1
00:00:00,080 --> 00:00:01,997
The following is a conversation with Yan

2
00:00:02,080 --> 00:00:04,477
Lakun. He's considered to be one of the

3
00:00:04,560 --> 00:00:09,357
fathers of deep learning, which if
you've been hiding under a rock, is the

4
00:00:09,440 --> 00:00:12,637
recent revolution in AI that has
captivated the world with the

5
00:00:12,720 --> 00:00:14,717
possibility of what machines can learn

6
00:00:14,800 --> 00:00:17,677
from data. He's a professor at New York

7
00:00:17,760 --> 00:00:19,757
University, a vice president and chief

8
00:00:19,840 --> 00:00:23,197
AI scientist at Facebook, and co-reient

9
00:00:23,280 --> 00:00:27,677
of the touring award for his work
on deep learning. He's probably best known

10
00:00:27,760 --> 00:00:31,917
as the founding father of convolutional
neural networks, in particular their

11
00:00:32,000 --> 00:00:33,437
application to optical character

12
00:00:33,520 --> 00:00:36,557
recognition and the famed emnest data

13
00:00:36,640 --> 00:00:40,077
set. He is also an outspoken personality

14
00:00:40,160 --> 00:00:44,397
unafraid to speak his mind in a
distinctive French accent and explore

15
00:00:44,480 --> 00:00:46,637
provocative ideas both in the rigorous

16
00:00:46,720 --> 00:00:49,437
medium of academic research and the

17
00:00:49,520 --> 00:00:51,437
somewhat less rigorous medium of Twitter

18
00:00:51,520 --> 00:00:54,077
and Facebook. This is the artificial

19
00:00:54,160 --> 00:00:58,717
intelligence podcast. If you enjoy it,
subscribe on YouTube, give it five stars

20
00:00:58,800 --> 00:01:01,037
on iTunes, support it on Patreon, or

21
00:01:01,120 --> 00:01:03,277
simply connect with me on Twitter at Lex

22
00:01:03,360 --> 00:01:07,517
Freedman, spelled F R I D M. And now,

23
00:01:07,600 --> 00:01:13,000
here's my conversation with Yan Lun.

24
00:01:21,220 --> 00:01:23,240
[music]

25
00:01:28,880 --> 00:01:31,277
You said that 2001 Space Odyssey is one

26
00:01:31,360 --> 00:01:33,357
of your favorite movies.

27
00:01:33,440 --> 00:01:36,877
Hell 9000 decides to get rid of the

28
00:01:36,960 --> 00:01:38,397
astronauts for people who haven't seen

29
00:01:38,480 --> 00:01:42,797
the movie. Spoiler alert, because he it

30
00:01:42,880 --> 00:01:46,637
she believes that the astronauts they

31
00:01:46,720 --> 00:01:49,037
will interfere with the mission. Do you

32
00:01:49,120 --> 00:01:54,637
see Hal as flawed in some fundamental
way or even evil or did he do the right

33
00:01:54,720 --> 00:01:58,717
thing? >> Neither.
There's no notion of evil in

34
00:01:58,800 --> 00:02:01,117
that in that context other than the fact

35
00:02:01,200 --> 00:02:06,317
that people die. But it was an example
of what people call uh value

36
00:02:06,400 --> 00:02:10,077
misalignment, right? And you give
an objective to a machine and the machine

37
00:02:10,160 --> 00:02:14,797
strives to achieve this objective and if
you don't put any constraints on this

38
00:02:14,880 --> 00:02:16,717
objective like don't kill people and

39
00:02:16,800 --> 00:02:19,677
don't do things like this.

40
00:02:19,760 --> 00:02:24,717
The machine given the power will do
stupid things just to achieve this this

41
00:02:24,800 --> 00:02:28,397
objective or damaging things to achieve
this objective. It's a little bit like I

42
00:02:28,480 --> 00:02:30,237
mean we are used to this in the context

43
00:02:30,320 --> 00:02:32,957
of human society.

44
00:02:33,040 --> 00:02:36,877
We we put in place

45
00:02:36,960 --> 00:02:40,397
laws to prevent people from doing bad
things because spontaneously they would

46
00:02:40,480 --> 00:02:44,877
do those bad things, right?
So we have to shape their

47
00:02:44,960 --> 00:02:48,077
cost function, their objective function
if you want through laws to kind of

48
00:02:48,160 --> 00:02:50,797
correct and education obviously to sort

49
00:02:50,880 --> 00:02:55,117
of correct for for those. So maybe just

50
00:02:55,200 --> 00:02:59,197
pushing a little further on on that point,

51
00:02:59,280 --> 00:03:01,517
Hal, you know, there's a mission.

52
00:03:01,600 --> 00:03:04,157
there's a this fuzziness around the the

53
00:03:04,240 --> 00:03:06,077
ambiguity around what the actual mission

54
00:03:06,160 --> 00:03:08,557
is but

55
00:03:08,640 --> 00:03:11,357
you know do you think that there will be

56
00:03:11,440 --> 00:03:15,917
a time from a utilitarian perspective
where an AI system where it is not

57
00:03:16,000 --> 00:03:20,557
misalignment where it is alignment
for the greater good of society that an AI

58
00:03:20,640 --> 00:03:23,037
system will make decisions
that are difficult

59
00:03:23,120 --> 00:03:27,997
>> well that's the trick I mean eventually
we'll have to figure out how to do this

60
00:03:28,080 --> 00:03:30,797
and again we're not starting from
scratch because we've been doing this

61
00:03:30,880 --> 00:03:33,677
with humans for for millennia.

62
00:03:33,760 --> 00:03:35,837
>> So designing objective functions for

63
00:03:35,920 --> 00:03:37,437
people is something that we know how to

64
00:03:37,520 --> 00:03:40,957
do and we don't do it by um you know

65
00:03:41,040 --> 00:03:43,597
programming things although

66
00:03:43,680 --> 00:03:48,397
the legal code is called code.
Um so that tells you something and it's

67
00:03:48,480 --> 00:03:51,277
actually the design of an objective
function. That's really what legal code

68
00:03:51,360 --> 00:03:54,877
is right. It tells you here is what you
can do here is what you can't do. If you

69
00:03:54,960 --> 00:03:58,877
do it you pay that much.
That's that's an objective function.

70
00:03:58,960 --> 00:04:02,957
So there is this idea somehow that it's
a new thing for people to try to design

71
00:04:03,040 --> 00:04:05,917
objective functions that are aligned
with the common good. But no, we've been

72
00:04:06,000 --> 00:04:07,357
writing laws for millennia and that's

73
00:04:07,440 --> 00:04:09,997
exactly what it is. Um so this that's

74
00:04:10,080 --> 00:04:13,837
where you know uh the science of uh law

75
00:04:13,920 --> 00:04:18,477
making and uh and computer
science will >> come together

76
00:04:18,560 --> 00:04:21,597
>> will come together.
>> So it's nothing there's nothing special

77
00:04:21,680 --> 00:04:24,637
about hell or AI systems. It's just a

78
00:04:24,720 --> 00:04:28,797
continuation of tools used to make some
of these difficult ethical judgments

79
00:04:28,880 --> 00:04:31,757
that laws make. >> Yeah.
And we and we have systems like

80
00:04:31,840 --> 00:04:37,117
this already that you know uh make many
decisions for for ourselves in society

81
00:04:37,200 --> 00:04:40,957
that you know need to be designed in a
way that they like you know rules about

82
00:04:41,040 --> 00:04:45,837
things that sometime sometimes have bad
side effects and we have to be flexible

83
00:04:45,920 --> 00:04:48,957
enough about those rules so that they
can be broken when it's obvious that

84
00:04:49,040 --> 00:04:52,637
they shouldn't be applied. So,
you don't see this on the camera

85
00:04:52,720 --> 00:04:56,317
here, but all the decoration in this room
is all pictures from 2001 of Space

86
00:04:56,400 --> 00:05:00,557
Odyssey. [laughter] >> Wow.
Is that by accident or is there a

87
00:05:00,640 --> 00:05:00,797
lot?

88
00:05:00,880 --> 00:05:04,557
>> It's not by accident. It's by design.

89
00:05:04,640 --> 00:05:07,757
>> Oh, wow. So, if you were um if you were

90
00:05:07,840 --> 00:05:12,637
to build HAL 10,000,
so an improvement of HAL 9,000,

91
00:05:12,720 --> 00:05:14,717
what would you improve? Well, first of

92
00:05:14,800 --> 00:05:17,757
all, I wouldn't ask it to hold secrets

93
00:05:17,840 --> 00:05:21,757
and tell lies because that's really
what breaks it in the end. That's the the

94
00:05:21,840 --> 00:05:26,157
fact that it's asking itself questions
about the purpose of the mission and

95
00:05:26,240 --> 00:05:29,997
it's you know pieces things together
that it's heard you know all the secrecy

96
00:05:30,080 --> 00:05:34,077
of the preparation of the mission and
the fact that it was discovery on on the

97
00:05:34,160 --> 00:05:35,677
lunar surface that really was kept

98
00:05:35,760 --> 00:05:38,957
secret and and one part of how's memory

99
00:05:39,040 --> 00:05:43,117
knows this and the other part is does
not know it and is supposed to not tell

100
00:05:43,200 --> 00:05:45,197
anyone and that creates internal

101
00:05:45,280 --> 00:05:46,797
conflict. So you think there's never

102
00:05:46,880 --> 00:05:50,317
should be a set of things that an AI

103
00:05:50,400 --> 00:05:53,677
system should not be allowed

104
00:05:53,760 --> 00:05:56,717
like a a set of facts that should not be

105
00:05:56,800 --> 00:05:59,357
shared with the human operators.

106
00:05:59,440 --> 00:06:01,437
Well, I think no. I think that I think

107
00:06:01,520 --> 00:06:06,797
it should be a bit like in the design of

108
00:06:06,880 --> 00:06:09,357
autonomous AI systems. Um, there should

109
00:06:09,440 --> 00:06:12,237
be the equivalent of, you know, the the

110
00:06:12,320 --> 00:06:16,237
the the oath that hypocratic oath Yeah.

111
00:06:16,320 --> 00:06:21,197
>> that uh doctors sign up to, right?
So, there's certain things, certain rules

112
00:06:21,280 --> 00:06:25,197
that that that you have to abide by and we
can sort of hardwire this into into

113
00:06:25,280 --> 00:06:29,197
our into our machines to kind of make
sure they don't go. So I'm not you know

114
00:06:29,280 --> 00:06:33,837
an advocate of the the three law three
laws of robotics you know the azimov

115
00:06:33,920 --> 00:06:35,277
kind of thing because I don't think it's

116
00:06:35,360 --> 00:06:38,477
practical but but you know some some

117
00:06:38,560 --> 00:06:41,757
level of uh of limits but but to be

118
00:06:41,840 --> 00:06:44,637
clear uh this is not these are not

119
00:06:44,720 --> 00:06:48,317
questions that are kind of really worth

120
00:06:48,400 --> 00:06:52,237
asking today because we just don't have
the technology to do this. We don't we

121
00:06:52,320 --> 00:06:54,637
don't have autonomous intelligent
machines. We have intelligent machines,

122
00:06:54,720 --> 00:06:57,997
semi- intelligent machines
that are very specialized,

123
00:06:58,080 --> 00:07:01,277
but they don't they don't really sort
of satisfy an objective. They're just, you

124
00:07:01,360 --> 00:07:03,757
know, kind of trained to do one thing.

125
00:07:03,840 --> 00:07:07,037
>> So, until we have some idea for design

126
00:07:07,120 --> 00:07:11,837
of a full-fledged autonomous intelligent
system, asking the question of how how

127
00:07:11,920 --> 00:07:15,677
we design is objective, I think
is a little a little too abstract.

