episode
stringlengths
45
100
text
stringlengths
1
528
timestamp_link
stringlengths
56
56
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
You just can't remember a sequence of moves that's 50 moves long. And you can't imagine
https://karpathy.ai/lexicap/0009-large.html#00:09:02.000
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
the board correctly for that many moves into the future.
https://karpathy.ai/lexicap/0009-large.html#00:09:08.960
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
Of course, the top players, I'm much more familiar with chess, but the top players probably have,
https://karpathy.ai/lexicap/0009-large.html#00:09:13.040
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
they have echoes of the same kind of intuition instinct that in a moment's time AlphaGo applies
https://karpathy.ai/lexicap/0009-large.html#00:09:19.680
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
when they see a board. I mean, they've seen those patterns, human beings have seen those patterns
https://karpathy.ai/lexicap/0009-large.html#00:09:26.720
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
before at the top, at the grandmaster level. It seems that there is some similarities or maybe
https://karpathy.ai/lexicap/0009-large.html#00:09:31.600
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
it's our imagination creates a vision of those similarities, but it feels like this kind of
https://karpathy.ai/lexicap/0009-large.html#00:09:41.360
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
pattern recognition that the AlphaGo approaches are using is similar to what human beings at the
https://karpathy.ai/lexicap/0009-large.html#00:09:47.360
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
top level are using.
https://karpathy.ai/lexicap/0009-large.html#00:09:53.920
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
I think there's, there's some truth to that, but not entirely. Yeah. I mean, I think the,
https://karpathy.ai/lexicap/0009-large.html#00:09:55.360
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
the extent to which a human grandmaster can reliably instantly recognize the right move
https://karpathy.ai/lexicap/0009-large.html#00:10:03.040
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
and instantly recognize the value of the position. I think that's a little bit overrated.
https://karpathy.ai/lexicap/0009-large.html#00:10:10.720
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
But if you sacrifice a queen, for example, I mean, there's these, there's these beautiful games of
https://karpathy.ai/lexicap/0009-large.html#00:10:15.840
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
chess with Bobby Fischer, somebody where it's seeming to make a bad move. And I'm not sure
https://karpathy.ai/lexicap/0009-large.html#00:10:20.480
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
there's a perfect degree of calculation involved where they've calculated all the possible things
https://karpathy.ai/lexicap/0009-large.html#00:10:28.400
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
that happen, but there's an instinct there, right? That somehow adds up to
https://karpathy.ai/lexicap/0009-large.html#00:10:34.720
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
Yeah. So I think what happens is you, you, you get a sense that there's some possibility in the
https://karpathy.ai/lexicap/0009-large.html#00:10:40.640
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
position, even if you make a weird looking move, that it opens up some, some lines of,
https://karpathy.ai/lexicap/0009-large.html#00:10:46.160
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
of calculation that otherwise would be definitely bad. And, and it's that intuition that there's
https://karpathy.ai/lexicap/0009-large.html#00:10:56.080
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
something here in this position that might, might yield a win.
https://karpathy.ai/lexicap/0009-large.html#00:11:05.040
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
And then you follow that, right? And, and in some sense, when a, when a chess player is
https://karpathy.ai/lexicap/0009-large.html#00:11:10.880
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
following a line and in his or her mind, they're, they're mentally simulating what the other person
https://karpathy.ai/lexicap/0009-large.html#00:11:16.080
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
is going to do, what the opponent is going to do. And they can do that as long as the moves are kind
https://karpathy.ai/lexicap/0009-large.html#00:11:23.440
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
of forced, right? As long as there's, you know, there's a, a fort we call a forcing variation
https://karpathy.ai/lexicap/0009-large.html#00:11:29.200
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
where the opponent doesn't really have much choice how to respond. And then you follow that,
https://karpathy.ai/lexicap/0009-large.html#00:11:34.640
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
how to respond. And then you see if you can force them into a situation where you win.
https://karpathy.ai/lexicap/0009-large.html#00:11:39.120
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
You know, we see plenty of mistakes even, even in grandmaster games where they just miss some
https://karpathy.ai/lexicap/0009-large.html#00:11:43.520
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
simple three, four, five move combination that, you know, wasn't particularly apparent in,
https://karpathy.ai/lexicap/0009-large.html#00:11:51.920
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
in the position, but was still there. That's the thing that makes us human.
https://karpathy.ai/lexicap/0009-large.html#00:11:58.560
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
Yeah. So when you mentioned that in Othello, those games were after some matter reasoning
https://karpathy.ai/lexicap/0009-large.html#00:12:02.560
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
improvements and research was able to beat you. How did that make you feel?
