episode
stringlengths
45
100
text
stringlengths
1
528
timestamp_link
stringlengths
56
56
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
about existential threat of artificial intelligence. You've seen this community
https://karpathy.ai/lexicap/0004-large.html#00:21:32.800
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
develop over time. You've seen you have a perspective. So what do you think is the best
https://karpathy.ai/lexicap/0004-large.html#00:21:37.600
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
way to talk about AI safety, to think about it, to have discourse about it within AI community
https://karpathy.ai/lexicap/0004-large.html#00:21:42.160
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
and outside and grounded in the fact that Ex Machina is one of the main sources of information
https://karpathy.ai/lexicap/0004-large.html#00:21:48.320
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
for the general public about AI?
https://karpathy.ai/lexicap/0004-large.html#00:21:54.560
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
So I think you're putting it right. There's a big difference between the sort of discussion
https://karpathy.ai/lexicap/0004-large.html#00:21:56.560
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
we ought to have within the AI community and the sort of discussion that really matter
https://karpathy.ai/lexicap/0004-large.html#00:22:02.240
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
in the general public. So I think the picture of Terminator and AI loose and killing people
https://karpathy.ai/lexicap/0004-large.html#00:22:07.600
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
and super intelligence that's going to destroy us, whatever we try, isn't really so useful
https://karpathy.ai/lexicap/0004-large.html#00:22:17.120
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
for the public discussion. Because for the public discussion, the things I believe really
https://karpathy.ai/lexicap/0004-large.html#00:22:24.560
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
matter are the short term and medium term, very likely negative impacts of AI on society,
https://karpathy.ai/lexicap/0004-large.html#00:22:30.000
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
whether it's from security, like, you know, big brother scenarios with face recognition
https://karpathy.ai/lexicap/0004-large.html#00:22:37.200
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
or killer robots, or the impact on the job market, or concentration of power and discrimination,
https://karpathy.ai/lexicap/0004-large.html#00:22:43.280
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
all kinds of social issues, which could actually, some of them could really threaten democracy,
https://karpathy.ai/lexicap/0004-large.html#00:22:50.000
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
for example.
https://karpathy.ai/lexicap/0004-large.html#00:22:57.760
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
Just to clarify, when you said killer robots, you mean autonomous weapon, weapon systems.
https://karpathy.ai/lexicap/0004-large.html#00:22:58.800
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
Yes, I don't mean that's right.
https://karpathy.ai/lexicap/0004-large.html#00:23:04.000
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
So I think these short and medium term concerns should be important parts of the public debate.
https://karpathy.ai/lexicap/0004-large.html#00:23:06.320
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
Now, existential risk, for me is a very unlikely consideration, but still worth academic investigation
https://karpathy.ai/lexicap/0004-large.html#00:23:13.040
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
in the same way that you could say, should we study what could happen if meteorite, you
https://karpathy.ai/lexicap/0004-large.html#00:23:24.640
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
know, came to earth and destroyed it. So I think it's very unlikely that this is going
https://karpathy.ai/lexicap/0004-large.html#00:23:30.080
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
to happen in or happen in a reasonable future. The sort of scenario of an AI getting loose
https://karpathy.ai/lexicap/0004-large.html#00:23:33.920
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
goes against my understanding of at least current machine learning and current neural
https://karpathy.ai/lexicap/0004-large.html#00:23:43.040
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
nets and so on. It's not plausible to me. But of course, I don't have a crystal ball
https://karpathy.ai/lexicap/0004-large.html#00:23:46.560
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
and who knows what AI will be in 50 years from now. So I think it is worth that scientists
https://karpathy.ai/lexicap/0004-large.html#00:23:51.120
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
study those problems. It's just not a pressing question as far as I'm concerned.
https://karpathy.ai/lexicap/0004-large.html#00:23:55.520
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
So before I continue down that line, I have a few questions there. But what do you like
https://karpathy.ai/lexicap/0004-large.html#00:23:59.680
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
and not like about Ex Machina as a movie? Because I actually watched it for the second
https://karpathy.ai/lexicap/0004-large.html#00:24:05.840
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
time and enjoyed it. I hated it the first time, and I enjoyed it quite a bit more the
https://karpathy.ai/lexicap/0004-large.html#00:24:09.840
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
second time when I sort of learned to accept certain pieces of it, see it as a concept
https://karpathy.ai/lexicap/0004-large.html#00:24:15.600
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
movie. What was your experience? What were your thoughts?