128
00:07:15,760 --> 00:07:17,117
>> It's a little too abstract. There's

129
00:07:17,200 --> 00:07:19,757
useful elements to it in that it helps

130
00:07:19,840 --> 00:07:23,997
us understand our own ethical codes,

131
00:07:24,080 --> 00:07:26,397
humans. So even just as a thought

132
00:07:26,480 --> 00:07:29,437
experiment, if you imagine that an AGI

133
00:07:29,520 --> 00:07:32,397
system is here today, h how would we

134
00:07:32,480 --> 00:07:34,317
program it is a kind of nice thought

135
00:07:34,400 --> 00:07:37,677
experiment of constructing how should we

136
00:07:37,760 --> 00:07:40,717
have a law have a system of laws for us

137
00:07:40,800 --> 00:07:42,877
humans. It's it's just a nice practical

138
00:07:42,960 --> 00:07:45,677
tool. And I think there's echoes of that

139
00:07:45,760 --> 00:07:49,197
idea too in the AI systems we have today

140
00:07:49,280 --> 00:07:53,357
that don't have to be that intelligent
like autonomous vehicles. These things

141
00:07:53,440 --> 00:07:55,757
start creeping in that they're worth

142
00:07:55,840 --> 00:07:57,037
thinking about but certainly they

143
00:07:57,120 --> 00:08:00,877
shouldn't be framed as as hell. Yeah.

144
00:08:00,960 --> 00:08:04,237
>> Looking back,
what is the most I'm sorry

145
00:08:04,320 --> 00:08:07,677
if it's a silly question, but what is

146
00:08:07,760 --> 00:08:12,317
the most beautiful or surprising idea
in deep learning or AI in general that

147
00:08:12,400 --> 00:08:14,077
you've ever come across sort of

148
00:08:14,160 --> 00:08:17,117
personally where you sat back

149
00:08:17,200 --> 00:08:21,997
and and just had this kind of a that's
pretty cool moment. That's nice

150
00:08:22,080 --> 00:08:24,397
surprising. >> I don't know
if it's an idea rather than

151
00:08:24,480 --> 00:08:27,277
a sort of empirical

152
00:08:27,360 --> 00:08:29,357
uh fact.

153
00:08:29,440 --> 00:08:31,997
The fact that

154
00:08:32,080 --> 00:08:36,797
you can build gigantic
neural nets train them on,

155
00:08:36,880 --> 00:08:40,557
you know, relatively small
amounts of data relatively

156
00:08:40,640 --> 00:08:44,557
with stoastic gradient and descent
and that it actually works. Uh breaks

157
00:08:44,640 --> 00:08:48,637
everything you read in every textbook,
right? Every pre deep learning textbook

158
00:08:48,720 --> 00:08:53,037
that told you you need to have fewer
parameters than you have data samples.

159
00:08:53,120 --> 00:08:56,397
Um you know if you have a non-convex
objective function, you have no

160
00:08:56,480 --> 00:08:59,197
guarantee of convergence. You know all
those things that you read in textbook

161
00:08:59,280 --> 00:09:02,397
and they tell you stay away
from this and they're all wrong.

162
00:09:02,480 --> 00:09:05,117
>> Huge number of parameters non-convex

163
00:09:05,200 --> 00:09:10,797
and somehow which just vary relative
to the number of uh parameters data it's

164
00:09:10,880 --> 00:09:14,717
able to learn anything >> right >>
does that still surprise you today?

165
00:09:14,800 --> 00:09:17,517
>> Well it it was kind of obvious to me

166
00:09:17,600 --> 00:09:21,997
before I knew anything that that
this is a good idea and then it became

167
00:09:22,080 --> 00:09:26,397
surprising that it worked because
I started reading those textbooks.

168
00:09:26,480 --> 00:09:29,997
>> Okay. Wait. So, okay. So, can you
talk through the intuition of why it was

169
00:09:30,080 --> 00:09:31,437
obvious to you if you remember?

170
00:09:31,520 --> 00:09:34,797
>> Well, okay. So, the intuition was it's
it's sort of like, you know, those

171
00:09:34,880 --> 00:09:37,437
people in the late 19th century who

172
00:09:37,520 --> 00:09:41,437
proved that heavier than than air flight

173
00:09:41,520 --> 00:09:45,597
was impossible, right? And of course,
you have birds, right? They do fly. And

174
00:09:45,680 --> 00:09:50,237
so, on the face of it, it it's obviously
wrong as an empirical question, right?

175
00:09:50,320 --> 00:09:54,397
And so, we have the same kind of thing
that, you know, the we know that the

176
00:09:54,480 --> 00:09:58,557
brain works. We don't know how but we
know it works and we know it's a large

177
00:09:58,640 --> 00:10:01,917
network of neurons in interaction
and that learning takes place by changing

178
00:10:02,000 --> 00:10:05,837
the connections. So kind of getting
this level of inspiration without covering

179
00:10:05,920 --> 00:10:11,837
the details but sort of trying
to derive basic principles

180
00:10:11,920 --> 00:10:15,677
you know that kind of gives you a clue
as to which direction to go. There's

181
00:10:15,760 --> 00:10:19,357
also the idea somehow that I' I've been
convinced of since I was in undergrad

182
00:10:19,440 --> 00:10:22,877
that even before that intelligence is

183
00:10:22,960 --> 00:10:25,277
inseparable from learning. So you the

184
00:10:25,360 --> 00:10:27,277
idea somehow that you can create an

185
00:10:27,360 --> 00:10:30,077
intelligent machine by basically

186
00:10:30,160 --> 00:10:33,597
programming for me was a non-starter you

187
00:10:33,680 --> 00:10:36,237
know from the start every intelligent

188
00:10:36,320 --> 00:10:40,717
entity that we know about arrives
at this intelligence through

189
00:10:40,800 --> 00:10:43,197
learning so learning you know machine

190
00:10:43,280 --> 00:10:46,397
learning was completely obvious path

191
00:10:46,480 --> 00:10:50,637
uh also because I'm lazy
so you know [laughter]

192
00:10:50,720 --> 00:10:54,477
>> you automate basically everything and
learning is the automation of >> exactly

193
00:10:54,560 --> 00:10:56,637
>> intelligence, >> right?

194
00:10:56,720 --> 00:10:59,677
>> So do you think so what is learning

195
00:10:59,760 --> 00:11:02,717
then? What falls under learning? Because

196
00:11:02,800 --> 00:11:06,397
do you think of reasoning is learning?

197
00:11:06,480 --> 00:11:08,957
>> Well, reasoning is certainly a a a

198
00:11:09,040 --> 00:11:11,597
consequence of of learning as well just

199
00:11:11,680 --> 00:11:14,397
like other functions of uh of the brain.

200
00:11:14,480 --> 00:11:19,197
The big question about reasoning is how
do you make reasoning compatible with

201
00:11:19,280 --> 00:11:22,077
gradient-based learning? >> Do you
think neural networks can be made

202
00:11:22,160 --> 00:11:25,837
to reason? Yes, there is no question
about that. Again, we have a good

203
00:11:25,920 --> 00:11:29,997
example, right? The question
is is how? So, the question

204
00:11:30,080 --> 00:11:34,237
is how much prior structure do you
have to put in the neural net so that

205
00:11:34,320 --> 00:11:38,557
something like human reasoning will
emerge from it you know from learning.

206
00:11:38,640 --> 00:11:41,117
Uh another question is all of our kind

207
00:11:41,200 --> 00:11:43,757
of model of what reasoning is that are

208
00:11:43,840 --> 00:11:49,437
based on logic are discrete and and and
are therefore incompatible with gradient

209
00:11:49,520 --> 00:11:52,157
based learning. Mhm. >> And I'm
a very strong believer in this

210
00:11:52,240 --> 00:11:57,117
idea gradient based learning. I don't
believe that uh other types of learning

211
00:11:57,200 --> 00:11:59,677
that don't use kind of gradient
information if you want.

212
00:11:59,760 --> 00:12:01,197
>> So you don't like discrete mathematics.

213
00:12:01,280 --> 00:12:02,797
You don't like anything discrete.

214
00:12:02,880 --> 00:12:05,917
>> Well, that's it's not that I don't
like it. It's just that it's it's

215
00:12:06,000 --> 00:12:07,917
incompatible with learning and I'm I'm a

216
00:12:08,000 --> 00:12:13,277
big fan of learning, right?
So in fact, that's perhaps one reason why

217
00:12:13,360 --> 00:12:15,997
deep learning has been kind of looked
at with suspicion by a lot of computer

218
00:12:16,080 --> 00:12:19,517
scientists because the math is very
different. the math that uh you use for

219
00:12:19,600 --> 00:12:21,357
deep learning, you know, kind of has

220
00:12:21,440 --> 00:12:24,237
more to do with, you know, cybernetics,

221
00:12:24,320 --> 00:12:28,477
uh the kind of math you do in electrical
engineering than the kind of math you do

222
00:12:28,560 --> 00:12:32,557
in computer science. [snorts] And and you
know, nothing in in machine learning

223
00:12:32,640 --> 00:12:34,477
is exact, right? Computer science is all

224
00:12:34,560 --> 00:12:39,597
about sort of, you know, obviously
compulsive attention to details of like,

225
00:12:39,680 --> 00:12:43,117
you know, every index has to be right
and you can prove that an algorithm is

226
00:12:43,200 --> 00:12:47,587
correct, right? Uh machine learning
is the science of sloppiness. Really?

227
00:12:47,670 --> 00:12:49,690
[laughter]

228
00:12:49,920 --> 00:12:53,917
>> That's beautiful. So okay, maybe let's

229
00:12:54,000 --> 00:12:56,557
feel around in the dark of what is a

230
00:12:56,640 --> 00:13:00,397
neural network that reasons or a system

231
00:13:00,480 --> 00:13:02,077
that

232
00:13:02,160 --> 00:13:05,037
is works with continuous functions

233
00:13:05,120 --> 00:13:09,597
that's able to do uh build knowledge.

234
00:13:09,680 --> 00:13:11,917
However we think about reasoning, build

235
00:13:12,000 --> 00:13:14,397
on previous knowledge, build on extra

236
00:13:14,480 --> 00:13:16,717
knowledge, create new knowledge,

237
00:13:16,800 --> 00:13:19,437
generalize outside of any training set

238
00:13:19,520 --> 00:13:22,717
ever built. What does that look like? If

239
00:13:22,800 --> 00:13:25,517
uh yeah, maybe do you have inklings of

240
00:13:25,600 --> 00:13:27,517
thoughts of what that might look like?

241
00:13:27,600 --> 00:13:31,917
>> Well, yeah. I mean, yes and no. If I
had precise ideas about this, I think, you

242
00:13:32,000 --> 00:13:36,157
know, we'd be building it right now. But
and there are people working on this or

243
00:13:36,240 --> 00:13:38,477
whose main research interest is actually

244
00:13:38,560 --> 00:13:43,197
exactly that right. So what you need to
have is a working memory. So you need to

245
00:13:43,280 --> 00:13:44,877
have

246
00:13:44,960 --> 00:13:47,357
some device if you want uh some

247
00:13:47,440 --> 00:13:51,197
subsystem that can store uh a relatively

248
00:13:51,280 --> 00:13:53,757
large number of factual episodic

249
00:13:53,840 --> 00:13:57,117
information for you know a reasonable

250
00:13:57,200 --> 00:14:00,077
amount of time. So you you know in the

251
00:14:00,160 --> 00:14:01,517
in the brain for example there are kind

252
00:14:01,600 --> 00:14:04,157
of three main types of memory. one is

253
00:14:04,240 --> 00:14:09,037
the the sort of memory
of the the the state of your

254
00:14:09,120 --> 00:14:12,637
cortex and that sort
of disappears within 20

255
00:14:12,720 --> 00:14:16,477
seconds. You can't remember things
for more than about 20 seconds or a minute

256
00:14:16,560 --> 00:14:20,157
if if if you don't have any other form
of memory. Uh [snorts] the second type

257
00:14:20,240 --> 00:14:23,757
of memory which is longer term but still
shortterm is the hippoc campus. So you

258
00:14:23,840 --> 00:14:25,517
can you know you came into this building

259
00:14:25,600 --> 00:14:27,597
you remember where the where the the

260
00:14:27,680 --> 00:14:31,037
exit is where the uh elevators are. Um,

261
00:14:31,120 --> 00:14:34,717
you have some map of that building
that's stored in your hippoc campus.

