a strong mathematical background
2:18:20
mathematical background
2:18:20
mathematical background they loathe the lack of a proof they
2:18:23
they loathe the lack of a proof they
2:18:23
they loathe the lack of a proof they just don't believe it unless there's a
2:18:24
just don't believe it unless there's a
2:18:24
just don't believe it unless there's a proof
2:18:25
proof
2:18:25
proof and neural nets are remarkably resistant
2:18:27
and neural nets are remarkably resistant
2:18:27
and neural nets are remarkably resistant to proofs
2:18:29
to proofs
2:18:29
to proofs it's a problem and people don't know
2:18:31
it's a problem and people don't know
2:18:31
it's a problem and people don't know there is a there is an
2:18:32
there is a there is an
2:18:32
there is a there is an entire theory around all machine
2:18:36
entire theory around all machine
2:18:36
entire theory around all machine learning models
2:18:37
learning models
2:18:37
learning models except for the neural network like you
2:18:39
except for the neural network like you
2:18:39
except for the neural network like you can there's something called pac
2:18:41
can there's something called pac
2:18:41
can there's something called pac learning which stands for probably
2:18:42
learning which stands for probably
2:18:42
learning which stands for probably approximately correct where you can tell
2:18:46
approximately correct where you can tell
2:18:46
approximately correct where you can tell how many data points you need for a
2:18:48
how many data points you need for a
2:18:48
how many data points you need for a particular model
2:18:50
particular model
2:18:50
particular model to give you the the probability that
2:18:52
to give you the the probability that
2:18:52
to give you the the probability that it's that the accurate percentage is
2:18:54
it's that the accurate percentage is
2:18:54
it's that the accurate percentage is something right so
2:18:56
something right so
2:18:56
something right so and you can literally mathematically
2:18:58
and you can literally mathematically
2:18:58
and you can literally mathematically define that and there's something called
2:18:59
define that and there's something called
2:18:59
define that and there's something called dc dimension there's a there's an
2:19:01
dc dimension there's a there's an
2:19:01
dc dimension there's a there's an entire math and this is chernovanskis
2:19:04
entire math and this is chernovanskis
2:19:04
entire math and this is chernovanskis and vapnek
2:19:05
and vapnek
2:19:05
and vapnek came up with this stuff in the 90s and
2:19:07
came up with this stuff in the 90s and
2:19:07
came up with this stuff in the 90s and so when i studied machine learning it
2:19:09
so when i studied machine learning it
2:19:09
so when i studied machine learning it was like
2:19:10
was like
2:19:10
was like okay there are some pretty cool things
2:19:13
okay there are some pretty cool things
2:19:13
okay there are some pretty cool things like i love svms because they're
2:19:14
like i love svms because they're
2:19:14
like i love svms because they're mathematical beauty and i've talked to
2:19:16
mathematical beauty and i've talked to
2:19:16
mathematical beauty and i've talked to touring award winners about
2:19:18
touring award winners about
2:19:18
touring award winners about deep learning and their answers tend to
2:19:20
deep learning and their answers tend to
2:19:20
deep learning and their answers tend to be quite squishy
2:19:22
be quite squishy
2:19:22
be quite squishy yeah and it's not normal right you you
2:19:25
yeah and it's not normal right you you
2:19:25
yeah and it's not normal right you you think about
2:19:25
think about
2:19:25
think about in the arc of people that i was
2:19:27
in the arc of people that i was
2:19:27
in the arc of people that i was describing there guys like touring
2:19:29
describing there guys like touring
2:19:29
describing there guys like touring i didn't mention von neumann and so
2:19:30
i didn't mention von neumann and so
2:19:30
i didn't mention von neumann and so forth but long before the hardware
2:19:33
forth but long before the hardware
2:19:33
forth but long before the hardware could do the computations they had built
2:19:36
could do the computations they had built
2:19:36
could do the computations they had built the models for computing
2:19:37
the models for computing
2:19:37
the models for computing computation right the turing complete
2:19:40
computation right the turing complete
2:19:40
computation right the turing complete computational model
2:19:41
computational model
2:19:41
computational model was an important part of knowing you
2:19:43
was an important part of knowing you
2:19:43
was an important part of knowing you built a valid computer
2:19:45
built a valid computer
2:19:45
built a valid computer and so the fact that we're struggling
2:19:46
and so the fact that we're struggling
2:19:46
and so the fact that we're struggling with mathematical proofs around these
2:19:48
with mathematical proofs around these
2:19:48
with mathematical proofs around these neural net models
2:19:50
neural net models
2:19:50
neural net models it's deeply concerning for the validity
2:19:52
it's deeply concerning for the validity
2:19:52
it's deeply concerning for the validity of the process
2:19:54
of the process
2:19:54
of the process and it's very squishy like here's an
2:19:56
and it's very squishy like here's an
2:19:56
and it's very squishy like here's an example
2:19:57
example
2:19:57
example a simple mathematical example everyone
2:19:59
a simple mathematical example everyone
2:20:00
a simple mathematical example everyone has heard of a parabola
2:20:01
has heard of a parabola
2:20:01
has heard of a parabola it's a big u yeah if i were to ask you
2:20:04
it's a big u yeah if i were to ask you
2:20:04
it's a big u yeah if i were to ask you to mathematically give me
2:20:07
to mathematically give me
2:20:07
to mathematically give me the minimum value of that function it's
2:20:09
the minimum value of that function it's
2:20:09
the minimum value of that function it's easy to do
2:20:10
easy to do
2:20:10
easy to do yes of course set it equal to zero it
2:20:13
yes of course set it equal to zero it
2:20:13
yes of course set it equal to zero it tells you where it is of the u
2:20:15
tells you where it is of the u
2:20:15
tells you where it is of the u yeah and those are called convex
2:20:17
yeah and those are called convex
2:20:17
yeah and those are called convex functions because they basically
2:20:19
functions because they basically
2:20:19
functions because they basically have a single answer the thing about
2:20:21
have a single answer the thing about
2:20:21
have a single answer the thing about neural networks is that they are
2:20:23
neural networks is that they are
2:20:23
neural networks is that they are non-convex complex sheets in multi
2:20:26
non-convex complex sheets in multi
2:20:26
non-convex complex sheets in multi multiple dimensions and so there's
2:20:30
multiple dimensions and so there's
2:20:30
multiple dimensions and so there's no way to know number one are you in the
2:20:33
no way to know number one are you in the
2:20:33
no way to know number one are you in the right
2:20:33
right
2:20:33
right bottom you because you might just be in
2:20:35
bottom you because you might just be in
2:20:35
bottom you because you might just be in a hill or a little valley if there's a
2:20:37
a hill or a little valley if there's a
2:20:37
a hill or a little valley if there's a deeper valley and then the other
2:20:39
deeper valley and then the other
2:20:39
deeper valley and then the other question that i've always had is
2:20:41
question that i've always had is
2:20:41
question that i've always had is like how do you empirically it's only
2:20:44
like how do you empirically it's only
2:20:44
like how do you empirically it's only empirically but you have to empirically
2:20:46
empirically but you have to empirically
2:20:46
empirically but you have to empirically figure out how to structure these
2:20:49
figure out how to structure these
2:20:49
figure out how to structure these miles and i asked a touring award winner
2:20:51
miles and i asked a touring award winner
2:20:51
miles and i asked a touring award winner how do you how do you
2:20:53
how do you how do you
2:20:53
how do you how do you know and he's like well you kind of just
2:20:56
know and he's like well you kind of just
2:20:56
know and he's like well you kind of just do what
2:20:57
do what
2:20:57
do what people have done before and you go from
2:20:59
people have done before and you go from
2:20:59
people have done before and you go from there and i was like
2:21:02
there and i was like
2:21:02
there and i was like you're like a world renowned expert
2:21:05
you're like a world renowned expert
2:21:05
you're like a world renowned expert indeed the answer is close enough
2:21:09
indeed the answer is close enough
2:21:09
indeed the answer is close enough it's no the answer is like
2:21:13
well you know i mentioned arthur c
2:21:15
well you know i mentioned arthur c
2:21:15
well you know i mentioned arthur c clarke early on in 2001 space status
2:21:17
clarke early on in 2001 space status
2:21:17
clarke early on in 2001 space status he's also the
2:21:18
he's also the
2:21:18
he's also the fellow who said first any sufficiently
2:21:21
fellow who said first any sufficiently
2:21:21
fellow who said first any sufficiently advanced technology is indistinguishable
2:21:22
advanced technology is indistinguishable
2:21:22
advanced technology is indistinguishable from magic
2:21:24
from magic
2:21:24
from magic and uh and here we are there's now
2:21:27
and uh and here we are there's now
2:21:27
and uh and here we are there's now literally magic as a service
2:21:29
literally magic as a service
2:21:29
literally magic as a service it almost sounds like and that's what
2:21:31
it almost sounds like and that's what
2:21:31
it almost sounds like and that's what worries me because anthropomorphization
2:21:33
worries me because anthropomorphization
2:21:33
worries me because anthropomorphization of ai
2:21:34
of ai
2:21:34
of ai is probably the worst thing we can do
2:21:36
is probably the worst thing we can do
2:21:36
is probably the worst thing we can do for ethics well without a doubt and
2:21:38
for ethics well without a doubt and
2:21:38
for ethics well without a doubt and and we give this technology credibility
2:21:41
and we give this technology credibility
2:21:41
and we give this technology credibility that we don't have well the machine
2:21:42
that we don't have well the machine
2:21:42
that we don't have well the machine said so it must be true right we take
2:21:44
said so it must be true right we take
2:21:44
said so it must be true right we take the errors we have made in our past
2:21:46
the errors we have made in our past
2:21:46
the errors we have made in our past stored as data
2:21:48
stored as data
2:21:48
stored as data build models around it to amplify those
2:21:50
build models around it to amplify those
2:21:50
build models around it to amplify those errors
2:21:53
amen and like i was in a panel once and
2:21:56
amen and like i was in a panel once and
2:21:56
amen and like i was in a panel once and i think i got invited to the wrong panel
2:21:58
i think i got invited to the wrong panel
2:21:58
i think i got invited to the wrong panel it was like in texas
2:21:59
it was like in texas
2:21:59
it was like in texas for like educators right and like there
2:22:02
for like educators right and like there
2:22:02
for like educators right and like there was the
2:22:03
was the
2:22:03
was the there was like other company sales
2:22:05
there was like other company sales
2:22:05
there was like other company sales people and i was like
2:22:06
people and i was like
2:22:06
people and i was like i think i'm at the wrong meeting but
2:22:08
i think i'm at the wrong meeting but
2:22:08
