live_tv
Livestream Starting Soon
00
Hours
:
00
Minutes
:
00
Seconds
Up next in 10
We all interact with AI every day, but how much do we really know about its journey? Join us on a fascinating trip through the past, present, and future of Artificial Intelligence.
In this episode, we'll peel back the layers on some of AI's biggest milestones—from the early days of "boosting" algorithms to the massive data sets that made modern AI possible. But this isn't just a history lesson. We’ll also tackle the big questions: Is AI becoming too biased? Can a machine ever truly be "conscious"? And as AI becomes more and more a part of our lives, what are the rules we need to be following?
Tune in to explore the incredible progression of AI technology and the urgent ethical conversations that are happening right now.
#AIhistory #AIEthics #TechPodcast
See more:
Propositional Logic in Artificial Intelligence: Syntax, Semantics & Applications
https://youtu.be/omh6ldFGhAI
Show More Show Less View Video Transcript
0:00
Okay, let's unpack this. Welcome to the
0:02
deep dive. We're your shortcut to
0:04
understanding the big topics shaping our
0:06
world today. It's all about artificial
0:09
intelligence. We're going on a journey
0:11
really from these ancient dreams of
0:12
artificial life right up to the
0:14
incredibly powerful AI we see now. And
0:17
then importantly, we're tackling the
0:19
tough ethical questions that come along
0:20
with it. Our mission to get you beyond
0:23
the headlines, give you a solid
0:26
digestible grasp of AI's past, its
0:28
present boom, and the real societal
0:30
shifts we're already dealing with. You
0:32
might be surprised how it all connects.
0:33
It's absolutely true. AI is changing
0:35
things fast. How we live, how we work,
0:37
and figuring out where it came from and
0:39
these ethical knots. It's not just for
0:41
the tech folks anymore. It's something
0:42
everyone needs to grapple with. We'll
0:44
connect those early, sometimes abstract
0:46
concepts directly to the big debates
0:48
happening right now. So, where does the
0:50
story really kick off? I mean, the idea
0:52
of making artificial beings, it's
0:53
ancient, right? You think about myths
0:55
like Taos, that bronze giant, or the
0:57
golem. People have always been
0:58
fascinated by creating intelligence,
1:01
animating the inanimate.
1:02
That fascination is deep-seated.
1:04
But for the science part, we really need
1:06
to jump to the mid 20th century. You've
1:08
got Donald Heb in the 1940s. He came up
1:10
with this really key idea about how
1:12
brain neurons learn. And he said uh when
1:15
one cell repeatedly assists in firing
1:17
another. Basically describing how
1:19
connections get stronger or weaker.
1:21
That's the core idea behind weights in
1:24
neural networks today, isn't it? How the
1:26
system learns and remembers.
1:27
Exactly. Then 1950 Alan Turing drops his
1:30
huge paper computing machinery and
1:33
intelligence and in it the Turing test.
1:36
Such a simple yet profound concept. If a
1:40
machine can talk to you and you can't
1:41
tell it's not human, well, is it
1:43
thinking?
1:44
It really shifted the whole
1:45
philosophical debate about machine
1:46
intelligence.
1:47
At the same time, you had Arthur Samuel
1:48
in 52 coining machine learning with his
1:50
checkers program. It learned by
1:52
remembering past moves, wrote learning
1:55
he called it.
1:56
And what's crucial here is seeing how
1:57
these theoretical sparks, Heb's neurons,
2:00
Turing's test, Samuels learning, they
2:02
weren't just thought experiments. They
2:04
were building the foundation. It was
2:05
this massive conceptual shift, wasn't
2:07
it? from machines that just calculate to
2:09
machines that might actually learn.
2:11
It really was. And then the person who
2:13
kind of pulled these threads together
2:15
and gave it a name was John McCarthy.
2:17
1956 the Dartmouth workshop. He proposed
2:21
the term artificial intelligence AI
2:23
partly to distinguish it from uh
2:25
cybernetics. That workshop is basically
2:28
seen as the birth of AI as its own
2:30
academic field.
2:31
And McCarthy didn't stop there. lisp
2:34
garbage collection early cloud computing
2:36
ideas a real visionary
2:38
he really was his work shaped how we
2:41
even approached building these systems
2:44
okay so the field is born you'd think
2:46
maybe it was smooth sailing from there
2:47
not quite the first phase say 56 to 74
2:50
the optimism was just well astonishing
2:52
you had programs solving algebra proving
2:55
theorems learning basic English
2:56
there was this real feeling that human
2:57
level AI was maybe just a decade or two
3:00
away people like Simon and New were
3:01
making those kinds of predictions
3:02
yeah and things like Frank Rosenblat's
3:04
Mark1 Perceptron in 57 for image
3:06
recognition. That felt like concrete
3:07
progress.