262
00:14:34,800 --> 00:14:38,477
>> You might remember something about
what I said, you know, a few minutes ago.

263
00:14:38,560 --> 00:14:41,917
>> I forgot it all already. >> It's
been erased, but you know, but that

264
00:14:42,000 --> 00:14:44,397
that would be in your
in your hypoc campus.

265
00:14:44,480 --> 00:14:46,237
>> Uh, and then the the longer term memory

266
00:14:46,320 --> 00:14:49,517
is in the synapse, the synapses, right?

267
00:14:49,600 --> 00:14:52,877
>> Um, so what you need if you want
a system that's capable of reasoning is

268
00:14:52,960 --> 00:14:57,277
that you want the hypocampus
like thing, right?

269
00:14:57,360 --> 00:15:00,637
And [snorts] that's what people have
tried to do with memory networks and you

270
00:15:00,720 --> 00:15:02,637
know neural machines and stuff like that

271
00:15:02,720 --> 00:15:07,677
right and and now with transformers
which have sort of a a memory in their

272
00:15:07,760 --> 00:15:12,797
kind of self attention system you
can you can think of it this way so

273
00:15:12,880 --> 00:15:16,957
so that's one element you need another
thing you need is some sort of network

274
00:15:17,040 --> 00:15:19,037
that can

275
00:15:19,120 --> 00:15:22,237
access this memory

276
00:15:22,320 --> 00:15:26,317
get an information back
and then kind of crunch on it and

277
00:15:26,400 --> 00:15:28,077
do this iteratively multiple times

278
00:15:28,160 --> 00:15:31,037
because a chain of reasoning

279
00:15:31,120 --> 00:15:34,717
is is a process by which you you you

280
00:15:34,800 --> 00:15:38,637
kind of you update your knowledge about
the state of the world about you know

281
00:15:38,720 --> 00:15:42,957
what's going to happen etc and that
that has to be this sort of recurrent

282
00:15:43,040 --> 00:15:46,637
operation basically >> and you
think that kind of if if we

283
00:15:46,720 --> 00:15:48,877
think about a transformer so that seems

284
00:15:48,960 --> 00:15:54,077
to be too small to contain the knowledge
that's that's uh to represent the

285
00:15:54,160 --> 00:15:56,317
knowledge that's contained
in Wikipedia media for example

286
00:15:56,400 --> 00:15:59,997
>> but transformer doesn't have this idea
of recurrence it's got a fixed number of

287
00:16:00,080 --> 00:16:03,597
layers and that's the number of steps
that you know limits basically it

288
00:16:03,680 --> 00:16:07,197
representation >>
but recurrence would build on the

289
00:16:07,280 --> 00:16:09,197
knowledge somehow I mean it would it

290
00:16:09,280 --> 00:16:13,277
would evolve the knowledge and expand

291
00:16:13,360 --> 00:16:17,597
the amount of information perhaps or
useful information within that knowledge

292
00:16:17,680 --> 00:16:22,557
>> but is is this something that just
can emerge with size because it seems like

293
00:16:22,640 --> 00:16:26,637
everything we have now is too small no
it's not It's not it's not clear I mean

294
00:16:26,720 --> 00:16:30,957
how how you access and write into
an associative memory in efficient way. I

295
00:16:31,040 --> 00:16:34,157
mean sort of the original memory network
maybe had something like the right

296
00:16:34,240 --> 00:16:38,637
architecture but uh if you try to scale
up a memory network so that the memory

297
00:16:38,720 --> 00:16:42,237
contains all Wikipedia it
doesn't quite work >> right

298
00:16:42,320 --> 00:16:46,397
>> so so there's there's a need for new
ideas there okay but it's not the only

299
00:16:46,480 --> 00:16:50,077
form of reasoning so there's another
form of reasoning which is through which

300
00:16:50,160 --> 00:16:52,877
is very classical also in some types of

301
00:16:52,960 --> 00:16:56,877
AI uh and it's based on u let's call it

302
00:16:56,960 --> 00:17:00,717
energy minimization okay so you have uh

303
00:17:00,800 --> 00:17:02,637
some sort of objective, some energy

304
00:17:02,720 --> 00:17:06,957
function that represents the the the um

305
00:17:07,040 --> 00:17:09,917
the quality or the negative quality.

306
00:17:10,000 --> 00:17:14,557
Okay, energy goes up when things get bad
and they get low when things get good.

307
00:17:14,640 --> 00:17:16,877
So, let's say you you want to figure

308
00:17:16,960 --> 00:17:18,797
out, you know, what gestures do I need

309
00:17:18,880 --> 00:17:23,677
to to do to grab an object or walk out

310
00:17:23,760 --> 00:17:27,117
the door. [snorts] If you have
a good model of uh your own

311
00:17:27,200 --> 00:17:31,437
body, a good model of the environment
using this kind of energy minimization,

312
00:17:31,520 --> 00:17:33,357
you can make a you can make you can do

313
00:17:33,440 --> 00:17:38,557
planning and it's uh in optimal control
is called it's called model model

314
00:17:38,640 --> 00:17:41,437
predictive control. You have a model of
what's going to happen in the world as a

315
00:17:41,520 --> 00:17:45,757
consequence of your actions and that
allows you to by energy minimization

316
00:17:45,840 --> 00:17:48,477
figure out a sequence of action
that optimizes a particular objective

317
00:17:48,560 --> 00:17:51,517
function which measures you know
minimizes number of times you're going

318
00:17:51,600 --> 00:17:55,437
to hit something and the energy you're
going to spend doing the gesture and

319
00:17:55,520 --> 00:17:59,597
etc. So so that's a form of reasoning.

320
00:17:59,680 --> 00:18:00,877
Planning is a form of reasoning. And

321
00:18:00,960 --> 00:18:03,837
perhaps what led to the ability of

322
00:18:03,920 --> 00:18:07,837
humans to reason is the fact that or you

323
00:18:07,920 --> 00:18:10,397
know species you know that appear before

324
00:18:10,480 --> 00:18:12,557
us had to do some sort of planning to be

325
00:18:12,640 --> 00:18:17,677
able to hunt and survive and survive
the winter in particular and so you know

326
00:18:17,760 --> 00:18:19,277
it's the same capacity that that you

327
00:18:19,360 --> 00:18:24,637
need to have. So in your intuition is um

328
00:18:24,720 --> 00:18:27,197
if we look at expert systems and

329
00:18:27,280 --> 00:18:30,637
encoding knowledge as logic systems

330
00:18:30,720 --> 00:18:34,477
as graphs or in this kind of way is not

331
00:18:34,560 --> 00:18:37,437
a useful way to think about knowledge.

332
00:18:37,520 --> 00:18:40,237
>> Graphs are a little brittle or or logic

333
00:18:40,320 --> 00:18:45,437
uh representation. So basically you know
variables that that have values and then

334
00:18:45,520 --> 00:18:48,877
constraint between them that are
represented by rules uh is a little too

335
00:18:48,960 --> 00:18:52,717
rigid and too brittle right. So one
of the you know some of the early efforts

336
00:18:52,800 --> 00:18:54,717
in that respect

337
00:18:54,800 --> 00:18:57,997
um were were to put probabilities on

338
00:18:58,080 --> 00:19:01,997
them. So a rule you know you know if you
have this and that symptom you know you

339
00:19:02,080 --> 00:19:05,277
have this uh disease with that
probability and you should prescribe

340
00:19:05,360 --> 00:19:07,437
that antibiotic with that probability

341
00:19:07,520 --> 00:19:09,677
right this the m system from the from

342
00:19:09,760 --> 00:19:13,037
the 70s um and that that's what that

343
00:19:13,120 --> 00:19:16,317
branch of AI led to uh you know basics

344
00:19:16,400 --> 00:19:19,037
and graphical models and causal

345
00:19:19,120 --> 00:19:21,437
inference and viral you know

346
00:19:21,520 --> 00:19:26,237
>> method so so there
there is I mean certainly

347
00:19:26,320 --> 00:19:28,157
a lot of interesting work going on in

348
00:19:28,240 --> 00:19:29,757
this area. The main issue with this is

349
00:19:29,840 --> 00:19:36,077
is knowledge acquisition. How do you
uh reduce a bunch of data to a graph of

350
00:19:36,160 --> 00:19:39,357
this type? >> Yeah.
It relies on the expert to on a

351
00:19:39,440 --> 00:19:42,157
human being to encode to add knowledge

352
00:19:42,240 --> 00:19:44,317
>> and that's essentially impractical.

353
00:19:44,400 --> 00:19:47,757
>> Yeah, it's not scalable. >> That's
that's a big question. The second

354
00:19:47,840 --> 00:19:52,637
question is do you want to represent
knowledge as symbols and do you want to

355
00:19:52,720 --> 00:19:54,717
manipulate them with logic? And again

356
00:19:54,800 --> 00:19:58,557
that's incompatible with learning. So uh

357
00:19:58,640 --> 00:20:00,637
one suggestion which Jeffington has been

358
00:20:00,720 --> 00:20:03,117
advocating for many decades is replace

359
00:20:03,200 --> 00:20:07,117
symbols by uh vectors. Think of it as

360
00:20:07,200 --> 00:20:09,597
pattern of activities in a bunch of

361
00:20:09,680 --> 00:20:14,797
neurons or units or whatever you
want to call them and replace logic by

362
00:20:14,880 --> 00:20:16,717
continuous functions. >> Mhm.

363
00:20:16,800 --> 00:20:19,037
>> Okay. And that becomes now compatible.

364
00:20:19,120 --> 00:20:23,117
There's a very good set of ideas by uh

365
00:20:23,200 --> 00:20:25,197
written in a paper about 10 years ago by

366
00:20:25,280 --> 00:20:29,837
Leon Bou on who is here at Facebook. Um

367
00:20:29,920 --> 00:20:32,957
um the title of the paper is from
machine learning to machine reasoning

368
00:20:33,040 --> 00:20:37,197
and his idea is that uh a learning
learning system should be able to

369
00:20:37,280 --> 00:20:41,197
manipulate objects that are in the same
spa in a space and then put the result

370
00:20:41,280 --> 00:20:45,437
back in the same space. So it's this idea
of working memory basically [snorts]

371
00:20:45,520 --> 00:20:47,757
and it's uh it's very enlightening

372
00:20:47,840 --> 00:20:50,477
>> and in a sense that might learn

373
00:20:50,560 --> 00:20:53,277
something like the simple expert

374
00:20:53,360 --> 00:20:57,837
systems. I mean it's
you can learn basic logic

375
00:20:57,920 --> 00:21:00,797
operations there. >> Yeah,
quite possibly. Yeah, there's a

376
00:21:00,880 --> 00:21:04,797
big debate on sort of how much prior
structure you have to put in for this

377
00:21:04,880 --> 00:21:08,077
kind of stuff to emerge. That's
the debate I have with Gary Marcus and

378
00:21:08,160 --> 00:21:12,717
people like that. >> Yeah. Yeah.
So and the other person so I

379
00:21:12,800 --> 00:21:14,637
just talked to Judea Pearl

380
00:21:14,720 --> 00:21:16,397
>> from you mentioned causal inference

381
00:21:16,480 --> 00:21:20,397
world. So his worry is that the current

382
00:21:20,480 --> 00:21:25,837
neural networks are not able to learn

383
00:21:25,920 --> 00:21:29,917
what causes what causal
inference between things.