i think i'm at the wrong meeting but they were talking about
2:22:09
they were talking about
2:22:10
they were talking about ai in such a way that i was like hey
2:22:13
ai in such a way that i was like hey
2:22:13
ai in such a way that i was like hey friends
2:22:13
friends
2:22:13
friends but i'm gonna stay sorry it's up here
2:22:17
but i'm gonna stay sorry it's up here
2:22:17
but i'm gonna stay sorry it's up here yeah humans you you draw a face
2:22:20
yeah humans you you draw a face
2:22:20
yeah humans you you draw a face on a rock with a sharpie and our
2:22:22
on a rock with a sharpie and our
2:22:22
on a rock with a sharpie and our immediate reaction is
2:22:24
immediate reaction is
2:22:24
immediate reaction is oh it's happy no it's a rock
2:22:28
oh it's happy no it's a rock
2:22:28
oh it's happy no it's a rock it's a rock and and and people get mad
2:22:30
it's a rock and and and people get mad
2:22:30
it's a rock and and and people get mad at me they're like well you're calling
2:22:31
at me they're like well you're calling
2:22:32
at me they're like well you're calling ai a rock
2:22:32
ai a rock
2:22:32
ai a rock yes i am yeah it's a rock it's not
2:22:36
yes i am yeah it's a rock it's not
2:22:36
yes i am yeah it's a rock it's not data burned into it it's very fancy rock
2:22:39
data burned into it it's very fancy rock
2:22:39
data burned into it it's very fancy rock but it is still fundamentally it's a
2:22:41
but it is still fundamentally it's a
2:22:41
but it is still fundamentally it's a tool it's a tool we made
2:22:44
tool it's a tool we made
2:22:44
tool it's a tool we made yeah and and then everyone all of a
2:22:46
yeah and and then everyone all of a
2:22:46
yeah and and then everyone all of a sudden we're like
2:22:47
sudden we're like
2:22:47
sudden we're like oh and then the sales dudes from the
2:22:49
oh and then the sales dudes from the
2:22:49
oh and then the sales dudes from the other companies were like well i don't
2:22:50
other companies were like well i don't
2:22:50
other companies were like well i don't know what to say after this
2:22:52
know what to say after this
2:22:52
know what to say after this how do we solve it when well and there
2:22:54
how do we solve it when well and there
2:22:54
how do we solve it when well and there lies the problem like we've
2:22:56
lies the problem like we've
2:22:56
lies the problem like we've we've taken a lot of technologies that
2:22:58
we've taken a lot of technologies that
2:22:58
we've taken a lot of technologies that already existed and
2:22:59
already existed and
2:22:59
already existed and bundled them under ai because it was hip
2:23:02
bundled them under ai because it was hip
2:23:02
bundled them under ai because it was hip and it's profitable right
2:23:03
and it's profitable right
2:23:03
and it's profitable right you know you get the joke of the the um
2:23:07
you know you get the joke of the the um
2:23:07
you know you get the joke of the the um the beverage company that just added ai
2:23:09
the beverage company that just added ai
2:23:09
the beverage company that just added ai to the name of their beverage and sold
2:23:11
to the name of their beverage and sold
2:23:11
to the name of their beverage and sold more
2:23:12
more
2:23:12
more it's like you know it's just a naming
2:23:15
it's like you know it's just a naming
2:23:15
it's like you know it's just a naming strategy
2:23:16
strategy
2:23:16
strategy and so you know there's a downside to
2:23:18
and so you know there's a downside to
2:23:18
and so you know there's a downside to that which is that it does
2:23:19
that which is that it does
2:23:19
that which is that it does tend to distort things but i also
2:23:21
tend to distort things but i also
2:23:21
tend to distort things but i also appreciate that as
2:23:22
appreciate that as
2:23:22
appreciate that as soon as it's a viable product in any way
2:23:25
soon as it's a viable product in any way
2:23:25
soon as it's a viable product in any way you give it a new name
2:23:27
you give it a new name
2:23:27
you give it a new name it gets pulled out that is true
2:23:31
it gets pulled out that is true
2:23:31
it gets pulled out that is true and i i would that was interesting that
2:23:32
and i i would that was interesting that
2:23:32
and i i would that was interesting that little that chart that you did it was
2:23:34
little that chart that you did it was
2:23:34
little that chart that you did it was interesting because i was like oh yeah
2:23:36
interesting because i was like oh yeah
2:23:36
interesting because i was like oh yeah once you give it a name and you strip
2:23:38
once you give it a name and you strip
2:23:38
once you give it a name and you strip like the mystique out of it you can
2:23:41
like the mystique out of it you can
2:23:41
like the mystique out of it you can you can make a pretty good living off of
2:23:43
you can make a pretty good living off of
2:23:43
you can make a pretty good living off of it and this is where i make sales
2:23:45
it and this is where i make sales
2:23:45
it and this is where i make sales upset and say we use artificial
2:23:46
upset and say we use artificial
2:23:46
upset and say we use artificial intelligences my experience is anything
2:23:48
intelligences my experience is anything
2:23:48
intelligences my experience is anything labeled artificial tells us stuff that
2:23:49
labeled artificial tells us stuff that
2:23:49
labeled artificial tells us stuff that doesn't work
2:23:52
when it does work it gets a new name
2:23:56
and that's why i guess i am an ai cloud
2:23:59
and that's why i guess i am an ai cloud
2:23:59
and that's why i guess i am an ai cloud advocate
2:24:00
advocate
2:24:00
advocate because i really don't work
2:24:03
because i really don't work
2:24:03
because i really don't work haven't worked for years seth we've done
2:24:05
haven't worked for years seth we've done
2:24:06
haven't worked for years seth we've done stuff together now for
2:24:07
stuff together now for
2:24:07
stuff together now for for more than a decade that's right
2:24:10
for more than a decade that's right
2:24:10
for more than a decade that's right and uh know you have way too much fun we
2:24:15
and uh know you have way too much fun we
2:24:15
and uh know you have way too much fun we can see it on your face
2:24:16
can see it on your face
2:24:16
can see it on your face that's right that's right richard we get
2:24:19
that's right that's right richard we get
2:24:19
that's right that's right richard we get into these discussions sometimes and
2:24:21
into these discussions sometimes and
2:24:21
into these discussions sometimes and it's like
2:24:22
it's like
2:24:22
it's like yeah this is this is unfair because
2:24:24
yeah this is this is unfair because
2:24:24
yeah this is this is unfair because we've we've known each other so long i
2:24:26
we've we've known each other so long i
2:24:26
we've we've known each other so long i mean i can tease him endlessly really
2:24:28
mean i can tease him endlessly really
2:24:28
mean i can tease him endlessly really that's right there is endless things to
2:24:30
that's right there is endless things to
2:24:30
that's right there is endless things to tease me about
2:24:32
tease me about
2:24:32
tease me about it's true in this case since my sound is
2:24:35
it's true in this case since my sound is
2:24:35
it's true in this case since my sound is really bad today
2:24:36
really bad today
2:24:36
really bad today i'm really happy that you're having this
2:24:37
i'm really happy that you're having this
2:24:37
i'm really happy that you're having this conversation actually and i'm learning a
2:24:40
conversation actually and i'm learning a
2:24:40
conversation actually and i'm learning a lot from this one uh
2:24:41
lot from this one uh
2:24:41
lot from this one uh especially and i it's totally fine so
2:24:44
especially and i it's totally fine so
2:24:44
especially and i it's totally fine so i've got a question
2:24:45
i've got a question
2:24:45
i've got a question um we talked about these neat guys
2:24:48
um we talked about these neat guys
2:24:48
um we talked about these neat guys instructor guys
2:24:49
instructor guys
2:24:49
instructor guys do you see any crossovers happening from
2:24:51
do you see any crossovers happening from
2:24:51
do you see any crossovers happening from the scrappy guys
2:24:52
the scrappy guys
2:24:52
the scrappy guys and i have to find this game since my
2:24:54
and i have to find this game since my
2:24:54
and i have to find this game since my sound is really bad to establish proof
2:24:56
sound is really bad to establish proof
2:24:56
sound is really bad to establish proof i'm really happy that you're having this
2:24:58
i'm really happy that you're having this
2:24:58
i'm really happy that you're having this conversation actually
2:24:59
conversation actually
2:24:59
conversation actually well are you asking just swing both ways
2:25:03
well are you asking just swing both ways
2:25:03
well are you asking just swing both ways yeah i think he does
2:25:04
yeah i think he does
2:25:04
yeah i think he does but i do and i were to go back to school
2:25:15
the next seven or eight years trying to
2:25:17
the next seven or eight years trying to
2:25:17
the next seven or eight years trying to come up with formal methods
2:25:19
come up with formal methods
2:25:19
come up with formal methods or verification methods for neural
2:25:21
or verification methods for neural
2:25:21
or verification methods for neural networks maybe
2:25:22
networks maybe
2:25:22
networks maybe for example pac learning works
2:25:25
for example pac learning works
2:25:25
for example pac learning works in some way shape or form with deep
2:25:27
in some way shape or form with deep
2:25:27
in some way shape or form with deep learning maybe there is an
2:25:28
learning maybe there is an
2:25:28
learning maybe there is an there is a structural model beyond
2:25:32
there is a structural model beyond
2:25:32
there is a structural model beyond deep learning that we can use because in
2:25:34
deep learning that we can use because in
2:25:34
deep learning that we can use because in the end it's all
2:25:35
the end it's all
2:25:35
the end it's all dot products and functions and like
2:25:38
dot products and functions and like
2:25:38
dot products and functions and like vector
2:25:38
vector
2:25:38
vector subtraction i mean that's all it really
2:25:40
subtraction i mean that's all it really
2:25:40
subtraction i mean that's all it really is i had read some of the papers coming
2:25:42
is i had read some of the papers coming
2:25:42
is i had read some of the papers coming out of the explainable ai effort from
2:25:44
out of the explainable ai effort from
2:25:44
out of the explainable ai effort from darpa
2:25:45
darpa
2:25:45
darpa that very much they were talking about
2:25:48
that very much they were talking about
2:25:48
that very much they were talking about that particular problem of we're trying
2:25:49
that particular problem of we're trying
2:25:49
that particular problem of we're trying to understand why neural nets are coming
2:25:51
to understand why neural nets are coming
2:25:51
to understand why neural nets are coming to the decisions that they're
2:25:52
to the decisions that they're
2:25:52
to the decisions that they're they are where they find their
2:25:53
they are where they find their
2:25:53
they are where they find their probabilities and they were using
2:25:54
probabilities and they were using
2:25:54
probabilities and they were using decision trees
2:25:55
decision trees
2:25:55
decision trees to explore the neural net like it's just
2:25:57
to explore the neural net like it's just
2:25:57
to explore the neural net like it's just interesting to me that
2:25:59
interesting to me that
2:25:59
interesting to me that those two disciplines feed against each
2:26:01
those two disciplines feed against each
2:26:01
those two disciplines feed against each other ultimately
2:26:02