3:08
But that huge optimism, it kind of
3:10
backfired, didn't it? It set
3:12
expectations incredibly high. And when
3:15
those massive breakthroughs didn't
3:16
happen quite on schedule,
3:18
the funding, the enthusiasm,
3:21
it started to dry up. It raises that
3:23
classic question about hype cycles in
3:25
tech, right?
3:26
How expectations drive the whole
3:27
research path.
3:28
Absolutely. And that led straight into
3:30
the first AI winter roughly 1974 to
3:34
1980. A big trigger was the book
3:36
perceptrons in ' 69 by Minsky and paper.
3:39
They showed some pretty serious
3:40
limitations of those early neural
3:42
networks.
3:42
Yeah, that book really put the brakes on
3:44
neural net research for quite a while.
3:46
And there are other big hurdles too.
3:47
Just raw computing power was a major
3:49
limitation. I read about one natural
3:52
language program that could only handle
3:53
like 20 words because of memory limit.
3:55
Wow. Just 20 words. Then you had the
3:57
cominatorial explosion problems just
3:59
getting too complex, too many
4:00
possibilities for early machines to
4:02
handle. And that weird thing, Morovx
4:05
paradox, AI was getting good at chess.
4:07
You know, smart stuff, but terrible at
4:09
basic things like recognizing a face or
4:12
walking,
4:12
right? The common sense problem. A
4:15
incredibly difficult to program what
4:18
humans just know.
4:20
So with all these roadblocks, the
4:21
funding got cut. Big reports like Alpac
4:24
in the US and Lighill in the UK were
4:26
pretty critical. Governments pulled
4:28
back.
4:28
You know, it's worth looking closer at
4:30
that AI winter idea. It sounds like
4:32
everything stopped, but research often
4:34
just continued under different names.
4:36
People shifted focus, maybe called it
4:38
computational intelligence or something
4:40
else to get funding.
4:41
It wasn't necessarily a total freeze,
4:43
more like a a rebranding in a quiet
4:46
continuation of work that actually laid
4:48
groundwork for later.
4:49
That's a really important point. It
4:51
wasn't dormant, just evolving
4:53
differently. Which brings us to the next
4:55
phase. The 80s saw a bit of a comeback
4:57
with expert systems. These were AI
5:00
programs focused on specific tasks,
5:02
mimicking human experts. They actually
5:04
got used quite a bit in industry,
5:05
right? That was a boom time for a while.
5:07
For a while, yeah. But the hype got
5:09
ahead of reality again. The business
5:11
side of it collapsed. Big national
5:13
projects like Japan's Fifth Generation
5:16
didn't meet their uh grandiose
5:18
objectives. And boom, second AI winter
5:21
in the 1990s. AI companies closed. Even
5:24
the term AI became a bit toxic again.
5:26
Researchers had to find other labels.
5:28
Another cycle of boom and bust.
5:30
But then things really started to change
5:32
in the early 2000s. This set the stage
5:34
for the massive boom we're in now. Three
5:37
things really. First, big data.
5:40
Suddenly, the internet and digital
5:41
everything meant oceans of data to train
5:43
AI on.
5:44
Crucial.
5:45
Think ImageNet Fay lies project from
5:47
2009. They crowdsourced labels for 14
5:50
million images. It became this vital
5:51
benchmark or Google's word tovec in 2013
5:54
learning word meanings from just vast
5:55
text
5:56
the fuel for the new engines.
5:57
Second powerful hardware. Jeffrey
5:59
Hinton, one of the godfathers of deep
6:01
learning said back in the9s data sets
6:03
were too small and computers were
6:04
millions of times too slow. By the 2010s
6:07
that changed GPUs, graphics cards
6:09
originally for gaming turned out to be
6:11
perfect for the kind of math needed for
6:12
deep learning.
6:13
Suddenly the compute power was there.
6:15
And third, advanced machine learning
6:17
techniques, especially deep learning,
6:19
really taking off around 2012. This was
6:22
the return of neural networks, but much
6:24
more powerful now, building on earlier
6:26
ideas like back propagation and
6:28
convolutional nets, but ready for the
6:30
big data and fast hardware.
6:32
The pieces finally came together.
6:34
And here's where it gets really
6:35
interesting. The imageet challenge in
6:38
2012. This team, Krishvsky, Sutskver and
6:41
Hinton entered their system, AlexNet. It
6:45
used deep convolutional neural networks
6:47
and it didn't just win, it crushed the
6:49
competition. The error rate plummeted.