384
00:21:30,000 --> 00:21:32,077
>> So I think I think he's right and wrong

385
00:21:32,160 --> 00:21:34,717
about this. uh if he's talking about the

386
00:21:34,800 --> 00:21:39,197
sort of classic uh type
of neural nets, people sort of

387
00:21:39,280 --> 00:21:42,317
didn't worry too much about this,
but there's a lot of people now working on

388
00:21:42,400 --> 00:21:45,757
calls on inference and there's a paper
that just came out last week by Leonu

389
00:21:45,840 --> 00:21:47,517
among others, David Lopez and a bunch of

390
00:21:47,600 --> 00:21:51,197
other people exactly on that problem of

391
00:21:51,280 --> 00:21:54,237
um how do you kind of you know get a

392
00:21:54,320 --> 00:21:56,957
neural net to sort of pay attention to

393
00:21:57,040 --> 00:21:59,597
real causal relationships which may also

394
00:21:59,680 --> 00:22:04,317
solve issues of bias.
in data and things like

395
00:22:04,400 --> 00:22:06,957
this. So >> I' I'd like
to read that paper because

396
00:22:07,040 --> 00:22:09,037
uh that ultimately the challenge there

397
00:22:09,120 --> 00:22:12,557
is also seems to fall back on the human

398
00:22:12,640 --> 00:22:17,677
expert to ultimately decide causality

399
00:22:17,760 --> 00:22:20,397
between things. >> People are
not very good at establishing

400
00:22:20,480 --> 00:22:24,477
causality first of all. So first of all
you talk to physicists and physicists

401
00:22:24,560 --> 00:22:27,677
actually don't believe
in causality because look at the

402
00:22:27,760 --> 00:22:32,637
>> all the basic laws of microfysics are
time reversible. So there is no causality.

403
00:22:32,720 --> 00:22:34,317
>> The arrow of time is not real.

404
00:22:34,400 --> 00:22:35,917
>> It's it's as soon as you start looking

405
00:22:36,000 --> 00:22:38,077
at microscopic systems where there is

406
00:22:38,160 --> 00:22:43,037
unpredictable randomness where there
is clearly an arrow of time but it's a big

407
00:22:43,120 --> 00:22:46,237
mystery in physics actually
how that emerges. Um

408
00:22:46,320 --> 00:22:49,437
>> is it emergent or is it part of the

409
00:22:49,520 --> 00:22:54,397
fundamental fabric of reality? Yeah. Or
is it a bias of intelligent systems that

410
00:22:54,480 --> 00:22:57,917
you know because of the second law
of thermodynamics we perceive a particular

411
00:22:58,000 --> 00:23:02,237
hour of time but in fact
it's kind of arbitrary right

412
00:23:02,320 --> 00:23:06,557
>> so yeah physicists mathematicians they
don't care about I mean the math doesn't

413
00:23:06,640 --> 00:23:10,237
care about the flow of time >> well
certainly certainly micro physics

414
00:23:10,320 --> 00:23:12,477
doesn't people themselves are not very

415
00:23:12,560 --> 00:23:15,597
good at establishing causal causal

416
00:23:15,680 --> 00:23:17,997
relationships if you ask this I think it

417
00:23:18,080 --> 00:23:21,757
was in one of simar paper's book on on

418
00:23:21,840 --> 00:23:26,397
um like children learning you know he
studied with Jean Pia you know he's the

419
00:23:26,480 --> 00:23:29,757
guy who co-authored [snorts] the book
perceptron with Marvin Minsky that kind

420
00:23:29,840 --> 00:23:31,357
of killed the first wave of neural nets

421
00:23:31,440 --> 00:23:34,077
but but he was actually a a learning

422
00:23:34,160 --> 00:23:39,117
person he uh in the sense of studying
learning in humans and and machines

423
00:23:39,200 --> 00:23:40,077
that's why he got interested in

424
00:23:40,160 --> 00:23:43,917
perceptron and uh and he he he wrote

425
00:23:44,000 --> 00:23:47,277
that if you ask a little kid uh about

426
00:23:47,360 --> 00:23:51,037
what is the cause of of the wind a lot

427
00:23:51,120 --> 00:23:54,317
of kids will say they will think for
a while and they'll say oh it's the the

428
00:23:54,400 --> 00:23:57,837
branches in the trees they move and
that creates wind right so they get the

429
00:23:57,920 --> 00:24:01,517
causal relationship backwards and it's
because their understanding of the world

430
00:24:01,600 --> 00:24:04,397
and intuitive physics is not that great
right I mean these are like you know

431
00:24:04,480 --> 00:24:08,317
four or five year old kids >> you
know it gets better and then you

432
00:24:08,400 --> 00:24:11,117
understand that this it can't be right

433
00:24:11,200 --> 00:24:14,637
but there are many things which we can

434
00:24:14,720 --> 00:24:19,357
because of our common sense
understanding of things what people call

435
00:24:19,440 --> 00:24:23,837
common sense >> and with
understanding of physics we can

436
00:24:23,920 --> 00:24:26,877
>> there's a lot of stuff that we can
figure out causality even with diseases

437
00:24:26,960 --> 00:24:30,717
we can figure out what's not causing what

438
00:24:30,800 --> 00:24:33,197
>> often there's a lot
of mystery of course

439
00:24:33,280 --> 00:24:37,757
but the idea is that you should be able
to encode that into systems it seems

440
00:24:37,840 --> 00:24:39,917
unlikely they'd be able
to figure that out themselves

441
00:24:40,000 --> 00:24:43,277
>> well whenever we can do intervention
but you know all of humanity has been

442
00:24:43,360 --> 00:24:45,517
completely deluded for millennia

443
00:24:45,600 --> 00:24:50,797
probably since existence about a very
very wrong causal relationship where

444
00:24:50,880 --> 00:24:55,837
whatever you can explain you attribute it
to you know some deity some divinity right

445
00:24:55,920 --> 00:24:59,197
>> um and that's a copout that's a way
of saying like I don't know the cause so

446
00:24:59,280 --> 00:25:01,517
you know God did it right so you

447
00:25:01,600 --> 00:25:05,037
mentioned Marvin Minsky and uh the irony

448
00:25:05,120 --> 00:25:07,277
of um

449
00:25:07,360 --> 00:25:10,477
uh you know maybe causing the first AI

450
00:25:10,560 --> 00:25:15,437
winter you were there in the '9s you
were there in the 80s of course uh in

451
00:25:15,520 --> 00:25:17,437
the '9s why do you think people lost

452
00:25:17,520 --> 00:25:22,637
faith in deep learning in the '9s
and found it again a decade later, over a

453
00:25:22,720 --> 00:25:24,637
decade later. >> Yeah,
it wasn't called deep learning

454
00:25:24,720 --> 00:25:28,717
yet. It was just called
neural nets, but um

455
00:25:28,800 --> 00:25:33,997
>> yeah, they lost interest. I mean,
I think I would put that around 1995, at

456
00:25:34,080 --> 00:25:35,277
least the machine learning community.

457
00:25:35,360 --> 00:25:39,357
There was always a neural net
community, but it but became

458
00:25:39,440 --> 00:25:41,837
kind of disconnected from sort of

459
00:25:41,920 --> 00:25:45,357
mainstream machine learning if you want.

460
00:25:45,440 --> 00:25:49,997
There were it was basically electrical
engineering that kept at it, >> right?

461
00:25:50,080 --> 00:25:52,797
>> Uh and computer science >> just gave up

462
00:25:52,880 --> 00:25:56,077
>> gave up on uh on neural nets. I don't I

463
00:25:56,160 --> 00:25:57,517
don't know, you know, I was too close to

464
00:25:57,600 --> 00:26:00,077
it to really

465
00:26:00,160 --> 00:26:03,357
sort of analyze it with sort of a

466
00:26:03,440 --> 00:26:07,197
unbiased eye if you want. But I I
would I would I would I would make a few

467
00:26:07,280 --> 00:26:10,317
guesses. So the first one is at the time

468
00:26:10,400 --> 00:26:12,397
neural nets

469
00:26:12,480 --> 00:26:14,957
were it was very hard to make them work

470
00:26:15,040 --> 00:26:18,077
in the sense that you would you know

471
00:26:18,160 --> 00:26:20,397
implement back prop in your favorite

472
00:26:20,480 --> 00:26:25,517
language um and that favorite language
was not Python it was not matlab it was

473
00:26:25,600 --> 00:26:26,877
not any of those things because they

474
00:26:26,960 --> 00:26:29,197
didn't exist right you had to write it

475
00:26:29,280 --> 00:26:33,437
in forc or something like this right um

476
00:26:33,520 --> 00:26:37,357
so you would uh experiment with it
you would probably make some very basic

477
00:26:37,440 --> 00:26:40,957
mistakes takes like you know badly
initialize your weights make the network

478
00:26:41,040 --> 00:26:43,277
too small because you read in the
textbook you know you don't want too

479
00:26:43,360 --> 00:26:46,637
many parameters right and of course you
know and you would train on exor because

480
00:26:46,720 --> 00:26:49,997
you didn't have any other data set
to train it on and of course you know it

481
00:26:50,080 --> 00:26:52,797
works half the time so you' say I give

482
00:26:52,880 --> 00:26:57,837
up also you train it with batch gradient
which you know isn't efficient so

483
00:26:57,920 --> 00:27:01,437
there's a lot of there's bag of tricks
that you had to know to make those

484
00:27:01,520 --> 00:27:03,997
things work or you you had to reinvent

485
00:27:04,080 --> 00:27:08,557
and a lot of people just didn't and they
just couldn't make it work. Um, so

486
00:27:08,640 --> 00:27:10,717
that's one thing. The investment in

487
00:27:10,800 --> 00:27:15,837
software platform to be able to kind of,
you know, display things, figure out why

488
00:27:15,920 --> 00:27:19,197
things don't work, kind of get a good
intuition for how to get them to work,

489
00:27:19,280 --> 00:27:22,637
have enough flexibility so you can create,
you know, network architectures

490
00:27:22,720 --> 00:27:26,717
like convolutional nets and stuff like
that. It was hard. I mean, you had to

491
00:27:26,800 --> 00:27:29,597
write everything from scratch and again
you didn't have any Python or MATLAB or

492
00:27:29,680 --> 00:27:32,957
anything, right? >> [snorts] >> I
read that, sorry to interrupt, but I

493
00:27:33,040 --> 00:27:35,997
read that you wrote in in lisp the your

494
00:27:36,080 --> 00:27:40,477
first uh versions of uh Lynette with
convolutional networks, which by the

495
00:27:40,560 --> 00:27:42,557
way, one of my favorite languages.