other ultimately
2:26:02
other ultimately or could feed against each other
2:26:03
or could feed against each other
2:26:04
or could feed against each other ultimately to help them both be better
2:26:06
ultimately to help them both be better
2:26:06
ultimately to help them both be better people don't know this but when it
2:26:08
people don't know this but when it
2:26:08
people don't know this but when it outputs probabilities
2:26:10
outputs probabilities
2:26:10
outputs probabilities they're not actually probabilities they
2:26:12
they're not actually probabilities they
2:26:12
they're not actually probabilities they just put a function at the end that
2:26:13
just put a function at the end that
2:26:13
just put a function at the end that forces them all to sum to one
2:26:15
forces them all to sum to one
2:26:16
forces them all to sum to one yeah it's just a number between zero and
2:26:18
yeah it's just a number between zero and
2:26:18
yeah it's just a number between zero and one
2:26:19
one
2:26:19
one never zero never won yeah and it's crazy
2:26:23
never zero never won yeah and it's crazy
2:26:23
never zero never won yeah and it's crazy that that that that people are like oh
2:26:24
that that that that people are like oh
2:26:24
that that that that people are like oh here's the probabilities and
2:26:26
here's the probabilities and
2:26:26
here's the probabilities and like if you talk to a pure statistician
2:26:28
like if you talk to a pure statistician
2:26:28
like if you talk to a pure statistician they'll literally just be like
2:26:29
they'll literally just be like
2:26:29
they'll literally just be like that's it they'll you know they'll throw
2:26:30
that's it they'll you know they'll throw
2:26:30
that's it they'll you know they'll throw their ball on the floor and leave the
2:26:32
their ball on the floor and leave the
2:26:32
their ball on the floor and leave the house right because there was a cool
2:26:34
house right because there was a cool
2:26:34
house right because there was a cool meme
2:26:35
meme
2:26:35
meme there was a cool meme i saw like there
2:26:36
there was a cool meme i saw like there
2:26:36
there was a cool meme i saw like there must have been a bar fight somewhere and
2:26:38
must have been a bar fight somewhere and
2:26:38
must have been a bar fight somewhere and people were throwing chairs at each
2:26:39
people were throwing chairs at each
2:26:39
people were throwing chairs at each other and
2:26:40
other and
2:26:40
other and the statistician said looks like
2:26:42
the statistician said looks like
2:26:42
the statistician said looks like somebody just said machine learning is
2:26:43
somebody just said machine learning is
2:26:44
somebody just said machine learning is just applied statistics and everyone's
2:26:45
just applied statistics and everyone's
2:26:45
just applied statistics and everyone's going to change each other
2:26:52
just a bunch of if statements
2:27:12
so what'd you say yeah i i think so i
2:27:17
so what'd you say yeah i i think so i
2:27:17
so what'd you say yeah i i think so i mean there are
2:27:17
mean there are
2:27:17
mean there are any and we're ultimately doing the
2:27:19
any and we're ultimately doing the
2:27:19
any and we're ultimately doing the engineering of applying these things and
2:27:21
engineering of applying these things and
2:27:21
engineering of applying these things and getting useful results
2:27:22
getting useful results
2:27:22
getting useful results and every time we do that i think we
2:27:24
and every time we do that i think we
2:27:24
and every time we do that i think we need help to some degree
2:27:26
need help to some degree
2:27:26
need help to some degree to to validate things there's plenty of
2:27:29
to to validate things there's plenty of
2:27:29
to to validate things there's plenty of bad models too
2:27:30
bad models too
2:27:30
bad models too right i i do feel like
2:27:34
right i i do feel like
2:27:34
right i i do feel like our profession is being improved by the
2:27:37
our profession is being improved by the
2:27:37
our profession is being improved by the challenges of dealing with
2:27:39
challenges of dealing with
2:27:39
challenges of dealing with data at the scale right that that we
2:27:41
data at the scale right that that we
2:27:41
data at the scale right that that we have
2:27:42
have
2:27:42
have the ethical conversation has always been
2:27:44
the ethical conversation has always been
2:27:44
the ethical conversation has always been there
2:27:45
there
2:27:45
there we're now surfacing it in a much higher
2:27:47
we're now surfacing it in a much higher
2:27:47
we're now surfacing it in a much higher level
2:27:48
level
2:27:48
level uh and and it is a part of this is the
2:27:51
uh and and it is a part of this is the
2:27:51
uh and and it is a part of this is the anthropomorphization effect
2:27:52
anthropomorphization effect
2:27:52
anthropomorphization effect when we threw the word intelligence in
2:27:54
when we threw the word intelligence in
2:27:54
when we threw the word intelligence in this and we we are
2:27:55
this and we we are
2:27:55
this and we we are handing these tools agency they
2:27:58
handing these tools agency they
2:27:58
handing these tools agency they ultimately we should not have
2:28:00
ultimately we should not have
2:28:00
ultimately we should not have it's made us look back at ourselves and
2:28:03
it's made us look back at ourselves and
2:28:03
it's made us look back at ourselves and say
2:28:04
say
2:28:04
say these are not the data's biases these
2:28:06
these are not the data's biases these
2:28:06
these are not the data's biases these are our biases
2:28:08
are our biases
2:28:08
are our biases that are manifest in data and now we're
2:28:10
that are manifest in data and now we're
2:28:10
that are manifest in data and now we're using tools
2:28:11
using tools
2:28:11
using tools to amplify them and i said i've used
2:28:12
to amplify them and i said i've used
2:28:12
to amplify them and i said i've used this line in the board rooms like
2:28:14
this line in the board rooms like
2:28:14
this line in the board rooms like computers are only amplifiers
2:28:15
computers are only amplifiers
2:28:15
computers are only amplifiers they can amplify your intelligence or
2:28:17
they can amplify your intelligence or
2:28:17
they can amplify your intelligence or they can amplify your stupidity
2:28:19
they can amplify your stupidity
2:28:19
they can amplify your stupidity which would you like yeah and i would
2:28:22
which would you like yeah and i would
2:28:22
which would you like yeah and i would add that even even in my limited look at
2:28:24
add that even even in my limited look at
2:28:24
add that even even in my limited look at lyman's shop they're they're basically
2:28:26
lyman's shop they're they're basically
2:28:26
lyman's shop they're they're basically looking
2:28:27
looking
2:28:27
looking at a model post facto after
2:28:30
at a model post facto after
2:28:30
at a model post facto after it's been built like for me i would i
2:28:33
it's been built like for me i would i
2:28:34
it's been built like for me i would i would like to know more
2:28:35
would like to know more
2:28:35
would like to know more about it pre facto is that a term
2:28:38
about it pre facto is that a term
2:28:38
about it pre facto is that a term right like i'd love to know like what is
2:28:41
right like i'd love to know like what is
2:28:41
right like i'd love to know like what is the
2:28:43
the
2:28:43
the the the the power in this thing for
2:28:45
the the the power in this thing for
2:28:45
the the the power in this thing for example like i said there there is
2:28:46
example like i said there there is
2:28:46
example like i said there there is something in pack learning called vc
2:28:48
something in pack learning called vc
2:28:48
something in pack learning called vc dimension
2:28:49
dimension
2:28:49
dimension where like a certain machine learning
2:28:51
where like a certain machine learning
2:28:51
where like a certain machine learning model can shatter a space it
2:28:52
model can shatter a space it
2:28:52
model can shatter a space it it's called a shattering and you use
2:28:54
it's called a shattering and you use
2:28:54
it's called a shattering and you use that that's how many spaces you can
2:28:56
that that's how many spaces you can
2:28:56
that that's how many spaces you can shatter
2:28:56
shatter
2:28:56
shatter to do some approximations that's a
2:28:58
to do some approximations that's a
2:28:58
to do some approximations that's a formal method
2:29:00
formal method
2:29:00
formal method of looking at models but you can't do
2:29:02
of looking at models but you can't do
2:29:02
of looking at models but you can't do that with deep learning right now
2:29:03
that with deep learning right now
2:29:03
that with deep learning right now because there's no way to know
2:29:05
because there's no way to know
2:29:05
because there's no way to know like how many what the vc dimension of a
2:29:07
like how many what the vc dimension of a
2:29:07
like how many what the vc dimension of a particular
2:29:08
particular
2:29:08
particular deep learning model is because it's it's
2:29:10
deep learning model is because it's it's
2:29:10
deep learning model is because it's it's so variable and that's
2:29:12
so variable and that's
2:29:12
so variable and that's like those those those are all post
2:29:15
like those those those are all post
2:29:15
like those those those are all post facto
2:29:15
facto
2:29:16
facto let's look at a model after it was built
2:29:18
let's look at a model after it was built
2:29:18
let's look at a model after it was built as opposed to what is it that we're
2:29:19
as opposed to what is it that we're
2:29:20
as opposed to what is it that we're trying to build
2:29:21
trying to build
2:29:21
trying to build and how expressive is it you know i
2:29:24
and how expressive is it you know i
2:29:24
and how expressive is it you know i wonder if
2:29:25
wonder if
2:29:25
wonder if it's almost a nutritional content
2:29:27
it's almost a nutritional content
2:29:27
it's almost a nutritional content problem right it's like
2:29:28
problem right it's like
2:29:28
problem right it's like you don't eat the food and then decide
2:29:31
you don't eat the food and then decide
2:29:31
you don't eat the food and then decide whether it was good for you or not
2:29:33
whether it was good for you or not
2:29:33
whether it was good for you or not you kind of read the label right we've
2:29:34
you kind of read the label right we've
2:29:34
you kind of read the label right we've created a construct that says this is
2:29:36
created a construct that says this is
2:29:36
created a construct that says this is the contents of these things this is the
2:29:38
the contents of these things this is the
2:29:38
the contents of these things this is the nutritional value
2:29:39
nutritional value
2:29:39
nutritional value and i'm starting to feel more and more
2:29:40
and i'm starting to feel more and more
2:29:40
and i'm starting to feel more and more about that about data like i don't want
2:29:42
about that about data like i don't want
2:29:42
about that about data like i don't want to read this data until i know
2:29:44
to read this data until i know
2:29:44
to read this data until i know its basic qualities i need its
2:29:46
its basic qualities i need its
2:29:46
its basic qualities i need its nutritional content
2:29:48
nutritional content
2:29:48
nutritional content because then the ultimately these models
2:29:51
because then the ultimately these models
2:29:51
because then the ultimately these models would be manifestations of the sources
2:29:52
would be manifestations of the sources
2:29:52
would be manifestations of the sources of information
2:29:54
of information
2:29:54
of information and so if we qualify the source of
2:29:55
and so if we qualify the source of
2:29:55