https://karpathy.ai/lexicap/0009-large.html#00:12:09.680
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
Part of the meta reasoning capability that it had was based on learning and, and you could
https://karpathy.ai/lexicap/0009-large.html#00:12:14.960
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
sit down the next day and you could just feel that it had got a lot smarter, you know, and all of a
https://karpathy.ai/lexicap/0009-large.html#00:12:23.680
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
sudden you really felt like you're sort of pressed against the wall because it was, it was much more
https://karpathy.ai/lexicap/0009-large.html#00:12:30.240
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
aggressive and, and was totally unforgiving of any minor mistake that you might make. And, and
https://karpathy.ai/lexicap/0009-large.html#00:12:37.280
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
actually it seemed understood the game better than I did. And Gary Kasparov has this quote where
https://karpathy.ai/lexicap/0009-large.html#00:12:43.440
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
during his match against Deep Blue, he said, he suddenly felt that there was a new kind of
https://karpathy.ai/lexicap/0009-large.html#00:12:52.000
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
intelligence across the board. Do you think that's a scary or an exciting
https://karpathy.ai/lexicap/0009-large.html#00:12:56.880
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
possibility for, for Kasparov and for yourself in, in the context of chess, purely sort of
https://karpathy.ai/lexicap/0009-large.html#00:13:03.120
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
in this, like that feeling, whatever that is? I think it's definitely an exciting feeling.
https://karpathy.ai/lexicap/0009-large.html#00:13:10.320
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
You know, this is what made me work on AI in the first place was as soon as I really understood
https://karpathy.ai/lexicap/0009-large.html#00:13:17.680
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
what a computer was, I wanted to make it smart. You know, I started out with the first program
https://karpathy.ai/lexicap/0009-large.html#00:13:23.680
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
I wrote was for the Sinclair programmable calculator. And I think you could write a
https://karpathy.ai/lexicap/0009-large.html#00:13:29.920
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
21 step algorithm. That was the biggest program you could write, something like that. And do
https://karpathy.ai/lexicap/0009-large.html#00:13:35.680
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
little arithmetic calculations. So I think I implemented Newton's method for a square
https://karpathy.ai/lexicap/0009-large.html#00:13:42.800
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
roots and a few other things like that. But then, you know, I thought, okay, if I just had more
https://karpathy.ai/lexicap/0009-large.html#00:13:48.080
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
space, I could make this thing intelligent. And so I started thinking about AI and,
https://karpathy.ai/lexicap/0009-large.html#00:13:54.240
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
and I think the, the, the thing that's scary is not, is not the chess program
https://karpathy.ai/lexicap/0009-large.html#00:14:04.880
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
because, you know, chess programs, they're not in the taking over the world business.
https://karpathy.ai/lexicap/0009-large.html#00:14:11.280
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
But if you extrapolate, you know, there are things about chess that don't resemble
https://karpathy.ai/lexicap/0009-large.html#00:14:19.520
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
the real world, right? We know, we know the rules of chess.
https://karpathy.ai/lexicap/0009-large.html#00:14:29.040
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
The chess board is completely visible to the program where of course the real world is not
https://karpathy.ai/lexicap/0009-large.html#00:14:35.120
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
most, most of the real world is, is not visible from wherever you're sitting, so to speak.
https://karpathy.ai/lexicap/0009-large.html#00:14:40.720
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
And to overcome those kinds of problems, you need qualitatively different algorithms. Another thing
https://karpathy.ai/lexicap/0009-large.html#00:14:47.520
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
about the real world is that, you know, we, we regularly plan ahead on the timescales involving
https://karpathy.ai/lexicap/0009-large.html#00:14:56.720
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
billions or trillions of steps. Now we don't plan those in detail, but you know, when you
https://karpathy.ai/lexicap/0009-large.html#00:15:05.520
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
choose to do a PhD at Berkeley, that's a five year commitment and that amounts to about a trillion
https://karpathy.ai/lexicap/0009-large.html#00:15:12.400
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
motor control steps that you will eventually be committed to. Including going up the stairs,
https://karpathy.ai/lexicap/0009-large.html#00:15:19.680
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
opening doors, drinking water. Yeah. I mean, every, every finger movement while you're typing,
https://karpathy.ai/lexicap/0009-large.html#00:15:26.240
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
every character of every paper and the thesis and everything. So you're not committing in
https://karpathy.ai/lexicap/0009-large.html#00:15:32.240
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
advance to the specific motor control steps, but you're still reasoning on a timescale that
https://karpathy.ai/lexicap/0009-large.html#00:15:36.240
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
will eventually reduce to trillions of motor control actions. And so for all of these reasons,
https://karpathy.ai/lexicap/0009-large.html#00:15:41.600
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
you know, AlphaGo and Deep Blue and so on don't represent any kind of threat to humanity,
https://karpathy.ai/lexicap/0009-large.html#00:15:52.160
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
but they are a step towards it, right? And progress in AI occurs by essentially removing
https://karpathy.ai/lexicap/0009-large.html#00:15:58.080
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
one by one these assumptions that make problems easy. Like the assumption of complete observability
https://karpathy.ai/lexicap/0009-large.html#00:16:07.040
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
of the situation, right? We remove that assumption, you need a much more complicated
https://karpathy.ai/lexicap/0009-large.html#00:16:14.640
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
kind of computing design. It needs, it needs something that actually keeps track of all the
https://karpathy.ai/lexicap/0009-large.html#00:16:20.160
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
things you can't see and tries to estimate what's going on. And there's inevitable uncertainty
https://karpathy.ai/lexicap/0009-large.html#00:16:25.120
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
in that. So it becomes a much more complicated problem. But, you know, we are removing those
https://karpathy.ai/lexicap/0009-large.html#00:16:31.040
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
assumptions. We are starting to have algorithms that can cope with much longer timescales,
https://karpathy.ai/lexicap/0009-large.html#00:16:36.880
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
that can cope with uncertainty, that can cope with partial observability.