496
00:27:42,640 --> 00:27:46,877
>> That's how I knew you were legit.
Uh touring award, whatever. This what you

497
00:27:46,960 --> 00:27:50,557
programmed in Lisp. That's >> it's
still my favorite language. But

498
00:27:50,640 --> 00:27:53,597
>> it's not that we programmed in Lisp,
it's that we had to write a list

499
00:27:53,680 --> 00:27:58,317
interpreter. >> Okay? Because it's
not like we used one that existed. So

500
00:27:58,400 --> 00:28:02,797
>> we wrote a list interpreter that we
hooked up to you know a backend library

501
00:28:02,880 --> 00:28:07,117
that we wrote also for sort of neural
net computation. And then after a few

502
00:28:07,200 --> 00:28:12,317
years around 1991 we invented this idea
of uh basically having modules that know

503
00:28:12,400 --> 00:28:15,197
how to forward propagate and back
propagate gradients and then

504
00:28:15,280 --> 00:28:17,037
interconnecting those modules in a

505
00:28:17,120 --> 00:28:22,637
graph. [snorts] Leotu had made proposals
on this about this in the late 80s and

506
00:28:22,720 --> 00:28:26,557
we're able to implement this using
a list system. Eventually we wanted to use

507
00:28:26,640 --> 00:28:30,637
that system to make build
production code for character

508
00:28:30,720 --> 00:28:33,597
recognition at Bell Lab. So we
actually wrote a compiler for that list

509
00:28:33,680 --> 00:28:37,837
interpreter so that Patricy Mart who's
now Microsoft kind of did did the bulk

510
00:28:37,920 --> 00:28:40,557
of it with Leon and and me and uh and so

511
00:28:40,640 --> 00:28:45,037
we could write our system in list and
then compile to C and then we'll have a

512
00:28:45,120 --> 00:28:47,277
self-contained complete system that

513
00:28:47,360 --> 00:28:50,557
could kind of do the entire thing.

514
00:28:50,640 --> 00:28:55,277
Neither PyTorch nor transform can
do this today yet. Okay, it's coming.

515
00:28:55,360 --> 00:28:58,797
>> Yeah, >> I mean there's
something like that in

516
00:28:58,880 --> 00:29:01,757
PyTorch called you know to script

517
00:29:01,840 --> 00:29:03,757
>> and so you know we
had to write all this

518
00:29:03,840 --> 00:29:07,437
interpreter. to this compiler. where
to invest a huge amount of effort to do

519
00:29:07,520 --> 00:29:12,477
this and not everybody if if you don't
completely believe in the concept you're

520
00:29:12,560 --> 00:29:14,397
not going to invest
the time to do this right

521
00:29:14,480 --> 00:29:17,677
>> now at the time also you know or today

522
00:29:17,760 --> 00:29:21,517
>> this would turn into torture pietor
torture tensorflow or whatever we put it

523
00:29:21,600 --> 00:29:22,957
in open source everybody would use it

524
00:29:23,040 --> 00:29:25,597
and you know realize it's good back

525
00:29:25,680 --> 00:29:28,957
before 1995 working at AT&T there's no

526
00:29:29,040 --> 00:29:31,357
way the lawyers would let you release

527
00:29:31,440 --> 00:29:36,477
anything in open source of this nature
And so we could not distribute our code

528
00:29:36,560 --> 00:29:39,997
really. >> And on that point
and sorry to go on a

529
00:29:40,080 --> 00:29:42,237
million tangents but on that point I

530
00:29:42,320 --> 00:29:47,837
also read that there was some almost
pat like a patent on convolution networks

531
00:29:47,920 --> 00:29:51,037
labs. >> [sighs and gasps] >> So that

532
00:29:51,120 --> 00:29:55,117
>> first of all I mean just >>
there's two actually [laughter]

533
00:29:55,200 --> 00:29:57,197
>> uh that ran out uh the that

534
00:29:57,280 --> 00:29:59,197
>> thankfully in 2007

535
00:29:59,280 --> 00:30:03,917
>> in 2007 I what

536
00:30:04,000 --> 00:30:07,357
can we can we just talk about that for I
know you're a Facebook but you're also

537
00:30:07,440 --> 00:30:09,917
at NYU and and

538
00:30:10,000 --> 00:30:12,877
what does it mean to patent ideas like

539
00:30:12,960 --> 00:30:14,397
these

540
00:30:14,480 --> 00:30:20,077
software ideas essentially or what
or mathematical ical ideas or what are

541
00:30:20,160 --> 00:30:22,877
they? >> Okay, so they're
not mathematical ideas.

542
00:30:22,960 --> 00:30:24,797
There are, you know, algorithms and

543
00:30:24,880 --> 00:30:27,917
there was a period where the US patent

544
00:30:28,000 --> 00:30:30,637
office would allow the patent of

545
00:30:30,720 --> 00:30:35,517
software as long as it was embodied.
Uh the Europeans are very different. They

546
00:30:35,600 --> 00:30:40,077
don't they don't quite accept that. They
have a different concept. But you know,

547
00:30:40,160 --> 00:30:43,677
I don't I no longer I mean I never
actually strongly believed in this, but

548
00:30:43,760 --> 00:30:45,117
I I don't believe in this kind of

549
00:30:45,200 --> 00:30:51,000
patent. Facebook basically doesn't
believe in this kind of patent.

550
00:30:51,280 --> 00:30:55,037
Google fights patents because they've been

551
00:30:55,120 --> 00:30:59,277
burned with Apple. And so now they do
this for defensive purpose. But usually

552
00:30:59,360 --> 00:31:02,157
they say we're not going
to sue you if you infringe.

553
00:31:02,240 --> 00:31:04,397
>> Facebook has a has a similar policy.

554
00:31:04,480 --> 00:31:08,077
They say, you know, we patent on certain
things for defensive purpose. We're not

555
00:31:08,160 --> 00:31:09,677
going to see you if you infringe unless

556
00:31:09,760 --> 00:31:15,117
you sue us. Um so the the industry does

557
00:31:15,200 --> 00:31:17,117
not believe in in patents. They're there

558
00:31:17,200 --> 00:31:22,557
because of you know the legal landscape
and and and various things but but I

559
00:31:22,640 --> 00:31:24,877
don't really believe in patents
for this kind of stuff.

560
00:31:24,960 --> 00:31:27,517
>> Okay. So that's that's
a great thing. So I I

561
00:31:27,600 --> 00:31:29,197
>> I'll tell you a worst
story actually. So

562
00:31:29,280 --> 00:31:33,517
what happens was the the first the first
patent about convolutional net was about

563
00:31:33,600 --> 00:31:35,997
kind of the early version
of convolutional net that didn't have

564
00:31:36,080 --> 00:31:40,717
separate pooling layers. It had
convolutional layers which tried more

565
00:31:40,800 --> 00:31:44,877
than one if you want right >>
and then there was a second one on

566
00:31:44,960 --> 00:31:46,877
convolutional nets with separate pooling

567
00:31:46,960 --> 00:31:50,077
layers uh train with back prop and they

568
00:31:50,160 --> 00:31:52,637
were filed filed in ' 89 and 1990

569
00:31:52,720 --> 00:31:57,757
something like this at the time the life
life of a patent was 17 years so here's

570
00:31:57,840 --> 00:32:01,757
what what happened over the next few
years is that we started developing um

571
00:32:01,840 --> 00:32:04,317
character recognition technology around

572
00:32:04,400 --> 00:32:09,277
conviction nets [snorts] and uh in 1994

573
00:32:09,360 --> 00:32:11,837
a check reading system was deployed in

574
00:32:11,920 --> 00:32:15,357
ATM machines. In 1995 it was for large

575
00:32:15,440 --> 00:32:20,637
check reading machines in back
offices etc. And those systems were

576
00:32:20,720 --> 00:32:24,237
developed by an engineering group that
we were collaborating with at AT&T and

577
00:32:24,320 --> 00:32:26,157
they were commercialized by NCR which at

578
00:32:26,240 --> 00:32:28,717
the time was a subsidiary of AT&T. Mhm.

579
00:32:28,800 --> 00:32:33,357
>> Now, AT&T split up in 1996,

580
00:32:33,440 --> 00:32:37,837
199 early 1996 and the lawyers just
looked at all the patents and they

581
00:32:37,920 --> 00:32:42,077
distributed the patents among
the various companies. They gave the the

582
00:32:42,160 --> 00:32:45,517
commercial net patent to NCR because
they were actually selling products that

583
00:32:45,600 --> 00:32:49,517
used it. But nobody at NCR had
any idea what a commercial net was.

584
00:32:49,600 --> 00:32:50,477
>> Yeah.

585
00:32:50,560 --> 00:32:55,037
>> Okay. So between 1996 and 2007,

586
00:32:55,120 --> 00:32:58,877
so there's a whole period until 2002
where I didn't actually work on machine

587
00:32:58,960 --> 00:33:02,957
learning or control it.
I resumed working on this around 2002

588
00:33:03,040 --> 00:33:07,037
and between 2002 and 2007 I was working
on them crossing my finger that nobody

589
00:33:07,120 --> 00:33:09,277
at NCR would notice and nobody noticed.

590
00:33:09,360 --> 00:33:12,157
>> Yeah. And I and I
hope that this kind of

591
00:33:12,240 --> 00:33:15,517
somewhat as you said lawyers aside

592
00:33:15,600 --> 00:33:18,797
relative openness of the community now

593
00:33:18,880 --> 00:33:21,277
will continue. It accelerates the entire

594
00:33:21,360 --> 00:33:24,957
progress of of the industry. And

595
00:33:25,040 --> 00:33:29,357
you know the problems that Facebook
and Google and others are

596
00:33:29,440 --> 00:33:34,237
facing today is not whether Facebook or
Google or Microsoft or IBM or whoever is

597
00:33:34,320 --> 00:33:37,357
ahead of the other. It's that we don't
have the technology to build the things

598
00:33:37,440 --> 00:33:40,717
we want to build. We want to build
intelligent virtual assistants that have

599
00:33:40,800 --> 00:33:44,557
common sense. We don't have monopoly on
good ideas for this. We don't believe we

600
00:33:44,640 --> 00:33:47,117
do. Maybe others do believe they do, but

601
00:33:47,200 --> 00:33:52,717
we don't. Okay. If a startup tells you
they have the secret to, you know, human

602
00:33:52,800 --> 00:33:54,077
level intelligence and common sense,

603
00:33:54,160 --> 00:33:56,477
don't believe them. They don't. And it's

604
00:33:56,560 --> 00:34:00,397
going to take the entire uh work of the

605
00:34:00,480 --> 00:34:04,717
world research community for a while
to get to the point where you can go off

606
00:34:04,800 --> 00:34:08,077
and in each of those companies kind of
start to build things on this. We're not

607
00:34:08,160 --> 00:34:09,037
there yet.

608
00:34:09,120 --> 00:34:11,117
>> Absolutely. And this this calls to the

609
00:34:11,200 --> 00:34:15,197
the gap between the space of ideas and

610
00:34:15,280 --> 00:34:19,757
the rigorous testing of those ideas
of practical application that you often

611
00:34:19,840 --> 00:34:24,157
speak to. You've uh written advice
saying don't get fooled by people who

612
00:34:24,240 --> 00:34:28,317
claim to have a solution to artificial
general intelligence who claim to have

613
00:34:28,400 --> 00:34:30,877
an AI system that work just like the

614
00:34:30,960 --> 00:34:32,637
human brain or who claim to have figured

615
00:34:32,720 --> 00:34:35,917
out how the brain works. Ask them what

616
00:34:36,000 --> 00:34:40,557
the error rate they get
on emnest or imageet. Yeah,

617
00:34:40,640 --> 00:34:42,637
>> this is a little dated
by the way [laughter]

618
00:34:42,720 --> 00:34:45,517
>> 2000 I mean five years who's counting

619
00:34:45,600 --> 00:34:47,997
okay but I think your opinion is still

620
00:34:48,080 --> 00:34:52,077
emnest and imageet uh yes maybe data

621
00:34:52,160 --> 00:34:54,077
there may be new benchmarks right but I

622
00:34:54,160 --> 00:34:56,797
think that philosophy is one you still

623
00:34:56,880 --> 00:35:00,397
and and somewhat hold that benchmarks

624
00:35:00,480 --> 00:35:04,077
and the practical testing the practical
application is where you really get to

625
00:35:04,160 --> 00:35:06,877
test the ideas >> well it may
not be completely practical

626
00:35:06,960 --> 00:35:10,557
like for examp you know it could be
a toy data set or but but it to be some

627
00:35:10,640 --> 00:35:15,197
sort of task that the community
as a whole has accepted as some sort of

628
00:35:15,280 --> 00:35:18,797
standard you know kind of benchmark if
you want it doesn't need to be real. So

629
00:35:18,880 --> 00:35:23,837
for example many years ago here at fair
um people you know chisen west on auto

630
00:35:23,920 --> 00:35:27,597
and board and a few others proposed
the the baby tasks which were kind of a a

631
00:35:27,680 --> 00:35:30,477
toy problem to test the ability of

632
00:35:30,560 --> 00:35:34,797
machines to reason actually to access
working memory and things like this and

633
00:35:34,880 --> 00:35:38,557
uh it was very useful even though it
wasn't a real task. amnest is kind of

634
00:35:38,640 --> 00:35:41,757
halfway a real task. Uh so you know toy

635
00:35:41,840 --> 00:35:46,157
problems can be very useful. It's just
that I was really struck by the fact

636
00:35:46,240 --> 00:35:49,997
that a lot of people particularly a lot
of people with money to invest would be

637
00:35:50,080 --> 00:35:54,717
fooled by people telling them oh we have
you know the algorithm of the cortex and

638
00:35:54,800 --> 00:35:56,557
you should give us 50 million.