and so if we qualify the source of information we should have the potential
2:29:57
information we should have the potential
2:29:57
information we should have the potential for better models
2:29:58
for better models
2:29:58
for better models my my advisor in grad school hal dalmate
2:30:01
my my advisor in grad school hal dalmate
2:30:01
my my advisor in grad school hal dalmate you can look him up he
2:30:02
you can look him up he
2:30:02
you can look him up he co-authored a paper paper called data
2:30:04
co-authored a paper paper called data
2:30:04
co-authored a paper paper called data sheets
2:30:05
sheets
2:30:05
sheets which in theory is supposed to solve
2:30:07
which in theory is supposed to solve
2:30:07
which in theory is supposed to solve this problem
2:30:08
this problem
2:30:08
this problem in in essence what you do is you have a
2:30:11
in in essence what you do is you have a
2:30:11
in in essence what you do is you have a data set
2:30:11
data set
2:30:11
data set that is accompanied with a thing called
2:30:13
that is accompanied with a thing called
2:30:13
that is accompanied with a thing called the data sheet that says
2:30:15
the data sheet that says
2:30:15
the data sheet that says this is what we were thinking when we
2:30:16
this is what we were thinking when we
2:30:16
this is what we were thinking when we made this data this is what's in the
2:30:18
made this data this is what's in the
2:30:18
made this data this is what's in the data
2:30:19
data
2:30:19
data this is where we got it from this is the
2:30:21
this is where we got it from this is the
2:30:21
this is where we got it from this is the segment of people that it came from
2:30:23
segment of people that it came from
2:30:23
segment of people that it came from right
2:30:23
right
2:30:23
right and so that's an interesting idea but
2:30:25
and so that's an interesting idea but
2:30:25
and so that's an interesting idea but notice that's not a
2:30:26
notice that's not a
2:30:26
notice that's not a that's that's an interesting like human
2:30:29
that's that's an interesting like human
2:30:29
that's that's an interesting like human approach to it but again there's no
2:30:32
approach to it but again there's no
2:30:32
approach to it but again there's no mathematical way
2:30:35
the way you described it sound like
2:30:37
the way you described it sound like
2:30:37
the way you described it sound like intent rather than
2:30:39
intent rather than
2:30:39
intent rather than content right what's actually in the box
2:30:43
content right what's actually in the box
2:30:44
content right what's actually in the box and those are called data sheets and you
2:30:45
and those are called data sheets and you
2:30:45
and those are called data sheets and you can you can look that up there's also
2:30:47
can you can look that up there's also
2:30:47
can you can look that up there's also something called model cards
2:30:48
something called model cards
2:30:48
something called model cards which are similar things but they
2:30:50
which are similar things but they
2:30:50
which are similar things but they describe a particular model
2:30:52
describe a particular model
2:30:52
describe a particular model and that the this is the research that's
2:30:54
and that the this is the research that's
2:30:54
and that the this is the research that's going on right now but to me
2:30:57
going on right now but to me
2:30:57
going on right now but to me like you hit the scruffs versus the
2:30:58
like you hit the scruffs versus the
2:30:58
like you hit the scruffs versus the needs thing very well like it it doesn't
2:31:01
needs thing very well like it it doesn't
2:31:01
needs thing very well like it it doesn't hit me in the right spot you know
2:31:04
hit me in the right spot you know
2:31:04
hit me in the right spot you know it's like i feel like it's still too
2:31:06
it's like i feel like it's still too
2:31:06
it's like i feel like it's still too fuzzy
2:31:08
fuzzy
2:31:08
fuzzy right well and both of them were
2:31:10
right well and both of them were
2:31:10
right well and both of them were consuming data
2:31:11
consuming data
2:31:11
consuming data uh largely from similar sources so you
2:31:14
uh largely from similar sources so you
2:31:14
uh largely from similar sources so you we still have this overarching problem
2:31:17
we still have this overarching problem
2:31:17
we still have this overarching problem and every data scientist i've ever
2:31:18
and every data scientist i've ever
2:31:18
and every data scientist i've ever spoken to says the same thing it's like
2:31:19
spoken to says the same thing it's like
2:31:19
spoken to says the same thing it's like you spend 80 percent of the time
2:31:21
you spend 80 percent of the time
2:31:22
you spend 80 percent of the time validating the data in some ways the
2:31:24
validating the data in some ways the
2:31:24
validating the data in some ways the model is
2:31:25
model is
2:31:25
model is kind of the easy part do you know what
2:31:28
kind of the easy part do you know what
2:31:28
kind of the easy part do you know what you have
2:31:28
you have
2:31:28
you have do you know the nutritional content of
2:31:31
do you know the nutritional content of
2:31:31
do you know the nutritional content of this data set
2:31:33
this data set
2:31:33
this data set yeah and that's that's the that's the
2:31:35
yeah and that's that's the that's the
2:31:36
yeah and that's that's the that's the rub
2:31:36
rub
2:31:36
rub most people don't and then because they
2:31:38
most people don't and then because they
2:31:38
most people don't and then because they don't whatever
2:31:40
don't whatever
2:31:40
don't whatever whatever bias is burned into the data
2:31:43
whatever bias is burned into the data
2:31:43
whatever bias is burned into the data i'll give you a simple example uh and
2:31:46
i'll give you a simple example uh and
2:31:46
i'll give you a simple example uh and this i can find you sources later on but
2:31:48
this i can find you sources later on but
2:31:48
this i can find you sources later on but it turns out that in the united states
2:31:49
it turns out that in the united states
2:31:50
it turns out that in the united states we over police a certain segment of our
2:31:51
we over police a certain segment of our
2:31:51
we over police a certain segment of our population
2:31:53
population
2:31:53
population so if a police if a police precinct were
2:31:55
so if a police if a police precinct were
2:31:55
so if a police if a police precinct were to say to me seth i want you to build an
2:31:57
to say to me seth i want you to build an
2:31:57
to say to me seth i want you to build an ai model
2:31:58
ai model
2:31:58
ai model to tell us where we should go find crime
2:32:01
to tell us where we should go find crime
2:32:01
to tell us where we should go find crime well if i use their data what do you
2:32:04
well if i use their data what do you
2:32:04
well if i use their data what do you think it's going to tell them
2:32:05
think it's going to tell them
2:32:05
think it's going to tell them to do or to where to go right wherever
2:32:07
to do or to where to go right wherever
2:32:08
to do or to where to go right wherever they went before
2:32:08
they went before
2:32:08
they went before for crime yeah and it turns out that if
2:32:11
for crime yeah and it turns out that if
2:32:11
for crime yeah and it turns out that if if they over
2:32:12
if they over
2:32:12
if they over index on a certain sub-population of
2:32:14
index on a certain sub-population of
2:32:14
index on a certain sub-population of american citizens
2:32:16
american citizens
2:32:16
american citizens the model will say yup that's where we
2:32:19
the model will say yup that's where we
2:32:19
the model will say yup that's where we go to find crime because it turns out
2:32:20
go to find crime because it turns out
2:32:20
go to find crime because it turns out that if you're looking for crime you're
2:32:22
that if you're looking for crime you're
2:32:22
that if you're looking for crime you're probably going to find it and if you are
2:32:24
probably going to find it and if you are
2:32:24
probably going to find it and if you are looking in a specific place
2:32:26
looking in a specific place
2:32:26
looking in a specific place every time then all of a sudden there's
2:32:29
every time then all of a sudden there's
2:32:29
every time then all of a sudden there's a double problem
2:32:30
a double problem
2:32:30
a double problem right now you have computers saying that
2:32:32
right now you have computers saying that
2:32:32
right now you have computers saying that these are where the crimes happen
2:32:34
these are where the crimes happen
2:32:34
these are where the crimes happen and that's just a simple example but
2:32:36
and that's just a simple example but
2:32:36
and that's just a simple example but then and it gives it this credibility
2:32:38
then and it gives it this credibility
2:32:38
then and it gives it this credibility it's like it came from the computer it
2:32:39
it's like it came from the computer it
2:32:39
it's like it came from the computer it must be correct
2:32:41
must be correct
2:32:41
must be correct so you're working with ai on a daily
2:32:44
so you're working with ai on a daily
2:32:44
so you're working with ai on a daily basis
2:32:45
basis
2:32:45
basis how do you for example come back why is
2:32:47
how do you for example come back why is
2:32:47
how do you for example come back why is in your data sets
2:32:50
in your data sets
2:32:50
in your data sets yeah that's a good question i think we
2:32:51
yeah that's a good question i think we
2:32:51
yeah that's a good question i think we started uh with that like
2:32:53
started uh with that like
2:32:53
started uh with that like are you sampling are you doing your
2:32:56
are you sampling are you doing your
2:32:56
are you sampling are you doing your sampling
2:32:56
sampling
2:32:56
sampling part right and again are you looking
2:32:58
part right and again are you looking
2:32:58
part right and again are you looking with the right right sheet i mean
2:33:01
with the right right sheet i mean
2:33:01
with the right right sheet i mean uh it's hard and and it's uh i agree
2:33:06
uh it's hard and and it's uh i agree
2:33:06
uh it's hard and and it's uh i agree with seth
2:33:06
with seth
2:33:06
with seth i mean you're training your model based
2:33:09
i mean you're training your model based
2:33:09
i mean you're training your model based on historic facts so
2:33:11
on historic facts so
2:33:11
on historic facts so yeah if you're looking for crime you
2:33:13
yeah if you're looking for crime you
2:33:13
yeah if you're looking for crime you will find it right i mean you find
2:33:15
will find it right i mean you find
2:33:15
will find it right i mean you find where it works so and as i said before i
2:33:18
where it works so and as i said before i
2:33:18
where it works so and as i said before i think
2:33:19
think
2:33:19
think we humans are already biased so
2:33:22
we humans are already biased so
2:33:22
we humans are already biased so how can you get the bias out of yourself
2:33:25
how can you get the bias out of yourself
2:33:25
how can you get the bias out of yourself and
2:33:26
and
2:33:26
and well i think that's also a challenge in
2:33:27
well i think that's also a challenge in
2:33:27
well i think that's also a challenge in itself
2:33:29
itself
2:33:29
itself i i you know the data sheet concept is
2:33:31
i i you know the data sheet concept is
2:33:31
i i you know the data sheet concept is great in the sense that at least you're
2:33:33
great in the sense that at least you're
2:33:33
great in the sense that at least you're writing down your bias
2:33:34
writing down your bias
2:33:34
writing down your bias or what you believe your biases i think
2:33:37
or what you believe your biases i think
2:33:37
or what you believe your biases i think that's also nice
2:33:38
that's also nice
2:33:38