https://karpathy.ai/lexicap/0004-large.html#00:24:23.440
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
So the negative is the picture it paints of science is totally wrong. Science in general
https://karpathy.ai/lexicap/0004-large.html#00:24:26.320
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
and AI in particular. Science is not happening in some hidden place by some, you know, really
https://karpathy.ai/lexicap/0004-large.html#00:24:36.080
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
smart guy, one person. This is totally unrealistic. This is not how it happens. Even a team of
https://karpathy.ai/lexicap/0004-large.html#00:24:44.160
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
people in some isolated place will not make it. Science moves by small steps, thanks to
https://karpathy.ai/lexicap/0004-large.html#00:24:52.160
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
the collaboration and community of a large number of people interacting. And all the
https://karpathy.ai/lexicap/0004-large.html#00:24:59.840
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
scientists who are expert in their field kind of know what is going on, even in the industrial
https://karpathy.ai/lexicap/0004-large.html#00:25:10.480
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
labs. It's information flows and leaks and so on. And the spirit of it is very different
https://karpathy.ai/lexicap/0004-large.html#00:25:14.560
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
from the way science is painted in this movie.
https://karpathy.ai/lexicap/0004-large.html#00:25:21.920
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
Yeah, let me ask on that point. It's been the case to this point that kind of even if
https://karpathy.ai/lexicap/0004-large.html#00:25:25.600
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
the research happens inside Google or Facebook, inside companies, it still kind of comes out,
https://karpathy.ai/lexicap/0004-large.html#00:25:32.400
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
ideas come out. Do you think that will always be the case with AI? Is it possible to bottle
https://karpathy.ai/lexicap/0004-large.html#00:25:36.800
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
ideas to the point where there's a set of breakthroughs that go completely undiscovered
https://karpathy.ai/lexicap/0004-large.html#00:25:41.680
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
by the general research community? Do you think that's even possible?
https://karpathy.ai/lexicap/0004-large.html#00:25:47.360
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
It's possible, but it's unlikely. It's not how it is done now. It's not how I can foresee
https://karpathy.ai/lexicap/0004-large.html#00:25:52.240
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
it in the foreseeable future. But of course, I don't have a crystal ball and science is
https://karpathy.ai/lexicap/0004-large.html#00:25:59.520
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
a crystal ball. And so who knows? This is science fiction after all.
https://karpathy.ai/lexicap/0004-large.html#00:26:09.520
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
I think it's ominous that the lights went off during that discussion.
https://karpathy.ai/lexicap/0004-large.html#00:26:14.960
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
So the problem, again, there's one thing is the movie and you could imagine all kinds
https://karpathy.ai/lexicap/0004-large.html#00:26:21.440
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
of science fiction. The problem for me, maybe similar to the question about existential
https://karpathy.ai/lexicap/0004-large.html#00:26:25.320
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
risk, is that this kind of movie paints such a wrong picture of what is the actual science
https://karpathy.ai/lexicap/0004-large.html#00:26:30.320
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
and how it's going on that it can have unfortunate effects on people's understanding of current
https://karpathy.ai/lexicap/0004-large.html#00:26:39.440
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
science. And so that's kind of sad.
https://karpathy.ai/lexicap/0004-large.html#00:26:45.640
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
There's an important principle in research, which is diversity. So in other words, research
https://karpathy.ai/lexicap/0004-large.html#00:26:50.800
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
is exploration. Research is exploration in the space of ideas. And different people will
https://karpathy.ai/lexicap/0004-large.html#00:26:58.440
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
focus on different directions. And this is not just good, it's essential. So I'm totally
https://karpathy.ai/lexicap/0004-large.html#00:27:03.720
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
fine with people exploring directions that are contrary to mine or look orthogonal to
https://karpathy.ai/lexicap/0004-large.html#00:27:09.520
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
mine. I am more than fine. I think it's important. I and my friends don't claim we have universal
https://karpathy.ai/lexicap/0004-large.html#00:27:16.440
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
truth about what will, especially about what will happen in the future. Now that being
https://karpathy.ai/lexicap/0004-large.html#00:27:24.920
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
said, we have our intuitions and then we act accordingly according to where we think we
https://karpathy.ai/lexicap/0004-large.html#00:27:29.560
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
can be most useful and where society has the most to gain or to lose. We should have those
https://karpathy.ai/lexicap/0004-large.html#00:27:36.560
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
debates and not end up in a society where there's only one voice and one way of thinking
https://karpathy.ai/lexicap/0004-large.html#00:27:42.480
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
and research money is spread out.
https://karpathy.ai/lexicap/0004-large.html#00:27:49.800
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
So disagreement is a sign of good research, good science.
https://karpathy.ai/lexicap/0004-large.html#00:27:53.520
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
Yes.