639
00:35:56,640 --> 00:36:01,117
>> Yes. Absolutely.
So there's a lot of people who

640
00:36:01,200 --> 00:36:03,437
who try to take advantage of the hype

641
00:36:03,520 --> 00:36:05,917
for business reasons and so on. But let

642
00:36:06,000 --> 00:36:08,957
me sort of talk to this idea

643
00:36:09,040 --> 00:36:11,837
that sort of new ideas the ideas that

644
00:36:11,920 --> 00:36:16,877
push the field forward may not yet have
a benchmark or it may be very difficult

645
00:36:16,960 --> 00:36:19,597
to establish a benchmark.
>> I agree that's part of the process.

646
00:36:19,680 --> 00:36:21,117
Establishing benchmarks is part of the

647
00:36:21,200 --> 00:36:24,397
process. So what are your thoughts about

648
00:36:24,480 --> 00:36:27,437
so we have these benchmarks on around

649
00:36:27,520 --> 00:36:32,557
stuff we can do with images from
classification to captioning to just

650
00:36:32,640 --> 00:36:35,997
every kind of information you can pull
off from images and the surface level

651
00:36:36,080 --> 00:36:37,997
there's audio data sets there's some

652
00:36:38,080 --> 00:36:39,997
video

653
00:36:40,080 --> 00:36:45,037
what can we start natural language
what kind of stuff what kind of

654
00:36:45,120 --> 00:36:46,637
benchmarks do you see that start

655
00:36:46,720 --> 00:36:49,517
creeping on to more something like

656
00:36:49,600 --> 00:36:53,757
intelligence like reasoning like maybe

657
00:36:53,840 --> 00:36:58,637
you don't like the term but AGI
echoes of that kind of uh

658
00:36:58,720 --> 00:37:01,757
>> a lot of people are working
on interactive environments in which you

659
00:37:01,840 --> 00:37:04,397
can you can train and test uh

660
00:37:04,480 --> 00:37:06,637
intelligent systems so so there for

661
00:37:06,720 --> 00:37:10,477
example uh you know it's

662
00:37:10,560 --> 00:37:15,117
the classical paradigm of supervised
learning is that you you have a data set

663
00:37:15,200 --> 00:37:16,637
you partition it into a training set

664
00:37:16,720 --> 00:37:18,477
validation set test set and there's a

665
00:37:18,560 --> 00:37:22,237
clear protocol right but What if the

666
00:37:22,320 --> 00:37:26,477
that assumes that the samples are
stat statistically independent? You can

667
00:37:26,560 --> 00:37:29,277
exchange them. The order in which you
see them doesn't shouldn't matter. You

668
00:37:29,360 --> 00:37:33,917
know, things like that. But what if
the answer you give determines the next

669
00:37:34,000 --> 00:37:37,837
sample you see, which is the case
for example in robotics, right? You robot

670
00:37:37,920 --> 00:37:41,997
does something and then it gets exposed
to a new room and depending on where it

671
00:37:42,080 --> 00:37:45,037
goes, the room will be different.
So that's the that creates the exploration

672
00:37:45,120 --> 00:37:47,277
problem. >> Mhm.

673
00:37:47,360 --> 00:37:50,717
the what if the samples. So that creates

674
00:37:50,800 --> 00:37:55,197
also a dependency between samples,
right? You you if you move if you can

675
00:37:55,280 --> 00:37:58,637
only move in in space, the next sample
you're going to see is going to be

676
00:37:58,720 --> 00:38:03,517
probably in the same building most
likely, right? So so the all the

677
00:38:03,600 --> 00:38:08,877
assumptions about the validity of this
training set test set hypothesis break

678
00:38:08,960 --> 00:38:11,757
>> whenever a machine can take an action
that has an influence in the in the

679
00:38:11,840 --> 00:38:13,437
world and it's what it's going to see.

680
00:38:13,520 --> 00:38:16,157
So people are setting up artificial

681
00:38:16,240 --> 00:38:18,717
uh environments where where that takes

682
00:38:18,800 --> 00:38:22,077
place, right? The robot runs around a 3D

683
00:38:22,160 --> 00:38:26,237
model of a house and can interact with
objects and things like this. So you do

684
00:38:26,320 --> 00:38:28,237
robotics by simulation. You have those,

685
00:38:28,320 --> 00:38:33,357
you know, open eye gym
type thing or muoko kind of

686
00:38:33,440 --> 00:38:37,197
simulated uh robots and you
have games, you know,

687
00:38:37,280 --> 00:38:40,877
things like that. Um so that that's
where the field is going really. um this

688
00:38:40,960 --> 00:38:43,837
kind of environment. [snorts] Now back

689
00:38:43,920 --> 00:38:48,397
to the question of AGI like
I don't like the term AGI

690
00:38:48,480 --> 00:38:52,717
because it implies that human
intelligence is general

691
00:38:52,800 --> 00:38:58,077
>> and human intelligence is nothing like
general. It's very very specialized. We

692
00:38:58,160 --> 00:39:00,397
think it's general. We like to think
of ourselves as having general

693
00:39:00,480 --> 00:39:03,117
intelligence. We don't.
We're very specialized.

694
00:39:03,200 --> 00:39:04,637
>> We're only slightly more general than

695
00:39:04,720 --> 00:39:07,197
>> Why does it feel general?
So you kind of

696
00:39:07,280 --> 00:39:10,397
uh the term general. I think what's

697
00:39:10,480 --> 00:39:11,997
impressive about humans is this ability

698
00:39:12,080 --> 00:39:14,397
to learn as we were talking about

699
00:39:14,480 --> 00:39:17,517
learning to learn in just so many

700
00:39:17,600 --> 00:39:22,397
different domains. It's perhaps not
arbitrarily general but just you can

701
00:39:22,480 --> 00:39:26,397
learn in many domains and integrate that
knowledge somehow. Okay. The knowledge

702
00:39:26,480 --> 00:39:28,877
persists. >> So let me take
a very specific example.

703
00:39:28,960 --> 00:39:32,237
>> Yes, >> it's not an example.
It's more like a a

704
00:39:32,320 --> 00:39:36,077
quasi mathematical demonstration. So you
have about 1 million fibers coming out

705
00:39:36,160 --> 00:39:39,597
of one of your eyes. Okay. 2 million
total. But let's let's talk about just

706
00:39:39,680 --> 00:39:44,237
one of them. >> It's 1 million
nerve fibers. Your optical nerve.

707
00:39:44,320 --> 00:39:47,837
>> Let's imagine that they are binary.
So they can be active or inactive, right?

708
00:39:47,920 --> 00:39:51,437
So the input to your visual
cortex is 1 million bits.

709
00:39:51,520 --> 00:39:53,997
>> Mhm.

710
00:39:54,080 --> 00:39:58,477
>> Now they connected to your brain
in a particular way. And your brain has

711
00:39:58,560 --> 00:40:01,597
connections that are kind of a little
bit like a conversion net. They're kind

712
00:40:01,680 --> 00:40:05,037
of local uh you know
in space and things like this.

713
00:40:05,120 --> 00:40:07,997
>> Now imagine I play a trick on you.

714
00:40:08,080 --> 00:40:12,637
It's a pretty nasty trick, I admit.
I uh I cut your optical nerve

715
00:40:12,720 --> 00:40:14,877
>> and I put a device that makes a random

716
00:40:14,960 --> 00:40:17,597
perturbation of a permutation of all the

717
00:40:17,680 --> 00:40:20,717
nerve fibers. So now what comes to your

718
00:40:20,800 --> 00:40:24,957
to your brain is a fixed but random
permutation of all the pixels.

719
00:40:25,040 --> 00:40:27,757
>> Mhm. >> There's no way
in hell that your visual

720
00:40:27,840 --> 00:40:30,557
cortex, even if I do this to you in

721
00:40:30,640 --> 00:40:33,757
infancy, will actually learn vision to

722
00:40:33,840 --> 00:40:37,117
the same level of quality that you can.

723
00:40:37,200 --> 00:40:39,917
>> Got it. And you're saying
there's no way you relearn that.

724
00:40:40,000 --> 00:40:44,157
>> No, because now two pixels that are
nearby in the world will end up in very

725
00:40:44,240 --> 00:40:47,997
different places in your visual
cortex and your neurons there have no

726
00:40:48,080 --> 00:40:50,557
connections with each other
because they only connected locally.

727
00:40:50,640 --> 00:40:55,597
>> So this whole our entire the hardware
is built in many ways to support

728
00:40:55,680 --> 00:40:57,037
>> the locality of the real world.

729
00:40:57,120 --> 00:40:59,677
>> Yeah. >> Yes. That's specialization.

730
00:40:59,760 --> 00:41:02,557
>> Yeah. But it's still pretty
damn impressive. So it's not perfect

731
00:41:02,640 --> 00:41:04,237
generalization. It's not even close.

732
00:41:04,320 --> 00:41:08,237
>> No. No. It's it's it's it's not
that it's not even close. It's not at all.

733
00:41:08,320 --> 00:41:11,757
>> It's not specialized.
>> So how many boolean functions? So let's

734
00:41:11,840 --> 00:41:14,477
imagine you want to train your visual

735
00:41:14,560 --> 00:41:17,757
system to uh you know recognize

736
00:41:17,840 --> 00:41:22,077
particular patterns of those 1
million bits. Okay. So that's a boolean

737
00:41:22,160 --> 00:41:25,437
function, right? Either the pattern
is here or not here. This a two two-way

738
00:41:25,520 --> 00:41:30,717
classification with 1
million binary inputs.

739
00:41:30,800 --> 00:41:33,437
How many such boolean functions are there?

740
00:41:33,520 --> 00:41:35,597
Okay, you have you have two to the 1

741
00:41:35,680 --> 00:41:38,717
million combinations of inputs. For each

742
00:41:38,800 --> 00:41:41,437
of those, you have an output bit and so

743
00:41:41,520 --> 00:41:44,717
you have two to the two to the 1 million

744
00:41:44,800 --> 00:41:47,117
boolean functions of this type. Okay,

745
00:41:47,200 --> 00:41:50,157
which is an unimaginably large number.