that's also nice yeah and because you're never going to
2:33:40
yeah and because you're never going to
2:33:40
yeah and because you're never going to eliminate bias the best you can do is
2:33:42
eliminate bias the best you can do is
2:33:42
eliminate bias the best you can do is label it
2:33:43
label it
2:33:43
label it and so the idea that we would
2:33:45
and so the idea that we would
2:33:45
and so the idea that we would deliberately label
2:33:46
deliberately label
2:33:46
deliberately label this is what we were thinking this is
2:33:48
this is what we were thinking this is
2:33:48
this is what we were thinking this is what we you know our approach
2:33:50
what we you know our approach
2:33:50
what we you know our approach that also tells you all the things they
2:33:52
that also tells you all the things they
2:33:52
that also tells you all the things they weren't doing so when you're
2:33:53
weren't doing so when you're
2:33:54
weren't doing so when you're going to look at that data possibly for
2:33:55
going to look at that data possibly for
2:33:55
going to look at that data possibly for a different purpose at least you know
2:33:57
a different purpose at least you know
2:33:57
a different purpose at least you know this was not the intent of this data so
2:33:59
this was not the intent of this data so
2:33:59
this was not the intent of this data so you know just tread carefully
2:34:01
you know just tread carefully
2:34:01
you know just tread carefully yeah and then i think now you mentioned
2:34:03
yeah and then i think now you mentioned
2:34:03
yeah and then i think now you mentioned also i think the intent of the data
2:34:05
also i think the intent of the data
2:34:05
also i think the intent of the data that's also let's say method
2:34:07
that's also let's say method
2:34:07
that's also let's say method methodological problem that
2:34:09
methodological problem that
2:34:09
methodological problem that you should use your data where you had
2:34:12
you should use your data where you had
2:34:12
you should use your data where you had the intent for so
2:34:13
the intent for so
2:34:13
the intent for so you cannot just cross use it without
2:34:17
you cannot just cross use it without
2:34:17
you cannot just cross use it without thinking i mean yeah but that's what i
2:34:19
thinking i mean yeah but that's what i
2:34:19
thinking i mean yeah but that's what i think the data sheet serves is like are
2:34:21
think the data sheet serves is like are
2:34:21
think the data sheet serves is like are you to the purpose of the data from this
2:34:23
you to the purpose of the data from this
2:34:23
you to the purpose of the data from this original intent
2:34:24
original intent
2:34:24
original intent or are you cross-purpose to it and so
2:34:26
or are you cross-purpose to it and so
2:34:26
or are you cross-purpose to it and so some additional steps need to be taken
2:34:28
some additional steps need to be taken
2:34:28
some additional steps need to be taken yeah overall i think it's a
2:34:31
yeah overall i think it's a
2:34:32
yeah overall i think it's a it's fascinating that we we've kind of
2:34:35
it's fascinating that we we've kind of
2:34:35
it's fascinating that we we've kind of we've all of a sudden we're now to the
2:34:37
we've all of a sudden we're now to the
2:34:37
we've all of a sudden we're now to the point where we have some pretty
2:34:39
point where we have some pretty
2:34:39
point where we have some pretty sophisticated models
2:34:40
sophisticated models
2:34:40
sophisticated models and now we're stopping to think hey
2:34:44
and now we're stopping to think hey
2:34:44
and now we're stopping to think hey maybe we should be smarter about how
2:34:46
maybe we should be smarter about how
2:34:46
maybe we should be smarter about how these things are used and why and when
2:34:48
these things are used and why and when
2:34:48
these things are used and why and when and where
2:34:49
and where
2:34:49
and where that's but that's also pretty normal in
2:34:51
that's but that's also pretty normal in
2:34:51
that's but that's also pretty normal in in human civilization it's only when
2:34:53
in human civilization it's only when
2:34:53
in human civilization it's only when something
2:34:54
something
2:34:54
something reaches a certain point that we go we
2:34:56
reaches a certain point that we go we
2:34:56
reaches a certain point that we go we should have some rules right i mean the
2:34:58
should have some rules right i mean the
2:34:58
should have some rules right i mean the car came before the driver's license
2:35:00
car came before the driver's license
2:35:00
car came before the driver's license it was only until we had a certain
2:35:02
it was only until we had a certain
2:35:02
it was only until we had a certain number of cars is like
2:35:03
number of cars is like
2:35:03
number of cars is like traffic lights that's an idea right
2:35:06
traffic lights that's an idea right
2:35:06
traffic lights that's an idea right standardization of controls
2:35:07
standardization of controls
2:35:07
standardization of controls good idea like all of those things
2:35:09
good idea like all of those things
2:35:09
good idea like all of those things emerge with
2:35:10
emerge with
2:35:10
emerge with the expansion of something so
2:35:14
the expansion of something so
2:35:14
the expansion of something so i do think we've hit a critical point
2:35:15
i do think we've hit a critical point
2:35:15
i do think we've hit a critical point where we have so much data where we now
2:35:17
where we have so much data where we now
2:35:17
where we have so much data where we now press against people's privacy
2:35:18
press against people's privacy
2:35:18
press against people's privacy and attention all of the time and so
2:35:20
and attention all of the time and so
2:35:20
and attention all of the time and so those are now valued things
2:35:23
those are now valued things
2:35:23
those are now valued things precious things and things that need to
2:35:24
precious things and things that need to
2:35:24
precious things and things that need to be protected
2:35:26
be protected
2:35:26
be protected so for both of you what are the what are
2:35:27
so for both of you what are the what are
2:35:27
so for both of you what are the what are the traffic lights
2:35:29
the traffic lights
2:35:29
the traffic lights and the driver's licenses for
2:35:32
and the driver's licenses for
2:35:32
and the driver's licenses for machine learning or ai models what do
2:35:34
machine learning or ai models what do
2:35:34
machine learning or ai models what do you think those
2:35:35
you think those
2:35:36
you think those should look like what should they be and
2:35:38
should look like what should they be and
2:35:38
should look like what should they be and we'll start with you
2:35:39
we'll start with you
2:35:39
we'll start with you miriam and then we'll get we'll go over
2:35:41
miriam and then we'll get we'll go over
2:35:41
miriam and then we'll get we'll go over to you richard
2:35:43
to you richard
2:35:43
to you richard wow that's a hard question it is um
2:35:46
wow that's a hard question it is um
2:35:46
wow that's a hard question it is um again an answer would be that depends uh
2:35:49
again an answer would be that depends uh
2:35:49
again an answer would be that depends uh it depends on the algorithm
2:35:51
it depends on the algorithm
2:35:51
it depends on the algorithm it depends on the problem you're solving
2:35:53
it depends on the problem you're solving
2:35:53
it depends on the problem you're solving um
2:35:55
um
2:35:55
um so based on the algorithm you would have
2:35:57
so based on the algorithm you would have
2:35:57
so based on the algorithm you would have i would say different rules
2:35:59
i would say different rules
2:35:59
i would say different rules and different controls or or
2:36:02
and different controls or or
2:36:02
and different controls or or stopping lights
2:36:05
i i i i don't know maybe richard you can
2:36:09
i i i i don't know maybe richard you can
2:36:09
i i i i don't know maybe richard you can hop
2:36:09
hop
2:36:09
hop in but well you know we've had a barrier
2:36:12
in but well you know we've had a barrier
2:36:12
in but well you know we've had a barrier to entry right now in complexity
2:36:14
to entry right now in complexity
2:36:14
to entry right now in complexity right if only certain people could be
2:36:17
right if only certain people could be
2:36:17
right if only certain people could be successful with this
2:36:18
successful with this
2:36:18
successful with this that it took certain levels of skill and
2:36:20
that it took certain levels of skill and
2:36:20
that it took certain levels of skill and and that level is starting
2:36:21
and that level is starting
2:36:22
and that level is starting to diminish is part of the reason that
2:36:23
to diminish is part of the reason that
2:36:23
to diminish is part of the reason that it's it's surfacing so thoroughly
2:36:25
it's it's surfacing so thoroughly
2:36:25
it's it's surfacing so thoroughly what we're ending up with is a
2:36:28
what we're ending up with is a
2:36:28
what we're ending up with is a gatekeeper
2:36:29
gatekeeper
2:36:29
gatekeeper concept right now and you're seeing the
2:36:31
concept right now and you're seeing the
2:36:31
concept right now and you're seeing the pressure on organizations
2:36:33
pressure on organizations
2:36:33
pressure on organizations like microsoft and ibm and google and
2:36:35
like microsoft and ibm and google and
2:36:35
like microsoft and ibm and google and facebook and so forth those that are
2:36:37
facebook and so forth those that are
2:36:37
facebook and so forth those that are largely serving as gatekeepers to the
2:36:39
largely serving as gatekeepers to the
2:36:39
largely serving as gatekeepers to the access to this data
2:36:40
access to this data
2:36:40
access to this data and these models to say take some
2:36:43
and these models to say take some
2:36:43
and these models to say take some responsibility on that then and largely
2:36:45
responsibility on that then and largely
2:36:45
responsibility on that then and largely we look at the
2:36:46
we look at the
2:36:46
we look at the partnership of ai it is those
2:36:48
partnership of ai it is those
2:36:48
partnership of ai it is those organizations saying we're going to pay
2:36:50
organizations saying we're going to pay
2:36:50
organizations saying we're going to pay attention
2:36:51
attention
2:36:51
attention to how these things are used and who's
2:36:53
to how these things are used and who's
2:36:53
to how these things are used and who's and how they're being built
2:36:55
and how they're being built
2:36:55
and how they're being built i don't know that it's a good solution
2:36:57
i don't know that it's a good solution
2:36:57
i don't know that it's a good solution but it is
2:36:58
but it is
2:36:58
but it is you know the current gatekeeper are the
2:37:00
you know the current gatekeeper are the
2:37:00
you know the current gatekeeper are the cloud providers
2:37:02
cloud providers
2:37:02
cloud providers honestly like just a just a simple you
2:37:05
honestly like just a just a simple you
2:37:05
honestly like just a just a simple you know how sometimes
2:37:06
know how sometimes
2:37:06
know how sometimes in california i lived in california for
2:37:08
in california i lived in california for
2:37:08
in california i lived in california for a number of years and everything
2:37:09
a number of years and everything
2:37:09
a number of years and everything according to the state of california
2:37:10
according to the state of california
2:37:10
according to the state of california may cause cancer and there's a there's
2:37:12
may cause cancer and there's a there's
2:37:12
may cause cancer and there's a there's like a label on it like
2:37:14
like a label on it like
2:37:14
like a label on it like on everything this is this is known to
2:37:16
on everything this is this is known to