https://karpathy.ai/lexicap/0004-large.html#00:27:59.040
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
The idea of bias in the human sense of bias. How do you think about instilling in machine
https://karpathy.ai/lexicap/0004-large.html#00:28:00.040
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
learning something that's aligned with human values in terms of bias? We intuitively as
https://karpathy.ai/lexicap/0004-large.html#00:28:08.600
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
human beings have a concept of what bias means, of what fundamental respect for other human
https://karpathy.ai/lexicap/0004-large.html#00:28:15.240
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
beings means. But how do we instill that into machine learning systems, do you think?
https://karpathy.ai/lexicap/0004-large.html#00:28:21.160
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
So I think there are short term things that are already happening and then there are long
https://karpathy.ai/lexicap/0004-large.html#00:28:26.760
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
term things that we need to do. In the short term, there are techniques that have been
https://karpathy.ai/lexicap/0004-large.html#00:28:32.360
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
proposed and I think will continue to be improved and maybe alternatives will come up to take
https://karpathy.ai/lexicap/0004-large.html#00:28:38.360
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
data sets in which we know there is bias, we can measure it. Pretty much any data set
https://karpathy.ai/lexicap/0004-large.html#00:28:44.200
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
where humans are being observed taking decisions will have some sort of bias, discrimination
https://karpathy.ai/lexicap/0004-large.html#00:28:50.120
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
against particular groups and so on.
https://karpathy.ai/lexicap/0004-large.html#00:28:55.520
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
And we can use machine learning techniques to try to build predictors, classifiers that
https://karpathy.ai/lexicap/0004-large.html#00:28:59.000
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
are going to be less biased. We can do it, for example, using adversarial methods to
https://karpathy.ai/lexicap/0004-large.html#00:29:04.240
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
make our systems less sensitive to these variables we should not be sensitive to.
https://karpathy.ai/lexicap/0004-large.html#00:29:11.600
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
So these are clear, well defined ways of trying to address the problem. Maybe they have weaknesses
https://karpathy.ai/lexicap/0004-large.html#00:29:18.360
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
and more research is needed and so on. But I think in fact they are sufficiently mature
https://karpathy.ai/lexicap/0004-large.html#00:29:23.520
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
that governments should start regulating companies where it matters, say like insurance companies,
https://karpathy.ai/lexicap/0004-large.html#00:29:28.840
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
so that they use those techniques. Because those techniques will probably reduce the
https://karpathy.ai/lexicap/0004-large.html#00:29:35.240
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
bias but at a cost. For example, maybe their predictions will be less accurate and so companies
https://karpathy.ai/lexicap/0004-large.html#00:29:40.480
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
will not do it until you force them.
https://karpathy.ai/lexicap/0004-large.html#00:29:46.440
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
All right, so this is short term. Long term, I'm really interested in thinking how we can
https://karpathy.ai/lexicap/0004-large.html#00:29:48.560
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
instill moral values into computers. Obviously, this is not something we'll achieve in the
https://karpathy.ai/lexicap/0004-large.html#00:29:56.040
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
next five or 10 years. How can we, you know, there's already work in detecting emotions,
https://karpathy.ai/lexicap/0004-large.html#00:30:01.560
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
for example, in images, in sounds, in texts, and also studying how different agents interacting
https://karpathy.ai/lexicap/0004-large.html#00:30:08.120
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
in different ways may correspond to patterns of, say, injustice, which could trigger anger.
https://karpathy.ai/lexicap/0004-large.html#00:30:19.880
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
So these are things we can do in the medium term and eventually train computers to model,
https://karpathy.ai/lexicap/0004-large.html#00:30:28.200
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
for example, how humans react emotionally. I would say the simplest thing is unfair situations
https://karpathy.ai/lexicap/0004-large.html#00:30:37.840
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
which trigger anger. This is one of the most basic emotions that we share with other animals.
https://karpathy.ai/lexicap/0004-large.html#00:30:46.960
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
I think it's quite feasible within the next few years that we can build systems that can
https://karpathy.ai/lexicap/0004-large.html#00:30:52.680
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
detect these kinds of things to the extent, unfortunately, that they understand enough
https://karpathy.ai/lexicap/0004-large.html#00:30:57.160
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
about the world around us, which is a long time away. But maybe we can initially do this
https://karpathy.ai/lexicap/0004-large.html#00:31:01.980
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
in virtual environments. So you can imagine a video game where agents interact in some
https://karpathy.ai/lexicap/0004-large.html#00:31:08.240
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
ways and then some situations trigger an emotion. I think we could train machines to detect
https://karpathy.ai/lexicap/0004-large.html#00:31:14.840
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
those situations and predict that the particular emotion will likely be felt if a human was
https://karpathy.ai/lexicap/0004-large.html#00:31:21.640
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
playing one of the characters.
https://karpathy.ai/lexicap/0004-large.html#00:31:27.400
Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4
You have shown excitement and done a lot of excellent work with unsupervised learning.
https://karpathy.ai/lexicap/0004-large.html#00:31:29.460