6:51
That was the watershed moment, wasn't
6:53
it? Alex Net proved deep learning worked
6:55
and worked dramatically better. Everyone
6:56
took notice.
6:57
Exactly. Deep learning went mainstream
6:59
almost overnight because of that.
7:00
Yeah. This connects to that idea of
7:02
hypernetics. How the hype, the language
7:04
we use, like switching from AI to
7:06
machine learning sometimes hides the
7:08
real steady progress happening
7:10
underneath the internet providing the
7:11
data, the better hardware. That was the
7:13
essential material shift that let deep
7:15
learning finally deliver.
7:17
Which brings us right up to the current
7:19
AI boom. Let's say from 2017 onwards. A
7:22
massive leap here was the transformer
7:24
architecture in 2017. This new design
7:27
allowed AI to handle sequences of text
7:29
like sentences, understanding context
7:32
much better. It's the core tech behind
7:34
large language models, LLMs,
7:36
the engine for things like Chat GPT,
7:38
right? And speaking of which, Chat GPT's
7:40
public release in November 2022. I mean,
7:43
wow. Fastest growing app ever. 100
7:45
million users in two months. Its ability
7:48
to chat, write code, generate creative
7:50
stuff. It put advanced AI in everyone's
7:53
hands and made it a global conversation
7:55
topic. unprecedented public impact
7:56
and the investment pouring in now is
7:58
just staggering. You hear about projects
8:00
like Stargate LLC planning maybe $500
8:02
billion in the US alone. China's
8:04
investing hundreds of billions. It's a
8:06
geopolitical and economic race.
8:08
Huge sums of money chasing the
8:10
potential.
8:10
But it's not all smooth sailing. There's
8:12
this intense debate now. You had that
8:13
open letter in March 2023, right? Musk,
8:16
Waznjak, Benjio, thousands signing
8:18
calling for a pause on giant AI
8:20
experiments
8:20
fighting profound risks to society and
8:22
humanity.
8:23
Yeah. serious concerns. But then you
8:25
have others like Jurgen Schmidt Huber
8:27
who are much more optimistic focusing on
8:29
how AI can make lives better, longer,
8:31
healthier.
8:32
That tension, the caution versus the
8:35
acceleration that really defines this
8:37
moment we're in.
8:38
So, okay, we've traced this incredible
8:40
path from myths to chat GTT. What does
8:43
it all mean for us practically,
8:45
ethically? Because these advancements
8:47
bring a whole wave of ethical challenges
8:49
we have to face.
8:50
Exactly. And that's where the field of
8:52
ethics of artificial intelligence comes
8:54
in. It helps us map out these challenges
8:56
short-term, long-term, everything from
8:58
bias in today's systems to, you know,
9:01
speculating about super intelligence.
9:03
The key thing is understanding this
9:05
stuff isn't academic. It's about
9:06
applying ethical thinking to guide how
9:09
we develop and use this incredibly
9:10
powerful technology.
9:11
Let's dig into some of those immediate
9:13
concerns. Machine bias feels like a
9:15
really big present danger. We've seen it
9:17
already, haven't we? Amazon's hiring
9:19
tool showing gender bias, loan
9:21
applications, parole decisions like the
9:23
compass system showing racial
9:24
disparities, facial recognition working
9:26
less well for darker skin tones.
9:28
It's pervasive and it often comes from
9:31
the data itself reflecting societal
9:33
biases or the algorithms having
9:35
unintended biases or even just
9:37
historical patterns creating unfair
9:39
outcomes. Three key sources there.
9:42
Then there's the blackbox problem. AI
9:44
making huge decisions, loans, university
9:47
spots, jobs, but we can't always see
9:49
why. It's opaque,
9:50
which feels fundamentally unfair,
9:52
doesn't it? If you're denied something,
9:53
but no one can tell you the reason.
9:55
Exactly. And on a societal level, it
9:58
raises this fear of alocracy being ruled
10:01
by algorithms we just don't understand.
10:03
So, there is work on things like
10:04
counterfactual explanations, trying to
10:06
show you what you'd need to change to
10:08
get a different outcome, even if the
10:09
internal logic is hidden. It's one
10:11
potential way forward. Another area is
10:13
autonomous systems and responsibility.
10:16
Those self-driving car accidents, the
10:18
Tesla fatality in 2016, the Uber
10:20
incident in 2018 where the AI kept
10:22
mclassifying the pedestrian.