2:37:16
on everything this is this is known to the state of california to cause
2:37:17
the state of california to cause
2:37:18
the state of california to cause cancer and you're like okay like
2:37:21
cancer and you're like okay like
2:37:21
cancer and you're like okay like just a label that says this decision
2:37:25
just a label that says this decision
2:37:25
just a label that says this decision that the computer is making is
2:37:27
that the computer is making is
2:37:27
that the computer is making is influenced by automatic
2:37:29
influenced by automatic
2:37:29
influenced by automatic models like that alone is helpful
2:37:32
models like that alone is helpful
2:37:32
models like that alone is helpful right just disclosing the fact that
2:37:34
right just disclosing the fact that
2:37:34
right just disclosing the fact that they're using automatic methods to
2:37:36
they're using automatic methods to
2:37:36
they're using automatic methods to make decisions i mean that label helps a
2:37:39
make decisions i mean that label helps a
2:37:39
make decisions i mean that label helps a lot because then you're like oh well why
2:37:41
lot because then you're like oh well why
2:37:41
lot because then you're like oh well why did it reject me
2:37:42
did it reject me
2:37:42
did it reject me and then you can ask your questions
2:37:43
and then you can ask your questions
2:37:44
and then you can ask your questions right i mean that's just to me like a
2:37:45
right i mean that's just to me like a
2:37:45
right i mean that's just to me like a simple
2:37:46
simple
2:37:46
simple like tiny stop sign or maybe a speed
2:37:48
like tiny stop sign or maybe a speed
2:37:48
like tiny stop sign or maybe a speed bump
2:37:49
bump
2:37:49
bump what are your thoughts yeah it's the
2:37:51
what are your thoughts yeah it's the
2:37:51
what are your thoughts yeah it's the argument in favor of
2:37:53
argument in favor of
2:37:53
argument in favor of against the computer says right is to
2:37:56
against the computer says right is to
2:37:56
against the computer says right is to say did you read the label the label
2:37:57
say did you read the label the label
2:37:57
say did you read the label the label says
2:37:58
says
2:37:58
says the computer is only amplifying our own
2:38:00
the computer is only amplifying our own
2:38:00
the computer is only amplifying our own data sets
2:38:02
data sets
2:38:02
data sets uh marion your thoughts yeah yeah i
2:38:05
uh marion your thoughts yeah yeah i
2:38:05
uh marion your thoughts yeah yeah i fully agree because i mean even if you
2:38:06
fully agree because i mean even if you
2:38:06
fully agree because i mean even if you take just a simple
2:38:07
take just a simple
2:38:07
take just a simple decision for you i mean they're normally
2:38:09
decision for you i mean they're normally
2:38:09
decision for you i mean they're normally very successful right but
2:38:11
very successful right but
2:38:11
very successful right but if they're wrong you can be really wrong
2:38:14
if they're wrong you can be really wrong
2:38:14
if they're wrong you can be really wrong and
2:38:14
and
2:38:14
and um it's not without a risk and let's not
2:38:17
um it's not without a risk and let's not
2:38:17
um it's not without a risk and let's not talk about neural networks which are
2:38:19
talk about neural networks which are
2:38:19
talk about neural networks which are way harder to explain but um i
2:38:22
way harder to explain but um i
2:38:22
way harder to explain but um i see and practice a lot of people that
2:38:24
see and practice a lot of people that
2:38:24
see and practice a lot of people that are using let's say
2:38:26
are using let's say
2:38:26
are using let's say any alteration on a decision tree
2:38:29
any alteration on a decision tree
2:38:29
any alteration on a decision tree not knowing that they can be wrong
2:38:33
not knowing that they can be wrong
2:38:33
not knowing that they can be wrong right and then the computer says no but
2:38:34
right and then the computer says no but
2:38:34
right and then the computer says no but maybe it should be yes so
2:38:37
maybe it should be yes so
2:38:37
maybe it should be yes so um yeah there is a risk it now it's a
2:38:40
um yeah there is a risk it now it's a
2:38:40
um yeah there is a risk it now it's a probability right
2:38:41
probability right
2:38:41
probability right we have to i think we have to make clear
2:38:43
we have to i think we have to make clear
2:38:43
we have to i think we have to make clear to people that it's not a yes or no even
2:38:45
to people that it's not a yes or no even
2:38:45
to people that it's not a yes or no even if we convert it to yes or no but we're
2:38:47
if we convert it to yes or no but we're
2:38:47
if we convert it to yes or no but we're talking about probabilities
2:38:50
talking about probabilities
2:38:50
talking about probabilities richard and i i like the way you put it
2:38:52
richard and i i like the way you put it
2:38:52
richard and i i like the way you put it though like adding a disclaimer in
2:38:53
though like adding a disclaimer in
2:38:53
though like adding a disclaimer in really small
2:38:54
really small
2:38:54
really small print is not really a good idea either
2:38:57
print is not really a good idea either
2:38:57
print is not really a good idea either uh i guess i'm just trying to in my head
2:39:00
uh i guess i'm just trying to in my head
2:39:00
uh i guess i'm just trying to in my head figure out what's the license look like
2:39:02
figure out what's the license look like
2:39:02
figure out what's the license look like what's the stop sign what's the stop
2:39:03
what's the stop sign what's the stop
2:39:03
what's the stop sign what's the stop lights you know well i i don't want to
2:39:05
lights you know well i i don't want to
2:39:05
lights you know well i i don't want to fight you on that because i
2:39:06
fight you on that because i
2:39:06
fight you on that because i what i want is and no one to be
2:39:10
what i want is and no one to be
2:39:10
what i want is and no one to be accepting of the response well the
2:39:11
accepting of the response well the
2:39:12
accepting of the response well the computer says
2:39:13
computer says
2:39:13
computer says yes right because that's a terrible
2:39:16
yes right because that's a terrible
2:39:16
yes right because that's a terrible answer and it should be
2:39:17
answer and it should be
2:39:17
answer and it should be immediately challenged i kind of want a
2:39:19
immediately challenged i kind of want a
2:39:19
immediately challenged i kind of want a little business card that says
2:39:20
little business card that says
2:39:20
little business card that says computers are often wrong right it's
2:39:23
computers are often wrong right it's
2:39:23
computers are often wrong right it's just
2:39:24
just
2:39:24
just you want to press back against that
2:39:25
you want to press back against that
2:39:26
you want to press back against that because they are only ultimately tools
2:39:27
because they are only ultimately tools
2:39:28
because they are only ultimately tools that we manifest in the first place
2:39:33
you know our industry
2:39:37
you know our industry
2:39:37
you know our industry is still not a profession the computing
2:39:40
is still not a profession the computing
2:39:40
is still not a profession the computing industry
2:39:41
industry
2:39:41
industry right we don't have a professional
2:39:43
right we don't have a professional
2:39:43
right we don't have a professional certification
2:39:44
certification
2:39:44
certification we don't have a group that is ultimately
2:39:47
we don't have a group that is ultimately
2:39:47
we don't have a group that is ultimately responsible for when its members make
2:39:49
responsible for when its members make
2:39:49
responsible for when its members make mistakes
2:39:49
mistakes
2:39:49
mistakes you know it's the difference between a
2:39:51
you know it's the difference between a
2:39:51
you know it's the difference between a software engineer
2:39:52
software engineer
2:39:52
software engineer and an electrical engineer or a civil
2:39:55
and an electrical engineer or a civil
2:39:55
and an electrical engineer or a civil engineer or a doctor
2:39:56
engineer or a doctor
2:39:56
engineer or a doctor or a lawyer that they're always backed
2:39:59
or a lawyer that they're always backed
2:39:59
or a lawyer that they're always backed by a group that
2:40:01
by a group that
2:40:01
by a group that certifies them and takes responsibility
2:40:03
certifies them and takes responsibility
2:40:03
certifies them and takes responsibility for them right and we've largely been
2:40:06
for them right and we've largely been
2:40:06
for them right and we've largely been isolated from that
2:40:07
isolated from that
2:40:07
isolated from that early on i worked at a at a place uh
2:40:10
early on i worked at a at a place uh
2:40:10
early on i worked at a at a place uh that
2:40:11
that
2:40:11
that that i was a the software guy there at a
2:40:14
that i was a the software guy there at a
2:40:14
that i was a the software guy there at a place that
2:40:14
place that
2:40:14
place that served actual civil engineers and i
2:40:17
served actual civil engineers and i
2:40:17
served actual civil engineers and i remember one time
2:40:18
remember one time
2:40:18
remember one time one of their engineers came in and i was
2:40:21
one of their engineers came in and i was
2:40:21
one of their engineers came in and i was like oh yeah i'm a software engineer
2:40:22
like oh yeah i'm a software engineer
2:40:22
like oh yeah i'm a software engineer he's like are you
2:40:23
he's like are you
2:40:24
he's like are you where did you get your certification
2:40:25
where did you get your certification
2:40:25
where did you get your certification from uh where's your license
2:40:28
from uh where's your license
2:40:28
from uh where's your license and i was like oh and and i obviously i
2:40:31
and i was like oh and and i obviously i
2:40:31
and i was like oh and and i obviously i was
2:40:31
was
2:40:31
was i was a young guy still in college then
2:40:34
i was a young guy still in college then
2:40:34
i was a young guy still in college then and i took offense to that but
2:40:36
and i took offense to that but
2:40:36
and i took offense to that but many years later i came to the
2:40:38
many years later i came to the
2:40:38
many years later i came to the realization that
2:40:39
realization that
2:40:39
realization that if a bridge breaks and someone gets hurt
2:40:43
if a bridge breaks and someone gets hurt
2:40:43
if a bridge breaks and someone gets hurt there is someone that's accountable yes
2:40:45
there is someone that's accountable yes
2:40:45
there is someone that's accountable yes not only the person who built it
2:40:47
not only the person who built it
2:40:47
not only the person who built it but the state organization that
2:40:48
but the state organization that
2:40:48
but the state organization that certified it yeah and the authority and
2:40:50
certified it yeah and the authority and
2:40:50
certified it yeah and the authority and the certifying body
2:40:52
the certifying body
2:40:52
the certifying body you know that's the nature of it then
2:40:54
you know that's the nature of it then
2:40:54
you know that's the nature of it then you understand
2:40:55
you understand
2:40:55
you understand software be through the end user license
2:40:58
software be through the end user license
2:40:58
software be through the end user license agreement
2:41:00
agreement
2:41:00
agreement has obviated that responsibility and for
2:41:03
has obviated that responsibility and for
2:41:03
has obviated that responsibility and for a good reason it
2:41:06
a good reason it
2:41:06
a good reason it allowed for rapid innovation
2:41:09
allowed for rapid innovation
2:41:09
allowed for rapid innovation right and you know we didn't start out
2:41:11