746
00:41:50,240 --> 00:41:52,157
How many of those functions can actually

747
00:41:52,240 --> 00:41:54,477
be computed by your visual cortex? And

748
00:41:54,560 --> 00:41:59,837
the answer is a tiny tiny tiny tiny
tiny tiny sliver. Like an enormously tiny

749
00:41:59,920 --> 00:42:05,357
sliver. >> Yeah. Yeah. So,
we are ridiculously specialized,

750
00:42:05,440 --> 00:42:08,557
you know. Okay. >> But, [laughter]

751
00:42:08,640 --> 00:42:10,637
okay, that's an argument against the

752
00:42:10,720 --> 00:42:12,877
word general. I think there there's a

753
00:42:12,960 --> 00:42:16,237
>> I I there's I agree
with your intuition,

754
00:42:16,320 --> 00:42:19,117
but I'm not sure it's it seems the bre

755
00:42:19,200 --> 00:42:23,757
the the brain is impressively

756
00:42:23,840 --> 00:42:26,637
uh capable of adjusting to things. So

757
00:42:26,720 --> 00:42:31,677
>> it's because we can't imagine
tasks that are outside of our

758
00:42:31,760 --> 00:42:34,717
comprehension, >> right?
So we think we think we're

759
00:42:34,800 --> 00:42:37,357
general because we're general of all
the things that we can apprehend.

760
00:42:37,440 --> 00:42:40,237
>> So yeah, >> but there
is a huge world out there of

761
00:42:40,320 --> 00:42:43,917
things that we have no idea.
We call that heat, by the way.

762
00:42:44,000 --> 00:42:47,437
>> Heat. >> Heat.
So uh at least physicists call

763
00:42:47,520 --> 00:42:50,477
that heat or they call it entropy which is

764
00:42:50,560 --> 00:42:53,517
>> you know you have a

765
00:42:53,600 --> 00:42:56,477
thing full of gas right

766
00:42:56,560 --> 00:42:57,997
>> closed system full of gas

767
00:42:58,080 --> 00:43:01,517
>> right closed or no closed it has uh you

768
00:43:01,600 --> 00:43:05,517
know pressure it has uh temperature it

769
00:43:05,600 --> 00:43:08,877
has you know and you can write equations

770
00:43:08,960 --> 00:43:12,557
PV equivalent on T you know
things like that right [snorts]

771
00:43:12,640 --> 00:43:15,517
when you reduce a volume the temperature
goes up the pressure goes up, you know,

772
00:43:15,600 --> 00:43:19,917
things like that, right? Uh for a
perfect gas at least, those are the

773
00:43:20,000 --> 00:43:24,237
things you can know about that system
and it's a tiny tiny number of bits

774
00:43:24,320 --> 00:43:28,237
compared to the complete information
of the state of the entire system because

775
00:43:28,320 --> 00:43:29,677
the state of the entire system will give

776
00:43:29,760 --> 00:43:31,997
you the position and momentum of every

777
00:43:32,080 --> 00:43:35,517
every molecule of the gas.

778
00:43:35,600 --> 00:43:37,517
And what you don't know about it is the

779
00:43:37,600 --> 00:43:41,117
entropy and you interpret it as heat.

780
00:43:41,200 --> 00:43:44,237
the energy contained in that thing is is

781
00:43:44,320 --> 00:43:46,877
what we call heat. Now,

782
00:43:46,960 --> 00:43:51,357
it's very possible that in fact there
is some very strong structure in how those

783
00:43:51,440 --> 00:43:55,277
molecules are moving. It's just that
they are in a way that we're just not

784
00:43:55,360 --> 00:43:59,037
wired to perceive. >> Yeah,
we're ignorant to it and there's

785
00:43:59,120 --> 00:44:02,637
in near infinite amount of things
we're not wired to perceive.

786
00:44:02,720 --> 00:44:03,917
>> And you're right,
that's a that's a nice

787
00:44:04,000 --> 00:44:05,677
way to put it. We're general to all the

788
00:44:05,760 --> 00:44:11,037
things we can imagine which
is a very tiny subset of all

789
00:44:11,120 --> 00:44:13,517
things that are possible.
>> So it's like chromograph complexity or

790
00:44:13,600 --> 00:44:16,877
the chromograph chhattin of complexity.

791
00:44:16,960 --> 00:44:19,437
>> Yeah. >> You know every

792
00:44:19,520 --> 00:44:22,717
bit string or every integer is random

793
00:44:22,800 --> 00:44:27,900
except for all the ones that you
can actually write down. [laughter]

794
00:44:28,800 --> 00:44:31,997
>> Yeah. Okay. So beautifully put but you
know so we can just call it artificial

795
00:44:32,080 --> 00:44:35,117
intelligence. We don't
need to have a general

796
00:44:35,200 --> 00:44:37,997
>> or human level human
level intelligence is

797
00:44:38,080 --> 00:44:41,117
>> you know you you'll start anytime you

798
00:44:41,200 --> 00:44:43,757
touch human it gets it gets interesting

799
00:44:43,840 --> 00:44:46,097
because um

800
00:44:46,180 --> 00:44:50,877
[snorts] uh you know it's it's because
we attach ourselves to human and it's

801
00:44:50,960 --> 00:44:53,197
difficult to define what
human intelligence is.

802
00:44:53,280 --> 00:44:54,397
>> Yeah.

803
00:44:54,480 --> 00:44:57,117
>> Nevertheless, my definition is uh maybe

804
00:44:57,200 --> 00:45:01,037
uh demo impressive intelligence. Okay.

805
00:45:01,120 --> 00:45:02,557
Damn impressive demonstration of

806
00:45:02,640 --> 00:45:04,957
intelligence whatever and so on that

807
00:45:05,040 --> 00:45:08,557
topic most successes in deep learning

808
00:45:08,640 --> 00:45:11,037
have been in supervised learning.

809
00:45:11,120 --> 00:45:14,637
>> What is your view on unsupervised

810
00:45:14,720 --> 00:45:17,757
learning? Is there hope to reduce

811
00:45:17,840 --> 00:45:20,877
involvement of human input and still

812
00:45:20,960 --> 00:45:25,437
have uh successful systems
that are have practical use?

813
00:45:25,520 --> 00:45:26,877
>> Yeah, I mean there's definitely a hope.

814
00:45:26,960 --> 00:45:28,397
It's it's more than a hope actually.

815
00:45:28,480 --> 00:45:33,277
it's uh it's you know mounting evidence
for it and that's basically all I do

816
00:45:33,360 --> 00:45:37,357
like the only thing I'm interested in at
the moment is I call it self-supervised

817
00:45:37,440 --> 00:45:40,717
learning not unsupervised because
unsupervised learning is is a loaded

818
00:45:40,800 --> 00:45:44,637
term people who know
something about machine

819
00:45:44,720 --> 00:45:47,837
learning you know tell you so
you're doing clustering or PCA

820
00:45:47,920 --> 00:45:49,997
>> which is not the case and the
wide public you know when you say

821
00:45:50,080 --> 00:45:52,397
unsupervised learning oh my god you
know machines are going to learn by

822
00:45:52,480 --> 00:45:56,477
themselves and without supervision
you know they see this as

823
00:45:56,560 --> 00:46:00,237
>> where's where's the parents Yeah.
So, so I call it self-supervised learning

824
00:46:00,320 --> 00:46:03,917
because in fact the underlying
algorithms that are used are the same

825
00:46:04,000 --> 00:46:08,637
algorithms as the supervised learning
algorithms. Um, except that what we

826
00:46:08,720 --> 00:46:10,637
train them to do is not predict a

827
00:46:10,720 --> 00:46:13,277
particular set of variables like the

828
00:46:13,360 --> 00:46:17,597
category of of an image

829
00:46:17,680 --> 00:46:20,957
and and not to predict a set
of variables that have been provided by

830
00:46:21,040 --> 00:46:24,237
human labelers. But what you train the

831
00:46:24,320 --> 00:46:26,637
machine to do is basically reconstruct a

832
00:46:26,720 --> 00:46:29,437
piece of its input that it's being

833
00:46:29,520 --> 00:46:31,357
is being maxed maxed out essentially.

834
00:46:31,440 --> 00:46:36,157
You can think of it this way, right?
So show a piece of video to a machine and

835
00:46:36,240 --> 00:46:39,597
ask it to predict what's going to happen
next. And of course after a while you

836
00:46:39,680 --> 00:46:43,677
can show what what happens and the
machine will kind of train itself to do

837
00:46:43,760 --> 00:46:48,477
better at that task.
You can do like all the latest most

838
00:46:48,560 --> 00:46:53,277
successful models in natural language
processing use self-supervised learning.

839
00:46:53,360 --> 00:46:55,197
you know, sort of BERT style systems,

840
00:46:55,280 --> 00:46:56,877
for example, right? You show it a window

841
00:46:56,960 --> 00:47:00,797
of a thousand words on a test corpus.

842
00:47:00,880 --> 00:47:03,757
You take out 15% of the words and then

843
00:47:03,840 --> 00:47:08,477
you train a machine to predict
the words that are missing.

844
00:47:08,560 --> 00:47:11,597
>> That's self-supervised learning.
It's not predicting the future. It's just,

845
00:47:11,680 --> 00:47:14,557
you know, predicting things in the middle,
but you could have it predict

846
00:47:14,640 --> 00:47:16,717
the future.
That's what language models do.

847
00:47:16,800 --> 00:47:18,797
>> So, you construct,
so in an unsupervised

848
00:47:18,880 --> 00:47:21,277
way, you construct a model of language.

849
00:47:21,360 --> 00:47:25,437
Do you think >> or video
or the physical world or

850
00:47:25,520 --> 00:47:29,757
whatever? Right. >> How far
do you think that can take us?

851
00:47:29,840 --> 00:47:31,837
Do you think >> art

852
00:47:31,920 --> 00:47:35,117
>> understands anything?

853
00:47:35,200 --> 00:47:37,597
>> To some level it has you know a shallow

854
00:47:37,680 --> 00:47:41,117
understanding of uh of text but it needs

855
00:47:41,200 --> 00:47:44,637
to I mean to have kind of true human
level intelligence I think you need to

856
00:47:44,720 --> 00:47:47,357
ground language in reality. So some

857
00:47:47,440 --> 00:47:51,757
people are attempting to do this right
having systems that kind of have some

858
00:47:51,840 --> 00:47:55,437
visual representation of what what
is being talked about which is one reason

859
00:47:55,520 --> 00:47:57,117
you need those interactive environments

860
00:47:57,200 --> 00:47:59,757
actually but this is like a huge

861
00:47:59,840 --> 00:48:04,877
technical problem that is not solved
and that explains why self-supervised

862
00:48:04,960 --> 00:48:08,077
learning works in the context of natural
language but does not work in the

863
00:48:08,160 --> 00:48:12,477
context or at least not well in the
context of image recognition and video

864
00:48:12,560 --> 00:48:17,117
although it's making progress quickly
and [snorts] the reason that reason is

865
00:48:17,200 --> 00:48:20,397
the fact that it's
much easier to represent

866
00:48:20,480 --> 00:48:24,477
uncertainty in the prediction in a
context of natural language than it is

867
00:48:24,560 --> 00:48:27,277
in the context of things
like video and images.