10:24
Tragic examples. And they raise that
10:26
core question, who is responsible when
10:28
an autonomous system causes harm,
10:30
especially when you extend it to
10:32
autonomous weapons killer robots. The
10:34
whole debate about needing meaningful
10:36
human control is about preventing these
10:38
responsibility gaps. If there's no human
10:41
directly involved, who's accountable?
10:43
It really forces us to ask, how do we
10:45
build fairness and accountability into
10:47
systems that might be biased and opaque?
10:49
Is any level of bias acceptable when
10:51
lives are affected? The goal maybe isn't
10:54
perfectly unbiased AI, which might be
10:56
impossible, but developing robust ways
10:58
to find bias, measure it, and reduce its
11:01
negative impact as much as possible.
11:02
Moving beyond the immediate, we get into
11:05
some really deep long-term questions
11:07
like machine consciousness. Could AI
11:10
actually become conscious, able to feel,
11:12
to suffer? You have researchers like
11:14
Assada designing robots to feel pain?
11:16
Sophia, the robots creators, aiming for
11:18
a living machine. It sounds like sci-fi,
11:20
but philosophers are taking it
11:22
seriously. Metser argues we have an
11:24
ethical duty not to create suffering
11:25
machines. Others debate our obligations
11:27
towards potentially conscious AI. It
11:29
forces us to define what consciousness
11:31
even is,
11:32
which leads right into the moral status
11:33
of AI. Do robots deserve rights? How
11:36
would we even decide?
11:37
Different ideas are being floated.
11:38
There's the content inspired autonomy
11:40
approach. If a machine is rational and
11:43
autonomous, maybe it deserves status
11:45
regardless of being biological.
11:47
Okay.
11:47
Then Kate Darling's indirect duties
11:50
approach. Protect social robots, not for
11:53
their sake, but because mistreating them
11:56
might make us worse humans.
11:58
Interesting. Like how we treat animals.
12:00
Somewhat analogous. Yes. And then the
12:01
relational approach from Kokberg and
12:03
Gungle. Maybe moral status isn't
12:06
inherent but emerges from how we relate
12:08
to these machines socially.
12:10
So it depends on our interactions
12:11
in that view. Yes,
12:12
it's complex territory.
12:13
And there are other risks too, right?
12:15
Like human infeeblement. The worry that
12:18
we'll rely so much on AI. We'll lose our
12:20
own skills, our own ability to make
12:21
judgments, our moral agency, even
12:24
a slow erosion of capability,
12:25
potentially becoming overly dependent.
12:28
And the big one, value alignment. How do
12:30
we make absolutely sure AI goals line up
12:33
with human values?
12:35
This is seen by many as the core
12:36
challenge for long-term safety. Stuart
12:39
Russell's ideas about AI needing to be
12:40
uncertain about human preferences,
12:42
always deferring to us.
12:43
Right. So it doesn't just optimize some
12:45
goal we gave it in a destructive way.
12:47
Exactly. Toby Orid compares the risk of
12:49
unaligned super intelligence to nuclear
12:51
war. An existential threat needing
12:54
careful management now.
12:55
Wow.
12:55
But if we connect this all together,
12:57
these aren't just future worries.
12:59
discussing moral status, value
13:01
alignment. It's part of responsible AI
13:04
development today. And encouragingly,
13:06
when you look at AI ethics guidelines
13:08
emerging worldwide, there's actually a
13:11
lot of agreement on core principles.
13:13
Transparency, fairness, preventing harm,
13:16
accountability, privacy. There seems to
13:19
be a growing global consensus on the
13:21
basics needed for ethical AI.
13:22
What an incredible journey we've
13:24
covered. Seriously, from ancient dreams
13:26
through the science, the ups and downs,
13:27
the AI winters to this current explosion
13:30
fueled by data and deep learning, and
13:32
now facing these profound ethical
13:34
questions. Hopefully, this deep dive has
13:36
given you that shortcut to really grasp
13:38
AI's impact.
13:39
It really has the potential to reshape
13:40
everything, including our own moral
13:42
ideas and our understanding of ourselves
13:44
as humans. It's a fundamental challenge.
13:46
So, here's something to think about as
13:48
we wrap up. Considering how far AI has
13:50
come and all these new challenges it
13:52
throws at us, what responsibility do you
13:54
think we have in shaping its ethical
13:56
path? Not just for us now, but for the
13:58
future and maybe, just maybe, for the
14:00
machines themselves. That's definitely
14:02
something worth continuing your own deep
14:03
dive into.
#Social Issues & Advocacy
#Ethics
#Machine Learning & Artificial Intelligence
#Machine Learning & Artificial Intelligence