right and you know we didn't start out
2:41:11
right and you know we didn't start out with rules for buildings
2:41:13
with rules for buildings
2:41:13
with rules for buildings we just got to a point in sophistication
2:41:15
we just got to a point in sophistication
2:41:15
we just got to a point in sophistication of buildings where falling down had
2:41:17
of buildings where falling down had
2:41:17
of buildings where falling down had serious consequences like we should have
2:41:19
serious consequences like we should have
2:41:19
serious consequences like we should have rules
2:41:20
rules
2:41:20
rules we are now reaching that point in
2:41:22
we are now reaching that point in
2:41:22
we are now reaching that point in software like there's a question of
2:41:24
software like there's a question of
2:41:24
software like there's a question of when do we become a profession in
2:41:27
when do we become a profession in
2:41:27
when do we become a profession in exchange for
2:41:28
exchange for
2:41:28
exchange for decelerating innovation to increase
2:41:32
decelerating innovation to increase
2:41:32
decelerating innovation to increase reliability
2:41:34
reliability
2:41:34
reliability right that's the trade we're going to
2:41:35
right that's the trade we're going to
2:41:35
right that's the trade we're going to make the aerospace engineering
2:41:38
make the aerospace engineering
2:41:38
make the aerospace engineering engineering field still innovates slowly
2:41:41
engineering field still innovates slowly
2:41:42
engineering field still innovates slowly carefully because the consequences of
2:41:44
carefully because the consequences of
2:41:44
carefully because the consequences of being wrong are serious
2:41:46
being wrong are serious
2:41:46
being wrong are serious software is running the world these
2:41:49
software is running the world these
2:41:49
software is running the world these models
2:41:49
models
2:41:50
models can shape cultures
2:41:53
can shape cultures
2:41:53
can shape cultures we may need to be taking more
2:41:54
we may need to be taking more
2:41:54
we may need to be taking more responsibility for that and that means
2:41:56
responsibility for that and that means
2:41:56
responsibility for that and that means changing that rule exchanging rate of
2:41:58
changing that rule exchanging rate of
2:41:58
changing that rule exchanging rate of innovation
2:41:59
innovation
2:41:59
innovation for safety and that's an interesting
2:42:03
for safety and that's an interesting
2:42:03
for safety and that's an interesting that's an interesting thought process i
2:42:05
that's an interesting thought process i
2:42:05
that's an interesting thought process i mean is it something we should be doing
2:42:06
mean is it something we should be doing
2:42:06
mean is it something we should be doing in ai as well
2:42:08
in ai as well
2:42:08
in ai as well well i think ai has brought it to the
2:42:10
well i think ai has brought it to the
2:42:10
well i think ai has brought it to the light right we're seeing the ultimate
2:42:12
light right we're seeing the ultimate
2:42:12
light right we're seeing the ultimate manifestation
2:42:13
manifestation
2:42:13
manifestation of computing so far right there are more
2:42:16
of computing so far right there are more
2:42:16
of computing so far right there are more manifestations coming
2:42:18
manifestations coming
2:42:18
manifestations coming when you start seeing the potential of
2:42:19
when you start seeing the potential of
2:42:19
when you start seeing the potential of quantum computing around its changes to
2:42:21
quantum computing around its changes to
2:42:21
quantum computing around its changes to the way we'll approach chemistry
2:42:23
the way we'll approach chemistry
2:42:23
the way we'll approach chemistry and and different physics models and so
2:42:25
and and different physics models and so
2:42:25
and and different physics models and so forth that also may be a new class of
2:42:28
forth that also may be a new class of
2:42:28
forth that also may be a new class of of concern as we get better at this as
2:42:32
of concern as we get better at this as
2:42:32
of concern as we get better at this as the same way that when the car
2:42:33
the same way that when the car
2:42:33
the same way that when the car got faster the rules became more
2:42:35
got faster the rules became more
2:42:36
got faster the rules became more important
2:42:36
important
2:42:36
important the potential computing is growing the
2:42:39
the potential computing is growing the
2:42:39
the potential computing is growing the rules are becoming more important
2:42:43
yeah i mean there's certain systems that
2:42:46
yeah i mean there's certain systems that
2:42:46
yeah i mean there's certain systems that literally could be like a bridge falling
2:42:47
literally could be like a bridge falling
2:42:47
literally could be like a bridge falling down effectively sure
2:42:49
down effectively sure
2:42:50
down effectively sure right you think about the efforts we're
2:42:52
right you think about the efforts we're
2:42:52
right you think about the efforts we're going to to
2:42:53
going to to
2:42:53
going to to to well you know our water supplies are
2:42:56
to well you know our water supplies are
2:42:56
to well you know our water supplies are i have technology in them now
2:42:57
i have technology in them now
2:42:57
i have technology in them now right they're not just bumps and valves
2:42:59
right they're not just bumps and valves
2:42:59
right they're not just bumps and valves their computers there their computers on
2:43:01
their computers there their computers on
2:43:01
their computers there their computers on every one of our power plants
2:43:02
every one of our power plants
2:43:02
every one of our power plants you know i'm doing these shows over in
2:43:05
you know i'm doing these shows over in
2:43:05
you know i'm doing these shows over in the run as land and dot and rock land
2:43:07
the run as land and dot and rock land
2:43:07
the run as land and dot and rock land talking to folks that are
2:43:08
talking to folks that are
2:43:08
talking to folks that are talking about how to scada systems get
2:43:10
talking about how to scada systems get
2:43:10
talking about how to scada systems get integrated under the internet and the
2:43:11
integrated under the internet and the
2:43:11
integrated under the internet and the security consequences that represents
2:43:13
security consequences that represents
2:43:14
security consequences that represents you know the reality is that off an
2:43:17
you know the reality is that off an
2:43:17
you know the reality is that off an awful lot of our civilization depends on
2:43:18
awful lot of our civilization depends on
2:43:18
awful lot of our civilization depends on an awful lot of compute
2:43:19
an awful lot of compute
2:43:19
an awful lot of compute that needs to work and so
2:43:23
that needs to work and so
2:43:23
that needs to work and so we are in a transitory period now we i
2:43:25
we are in a transitory period now we i
2:43:26
we are in a transitory period now we i suspect
2:43:26
suspect
2:43:26
suspect we are the generation that will see us
2:43:28
we are the generation that will see us
2:43:28
we are the generation that will see us become a profession
2:43:30
become a profession
2:43:30
become a profession interesting and like when you say that
2:43:32
interesting and like when you say that
2:43:32
interesting and like when you say that do you mean like in the next 20 years
2:43:34
do you mean like in the next 20 years
2:43:34
do you mean like in the next 20 years kind of thing
2:43:34
kind of thing
2:43:34
kind of thing the next five years okay i mean it's
2:43:37
the next five years okay i mean it's
2:43:37
the next five years okay i mean it's hard to predict exactly how quickly
2:43:39
hard to predict exactly how quickly
2:43:39
hard to predict exactly how quickly there's a bunch of different models of
2:43:40
there's a bunch of different models of
2:43:40
there's a bunch of different models of how to go about this and that's a whole
2:43:41
how to go about this and that's a whole
2:43:41
how to go about this and that's a whole other show
2:43:42
other show
2:43:42
other show but i think that artificial intelligence
2:43:45
but i think that artificial intelligence
2:43:45
but i think that artificial intelligence is just
2:43:46
is just
2:43:46
is just this mechanism now that's showing the
2:43:49
this mechanism now that's showing the
2:43:49
this mechanism now that's showing the power of compute
2:43:51
power of compute
2:43:51
power of compute and some of it unfairly right we get
2:43:52
and some of it unfairly right we get
2:43:52
and some of it unfairly right we get into the anthropomorphic things
2:43:54
into the anthropomorphic things
2:43:54
into the anthropomorphic things the terminators all that silliness but
2:43:56
the terminators all that silliness but
2:43:56
the terminators all that silliness but it's also
2:43:57
it's also
2:43:57
it's also useful for the public's fear
2:44:00
useful for the public's fear
2:44:00
useful for the public's fear to say how are we licensing this how are
2:44:03
to say how are we licensing this how are
2:44:03
to say how are we licensing this how are we
2:44:03
we
2:44:04
we controlling this it's always actually
2:44:06
controlling this it's always actually
2:44:06
controlling this it's always actually been a concern
2:44:07
been a concern
2:44:07
been a concern it's just becoming more acute
2:44:10
it's just becoming more acute
2:44:10
it's just becoming more acute interesting
2:44:11
interesting
2:44:11
interesting marines you build stuff because uh like
2:44:14
marines you build stuff because uh like
2:44:14
marines you build stuff because uh like clearly you're you build stuff i'm a
2:44:16
clearly you're you build stuff i'm a
2:44:16
clearly you're you build stuff i'm a demo where
2:44:17
demo where
2:44:17
demo where person and i have been for many years uh
2:44:20
person and i have been for many years uh
2:44:20
person and i have been for many years uh do you think about these kinds of things
2:44:22
do you think about these kinds of things
2:44:22
do you think about these kinds of things as you build stuff with you and your yes
2:44:24
as you build stuff with you and your yes
2:44:24
as you build stuff with you and your yes yes yes as mentioned before uh maybe
2:44:26
yes yes as mentioned before uh maybe
2:44:26
yes yes as mentioned before uh maybe just in a very simple form but
2:44:28
just in a very simple form but
2:44:28
just in a very simple form but um yeah to write down everything you
2:44:31
um yeah to write down everything you
2:44:31
um yeah to write down everything you make a decision upon
2:44:33
make a decision upon
2:44:33
make a decision upon so so somebody else can check that with
2:44:36
so so somebody else can check that with
2:44:36
so so somebody else can check that with you can go through it
2:44:37
you can go through it
2:44:37
you can go through it also for yourself i mean as you develop
2:44:39
also for yourself i mean as you develop
2:44:39
also for yourself i mean as you develop as a person you can look back and say
2:44:41
as a person you can look back and say
2:44:41
as a person you can look back and say hey i made this decision and
2:44:43
hey i made this decision and
2:44:43
hey i made this decision and contributing to to bias basically
2:44:46
contributing to to bias basically
2:44:46
contributing to to bias basically so yes i am aware of it i tried to
2:44:49
so yes i am aware of it i tried to
2:44:49
so yes i am aware of it i tried to minimize it to at least write down my
2:44:51
minimize it to at least write down my
2:44:51
minimize it to at least write down my decision so
2:44:52
decision so
2:44:52
decision so somebody else at least can see what my
2:44:53
somebody else at least can see what my
2:44:54
somebody else at least can see what my intent was and also
2:44:56