https://karpathy.ai/lexicap/0009-large.html#00:16:42.320
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
And so each of those steps sort of magnifies by a thousand the range of things that we can
https://karpathy.ai/lexicap/0009-large.html#00:16:47.520
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
do with AI systems. So the way I started in AI, I wanted to be a psychiatrist for a long time. I
https://karpathy.ai/lexicap/0009-large.html#00:16:54.240
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
wanted to understand the mind in high school and of course program and so on. And I showed up
https://karpathy.ai/lexicap/0009-large.html#00:16:58.880
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
University of Illinois to an AI lab and they said, okay, I don't have time for you,
https://karpathy.ai/lexicap/0009-large.html#00:17:04.960
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
but here's a book, AI and Modern Approach. I think it was the first edition at the time.
https://karpathy.ai/lexicap/0009-large.html#00:17:10.160
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
Here, go, go, go learn this. And I remember the lay of the land was, well, it's incredible that
https://karpathy.ai/lexicap/0009-large.html#00:17:16.640
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
we solved chess, but we'll never solve go. I mean, it was pretty certain that go
https://karpathy.ai/lexicap/0009-large.html#00:17:22.080
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
in the way we thought about systems that reason wasn't possible to solve. And now we've solved
https://karpathy.ai/lexicap/0009-large.html#00:17:27.360
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
this. So it's a very... Well, I think I would have said that it's unlikely we could take
https://karpathy.ai/lexicap/0009-large.html#00:17:33.440
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
the kind of algorithm that was used for chess and just get it to scale up and work well for go.
https://karpathy.ai/lexicap/0009-large.html#00:17:39.840
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
And at the time what we thought was that in order to solve go, we would have to do something similar
https://karpathy.ai/lexicap/0009-large.html#00:17:46.480
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
to the way humans manage the complexity of go, which is to break it down into kind of sub games.
https://karpathy.ai/lexicap/0009-large.html#00:17:56.800
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
So when a human thinks about a go board, they think about different parts of the board as sort
https://karpathy.ai/lexicap/0009-large.html#00:18:02.800
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
of weakly connected to each other. And they think about, okay, within this part of the board, here's
https://karpathy.ai/lexicap/0009-large.html#00:18:08.320
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
how things could go in that part of board, here's how things could go. And then you try to sort of
https://karpathy.ai/lexicap/0009-large.html#00:18:13.280
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
couple those two analyses together and deal with the interactions and maybe revise your views of
https://karpathy.ai/lexicap/0009-large.html#00:18:18.000
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
how things are going to go in each part. And then you've got maybe five, six, seven, ten parts of
https://karpathy.ai/lexicap/0009-large.html#00:18:24.000
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
the board. And that actually resembles the real world much more than chess does because in the
https://karpathy.ai/lexicap/0009-large.html#00:18:28.640
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
real world, we have work, we have home life, we have sport, different kinds of activities,
https://karpathy.ai/lexicap/0009-large.html#00:18:38.160
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
shopping, these all are connected to each other, but they're weakly connected. So when I'm typing
https://karpathy.ai/lexicap/0009-large.html#00:18:46.880
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
a paper, I don't simultaneously have to decide which order I'm going to get the milk and the
https://karpathy.ai/lexicap/0009-large.html#00:18:54.560
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
butter, that doesn't affect the typing. But I do need to realize, okay, I better finish this
https://karpathy.ai/lexicap/0009-large.html#00:19:01.280
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
before the shops close because I don't have anything, I don't have any food at home. So
https://karpathy.ai/lexicap/0009-large.html#00:19:08.240
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
there's some weak connection, but not in the way that chess works where everything is tied into a
https://karpathy.ai/lexicap/0009-large.html#00:19:12.320
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
single stream of thought. So the thought was that to solve go, we'd have to make progress on stuff
https://karpathy.ai/lexicap/0009-large.html#00:19:19.040
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
that would be useful for the real world. And in a way, AlphaGo is a little bit disappointing,
https://karpathy.ai/lexicap/0009-large.html#00:19:26.080
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
right? Because the program designed for AlphaGo is actually not that different from Deep Blue
https://karpathy.ai/lexicap/0009-large.html#00:19:29.520
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
or even from Arthur Samuel's checker playing program from the 1950s. And in fact, the two
https://karpathy.ai/lexicap/0009-large.html#00:19:39.680
Stuart Russell: Long-Term Future of Artificial Intelligence | Lex Fridman Podcast #9
things that make AlphaGo work is one is this amazing ability to evaluate the positions,
https://karpathy.ai/lexicap/0009-large.html#00:19:48.160