intent was and also
2:44:56
intent was and also i i i normally put a disclaimer in my
2:44:59
i i i normally put a disclaimer in my
2:44:59
i i i normally put a disclaimer in my code
2:44:59
code
2:44:59
code like that my intention for a model
2:45:02
like that my intention for a model
2:45:02
like that my intention for a model because again
2:45:03
because again
2:45:03
because again if you build a model to for example um
2:45:07
if you build a model to for example um
2:45:07
if you build a model to for example um find out which student needs some help
2:45:09
find out which student needs some help
2:45:09
find out which student needs some help right you can
2:45:11
right you can
2:45:11
right you can revert it and say hey i'm going to
2:45:12
revert it and say hey i'm going to
2:45:12
revert it and say hey i'm going to exclude this student
2:45:14
exclude this student
2:45:14
exclude this student so by including my intent in the code
2:45:17
so by including my intent in the code
2:45:17
so by including my intent in the code it's like yeah this is the attention of
2:45:18
it's like yeah this is the attention of
2:45:18
it's like yeah this is the attention of the model so
2:45:20
the model so
2:45:20
the model so please use it rightfully um
2:45:23
please use it rightfully um
2:45:23
please use it rightfully um i don't know we don't have specific
2:45:25
i don't know we don't have specific
2:45:25
i don't know we don't have specific rules yet for that i think i've
2:45:27
rules yet for that i think i've
2:45:27
rules yet for that i think i've completely agree with richard we don't
2:45:28
completely agree with richard we don't
2:45:28
completely agree with richard we don't have the the licenses we don't have the
2:45:30
have the the licenses we don't have the
2:45:30
have the the licenses we don't have the controls but
2:45:31
controls but
2:45:31
controls but at least i think we we have to be you
2:45:34
at least i think we we have to be you
2:45:34
at least i think we we have to be you know we have to maybe make them and
2:45:36
know we have to maybe make them and
2:45:36
know we have to maybe make them and try to make them and maybe future
2:45:39
try to make them and maybe future
2:45:39
try to make them and maybe future generations will build upon that
2:45:40
generations will build upon that
2:45:40
generations will build upon that but but you have to be aware of as a
2:45:42
but but you have to be aware of as a
2:45:42
but but you have to be aware of as a developer i think
2:45:44
developer i think
2:45:44
developer i think you mentioned richard i think you always
2:45:46
you mentioned richard i think you always
2:45:46
you mentioned richard i think you always have to take your responsibility
2:45:48
have to take your responsibility
2:45:48
have to take your responsibility um how far that goes i mean i don't know
2:45:51
um how far that goes i mean i don't know
2:45:51
um how far that goes i mean i don't know because if somebody takes my models and
2:45:53
because if somebody takes my models and
2:45:53
because if somebody takes my models and implement it in a way and
2:45:55
implement it in a way and
2:45:55
implement it in a way and abusing it that's hard i mean uh but but
2:45:58
abusing it that's hard i mean uh but but
2:45:58
abusing it that's hard i mean uh but but in in theory you can do a lot of damage
2:46:00
in in theory you can do a lot of damage
2:46:00
in in theory you can do a lot of damage you can do a lot of good but
2:46:02
you can do a lot of good but
2:46:02
you can do a lot of good but the same way you can you can damage a
2:46:04
the same way you can you can damage a
2:46:04
the same way you can you can damage a lot so not only to prote let's say as a
2:46:07
lot so not only to prote let's say as a
2:46:07
lot so not only to prote let's say as a driver you also put your seatbelt right
2:46:09
driver you also put your seatbelt right
2:46:09
driver you also put your seatbelt right so you want to protect yourself too
2:46:11
so you want to protect yourself too
2:46:12
so you want to protect yourself too um but also at least to make sure what
2:46:14
um but also at least to make sure what
2:46:14
um but also at least to make sure what your intent was
2:46:15
your intent was
2:46:15
your intent was and the decisions you made so before i
2:46:18
and the decisions you made so before i
2:46:18
and the decisions you made so before i go to william one last question cause i
2:46:19
go to william one last question cause i
2:46:19
go to william one last question cause i know i've been hogging you all let's
2:46:21
know i've been hogging you all let's
2:46:21
know i've been hogging you all let's start with you
2:46:21
start with you
2:46:21
start with you uh marian what what's the future of ai
2:46:24
uh marian what what's the future of ai
2:46:24
uh marian what what's the future of ai and then we'll go to you richard on that
2:46:27
well what i hope i mean that's i don't
2:46:30
well what i hope i mean that's i don't
2:46:30
well what i hope i mean that's i don't know what a future is because i'm not a
2:46:31
know what a future is because i'm not a
2:46:32
know what a future is because i'm not a visionary but i hope people will see it
2:46:34
visionary but i hope people will see it
2:46:34
visionary but i hope people will see it as a tool
2:46:35
as a tool
2:46:35
as a tool and not forget that they're human
2:46:39
and not forget that they're human
2:46:39
and not forget that they're human so um yeah it's it's great technologies
2:46:42
so um yeah it's it's great technologies
2:46:42
so um yeah it's it's great technologies you can do a lot of things but please
2:46:45
you can do a lot of things but please
2:46:45
you can do a lot of things but please don't forget that you're a human being
2:46:47
don't forget that you're a human being
2:46:47
don't forget that you're a human being and you can use technology to do better
2:46:50
and you can use technology to do better
2:46:50
and you can use technology to do better things but
2:46:52
things but
2:46:52
things but again don't just listen to a computer
2:46:56
i can't argue with anything you say
2:46:58
i can't argue with anything you say
2:46:58
i can't argue with anything you say there i am it's
2:46:59
there i am it's
2:46:59
there i am it's it's these are just tools and we've
2:47:01
it's these are just tools and we've
2:47:01
it's these are just tools and we've created a new set of tools
2:47:03
created a new set of tools
2:47:03
created a new set of tools that are empowering people to do
2:49:51
thank you so much
2:49:56
okay we're back thank you very much all
2:49:58
okay we're back thank you very much all
2:49:58
okay we're back thank you very much all of you for joining us
2:49:59
of you for joining us
2:49:59
of you for joining us uh thank you rich for the awesome
2:50:01
uh thank you rich for the awesome
2:50:01
uh thank you rich for the awesome session and uh depend on this question
2:50:03
session and uh depend on this question
2:50:03
session and uh depend on this question afterwards there's a lot more stuff to
2:50:05
afterwards there's a lot more stuff to
2:50:05
afterwards there's a lot more stuff to talk about i'm sure
2:50:06
talk about i'm sure
2:50:06
talk about i'm sure uh but we're out of time and with that i
2:50:09
uh but we're out of time and with that i
2:50:09
uh but we're out of time and with that i would like to
2:50:10
would like to
2:50:10
would like to announce the winner of the fifty dollar
2:50:12
announce the winner of the fifty dollar
2:50:12
announce the winner of the fifty dollar gift card for amazon
2:50:13
gift card for amazon
2:50:13
gift card for amazon uh you're going to get a message from
2:50:16
uh you're going to get a message from
2:50:16
uh you're going to get a message from from us
2:50:17
from us
2:50:17
from us because you you have an awesome tweet
2:50:19
because you you have an awesome tweet
2:50:19
because you you have an awesome tweet there's no doubt about that
2:50:21
there's no doubt about that
2:50:21
there's no doubt about that we will should make sure that we retweet
2:50:23
we will should make sure that we retweet
2:50:23
we will should make sure that we retweet that and make sure that
2:50:25
that and make sure that
2:50:25
that and make sure that it's going on my wall i think that it's
2:50:27
it's going on my wall i think that it's
2:50:27
it's going on my wall i think that it's that good
2:50:28
that good
2:50:28
that good so thank you very much everyone for
2:50:29
so thank you very much everyone for
2:50:29
so thank you very much everyone for joining us and next week
2:50:31
joining us and next week
2:50:31
joining us and next week we are back on first day again same time
2:50:34
we are back on first day again same time
2:50:34
we are back on first day again same time same place uh hopefully
2:50:36
same place uh hopefully
2:50:36
same place uh hopefully with a better internet connection and
2:50:38
with a better internet connection and
2:50:38
with a better internet connection and better microphones and everything
2:50:40
better microphones and everything
2:50:40
better microphones and everything um and then we're going to talk about
2:50:41
um and then we're going to talk about
2:50:41
um and then we're going to talk about computer vision uh we've got
2:50:43
computer vision uh we've got
2:50:44
computer vision uh we've got awesome people joining us so we've got
2:50:46
awesome people joining us so we've got
2:50:46
awesome people joining us so we've got someone from
2:50:47
someone from
2:50:47
someone from opencv no less the ceo of opencv
2:50:50
opencv no less the ceo of opencv
2:50:50
opencv no less the ceo of opencv uh sacha malik who was going to talk to
2:50:52
uh sacha malik who was going to talk to
2:50:52
uh sacha malik who was going to talk to us about their new ai innovations
2:50:55
us about their new ai innovations
2:50:55
us about their new ai innovations uh actually a camera that prevented
2:50:57
uh actually a camera that prevented
2:50:57
uh actually a camera that prevented burning models so that's pretty cool i
2:50:59
burning models so that's pretty cool i
2:50:59
burning models so that's pretty cool i think
2:50:59
think
2:50:59
think um and we have uh elizabeth from
2:51:03
um and we have uh elizabeth from
2:51:03
um and we have uh elizabeth from cisco who's going to talk about micro
2:51:05
cisco who's going to talk about micro
2:51:05
cisco who's going to talk about micro interactions
2:51:06
interactions
2:51:06
interactions and what that means for building ai
2:51:08
and what that means for building ai
2:51:08
and what that means for building ai models in computer vision it all sounds
2:51:10
models in computer vision it all sounds
2:51:10
models in computer vision it all sounds very very cool so
2:51:12
very very cool so
2:51:12
very very cool so we'll have lots more demos to show a lot
2:51:14
we'll have lots more demos to show a lot
2:51:14
we'll have lots more demos to show a lot more stuff to
2:51:15
more stuff to
2:51:15
more stuff to go through i mean thank you very much
2:51:17
go through i mean thank you very much
2:51:17
go through i mean thank you very much have a nice evening or morning or
2:51:19
have a nice evening or morning or
2:51:19
have a nice evening or morning or afternoon or
2:51:20
afternoon or
2:51:20
afternoon or wherever you are seth do you have any
2:51:22
wherever you are seth do you have any
2:51:22
wherever you are seth do you have any final words for us
2:51:23
final words for us
2:51:23
final words for us any wisdom that you want to share with
2:51:26
any wisdom that you want to share with
2:51:26
any wisdom that you want to share with us yeah
2:51:27
us yeah
2:51:27
us yeah life is awesome go forth and make cool
2:51:30
life is awesome go forth and make cool
2:51:30
life is awesome go forth and make cool things
2:51:31
things
2:51:31
things cool cool thanks have a nice day
#Programming
#Computer Education
#Machine Learning & Artificial Intelligence