868
00:48:27,360 --> 00:48:30,077
>> So for example, if I ask you to predict

869
00:48:30,160 --> 00:48:34,797
what words are missing, you know,
15% of the words that are taken out,

870
00:48:34,880 --> 00:48:38,237
>> the possibilities are small. That means
>> it's small, right? There is 100,000

871
00:48:38,320 --> 00:48:41,277
words in the in the lexicon and what the

872
00:48:41,360 --> 00:48:46,077
machine spits out is a a big probability
vector, right? It's a bunch of numbers

873
00:48:46,160 --> 00:48:49,757
between 0 and one that's some to one
and we know how to do how to do this with

874
00:48:49,840 --> 00:48:53,677
computers. [snorts] So there
representing uncertainty in the

875
00:48:53,760 --> 00:48:55,597
prediction is relatively easy and that's

876
00:48:55,680 --> 00:48:58,397
in my opinion why those techniques work

877
00:48:58,480 --> 00:49:02,877
for NLP for images if you ask if you

878
00:49:02,960 --> 00:49:06,077
block a piece of an image and you ask
a system reconstruct that piece of the

879
00:49:06,160 --> 00:49:08,877
image there are many possible answers

880
00:49:08,960 --> 00:49:12,717
that are all perfectly legit right and

881
00:49:12,800 --> 00:49:16,957
how do you represent that this set
of possible answers you you can't train a

882
00:49:17,040 --> 00:49:20,317
system to make one prediction you can't
train a neural net to say here it is

883
00:49:20,400 --> 00:49:23,357
that's the image because
it's there's a whole set of

884
00:49:23,440 --> 00:49:25,917
things that are compatible with it.
So how do you get the machine to represent

885
00:49:26,000 --> 00:49:27,597
not a single output but a whole set of

886
00:49:27,680 --> 00:49:30,237
outputs

887
00:49:30,320 --> 00:49:33,277
and you know similarly with uh video

888
00:49:33,360 --> 00:49:37,277
prediction there's a lot of things
that can happen in the future of a video

889
00:49:37,360 --> 00:49:40,717
you're looking at me right now I'm not
moving my head very much but you know I

890
00:49:40,800 --> 00:49:43,997
might you know turn my my head
to the left or to the right >> right

891
00:49:44,080 --> 00:49:47,437
>> if you don't have
a system that can predict this

892
00:49:47,520 --> 00:49:50,557
>> and you train it with le square
to kind of minimize the error with the

893
00:49:50,640 --> 00:49:54,077
prediction and what I'm doing what you
get is a blurry image of myself right

894
00:49:54,160 --> 00:49:57,437
>> in all possible future positions
that I might be in which is not a good

895
00:49:57,520 --> 00:49:59,837
prediction. But so there might be other

896
00:49:59,920 --> 00:50:02,797
ways to do the self-supervision, right,

897
00:50:02,880 --> 00:50:06,877
for visual scenes.
>> Like what? [laughter]

898
00:50:06,960 --> 00:50:09,117
>> If I to I mean if
I knew I wouldn't tell

899
00:50:09,200 --> 00:50:12,637
you I publish it first. I don't know. I

900
00:50:12,720 --> 00:50:17,597
>> No, there might be. [laughter] >>
So I mean these are kind of uh there

901
00:50:17,680 --> 00:50:20,877
might be artificial ways of like
selfplay in games to where you can

902
00:50:20,960 --> 00:50:22,877
simulate part of the environment. You can

903
00:50:22,960 --> 00:50:24,157
>> Oh, that doesn't
solve the problem. It's

904
00:50:24,240 --> 00:50:27,677
just a way of generating data.

905
00:50:27,760 --> 00:50:32,317
But because you have more of a control
mean you can control. Yeah. It's a way

906
00:50:32,400 --> 00:50:36,157
to generate data. That's right. And
because you can do huge amounts of data

907
00:50:36,240 --> 00:50:38,477
generation that doesn't you're right.

908
00:50:38,560 --> 00:50:43,517
This well it's it's a creeps up on the
problem from the side of data and you

909
00:50:43,600 --> 00:50:44,957
don't think that's
the right way to creep up.

910
00:50:45,040 --> 00:50:47,837
>> It doesn't solve this problem
of handling uncertainty in the world.

911
00:50:47,920 --> 00:50:51,037
Right. So if you if you have a machine

912
00:50:51,120 --> 00:50:55,757
learn a predictive model of the world
in a game that is deterministic or quasi

913
00:50:55,840 --> 00:50:57,517
deterministic

914
00:50:57,600 --> 00:51:00,637
uh it's easy right just you know give a

915
00:51:00,720 --> 00:51:05,117
few frames of the game to a connet put
a bunch of layers and then have the game

916
00:51:05,200 --> 00:51:06,877
generates the next few frames

917
00:51:06,960 --> 00:51:09,117
>> and and if the game is deterministic it

918
00:51:09,200 --> 00:51:13,197
works fine. Um, and that includes, you

919
00:51:13,280 --> 00:51:18,397
know, feeding the system with the action
that your little character is is going

920
00:51:18,480 --> 00:51:23,277
to take the problem comes
from the fact that the

921
00:51:23,360 --> 00:51:27,197
real world and certain most games are
not entirely predictable. And so there

922
00:51:27,280 --> 00:51:31,357
you get those blurry predictions and you
can't do planning with blurry predictions.

923
00:51:31,440 --> 00:51:33,117
Right? So if you have a perfect model of

924
00:51:33,200 --> 00:51:37,117
the world, you can in your head run this

925
00:51:37,200 --> 00:51:42,157
model with a hypothesis for a sequence
of actions. Are you going to predict the

926
00:51:42,240 --> 00:51:43,997
outcome of that sequence of actions?

927
00:51:44,080 --> 00:51:48,397
>> Mhm. But if your model is imperfect,

928
00:51:48,480 --> 00:51:51,917
how can you plan? >> Yeah.
It quickly explodes.

929
00:51:52,000 --> 00:51:56,477
What are your thoughts on the extension
of this which topic I'm super excited

930
00:51:56,560 --> 00:51:59,917
about? It's connected to something you
were talking about in terms of robotics

931
00:52:00,000 --> 00:52:02,797
is active learning. So as opposed to

932
00:52:02,880 --> 00:52:05,437
sort of unsup completely unsupervised or

933
00:52:05,520 --> 00:52:08,237
self-supervised learning,

934
00:52:08,320 --> 00:52:11,997
you ask the system for human help for

935
00:52:12,080 --> 00:52:14,637
for selecting parts you want annotated

936
00:52:14,720 --> 00:52:16,637
next. So if you think about a robot

937
00:52:16,720 --> 00:52:19,037
exploring a space or a baby exploring a

938
00:52:19,120 --> 00:52:24,397
space or a system exploring a data set
every once in a while asking for human

939
00:52:24,480 --> 00:52:25,197
input,

940
00:52:25,280 --> 00:52:29,277
>> do you see value in that kind of work?

941
00:52:29,360 --> 00:52:33,277
>> I don't see transformative value.
It it's going to make things that we can

942
00:52:33,360 --> 00:52:36,717
already do >> more
efficient or they will learn

943
00:52:36,800 --> 00:52:39,597
slightly more efficiently, but it's
not going to make machines sort of

944
00:52:39,680 --> 00:52:42,297
significantly more intelligent, I think.

945
00:52:42,380 --> 00:52:42,557
>> [snorts]

946
00:52:42,640 --> 00:52:45,517
>> And I and and by the way there is no

947
00:52:45,600 --> 00:52:50,237
opposition there's no um uh conflict

948
00:52:50,320 --> 00:52:52,957
between self-supervised learning
reinforcement learning and supervised

949
00:52:53,040 --> 00:52:56,157
learning or imitation
learning or active learning.

950
00:52:56,240 --> 00:52:58,877
>> I see self-supervised
learning as a as a

951
00:52:58,960 --> 00:53:01,037
preliminary to all of the above.

952
00:53:01,120 --> 00:53:03,677
>> Yes. So

953
00:53:03,760 --> 00:53:05,677
the example I use very often is how is

954
00:53:05,760 --> 00:53:10,077
it that so if you use

955
00:53:10,160 --> 00:53:11,917
classical reinforcement learning deep

956
00:53:12,000 --> 00:53:14,717
reinforcement learning if you want

957
00:53:14,800 --> 00:53:18,477
the best methods today

958
00:53:18,560 --> 00:53:23,037
soal model free reinforcement learning
to learn to play Atari games uh take

959
00:53:23,120 --> 00:53:26,317
about 80 hours of training to reach the
level that any human can reach in about

960
00:53:26,400 --> 00:53:28,477
15 minutes. Mhm.

961
00:53:28,560 --> 00:53:31,037
>> Um they get better than humans, but it

962
00:53:31,120 --> 00:53:34,477
takes them a long time.

963
00:53:34,560 --> 00:53:38,477
Alpha star. Okay. the, you know, Oreo

964
00:53:38,560 --> 00:53:41,437
vin and and his teams the system to play

965
00:53:41,520 --> 00:53:45,197
um to um to play Starcraft

966
00:53:45,280 --> 00:53:48,557
plays, you know, a single map, a single

967
00:53:48,640 --> 00:53:51,677
type of uh player

968
00:53:51,760 --> 00:53:54,157
and can reach

969
00:53:54,240 --> 00:53:57,197
better than human level with about the

970
00:53:57,280 --> 00:54:00,477
equivalent of 200 years of training

971
00:54:00,560 --> 00:54:05,117
playing against itself. It's 200 years,
right? It's not something that no no

972
00:54:05,200 --> 00:54:08,557
human can could ever do. >> I mean,
I'm not sure what lesson to take

973
00:54:08,640 --> 00:54:12,077
away from that. >> Okay.
Now, take those algorithms, the

974
00:54:12,160 --> 00:54:15,037
best RL algorithms we have today. To

975
00:54:15,120 --> 00:54:19,917
train a a car to drive itself, >> it
would probably have to drive millions

976
00:54:20,000 --> 00:54:23,757
of hours. It will have to kill thousands
of pedestrians. It will have to run into

977
00:54:23,840 --> 00:54:26,157
thousands of trees.
It will have to run off cliffs.

978
00:54:26,240 --> 00:54:28,557
>> Yeah. >> And it had
to run the cliff multiple

979
00:54:28,640 --> 00:54:32,317
times before it figures out that it's
a bad idea, first of all. Yeah.

980
00:54:32,400 --> 00:54:33,917
>> And second of all,
before it figures out

981
00:54:34,000 --> 00:54:38,877
how not to do it and so I mean this type
of learning obviously does not reflect

982
00:54:38,960 --> 00:54:42,477
the kind of learning that animals
and humans do. There is something missing

983
00:54:42,560 --> 00:54:44,557
that's really really important there.

984
00:54:44,640 --> 00:54:48,877
And my hypothesis which I've been
advocating for like five years now is

985
00:54:48,960 --> 00:54:53,037
that we have predictive
models of the world

986
00:54:53,120 --> 00:54:57,677
that include the ability to predict
under uncertainty. [snorts] And what

987
00:54:57,760 --> 00:55:02,477
allows us to not run off
a cliff when we learn to

988
00:55:02,560 --> 00:55:06,077
drive, most of us can learn to drive in
about 20 or 30 hours of training without

989
00:55:06,160 --> 00:55:09,277
ever crashing, causing any accident. And

990
00:55:09,360 --> 00:55:13,437
if we drive next to a cliff, we know
that if we turn the wheel to the right,

991
00:55:13,520 --> 00:55:16,797
the car is going to run off the cliff
and nothing good is going to come out of

992
00:55:16,880 --> 00:55:19,437
this because we have a pretty good model
of intuitive physics that tells us, you

993
00:55:19,520 --> 00:55:23,037
know, the car is going to fall. We know
we know about gravity. Babies learned

994
00:55:23,120 --> 00:55:24,637
this around the age of eight or nine

995
00:55:24,720 --> 00:55:27,357
months that objects don't float, they

996
00:55:27,440 --> 00:55:32,077
they fall. Um, and you know, we have
a pretty good idea of the effect of

997
00:55:32,160 --> 00:55:34,957
turning the wheel on the car and you
know, we know we need to stay on the

998
00:55:35,040 --> 00:55:38,077
road. So, there's a lot of things
that we bring to the table, which is

999
00:55:38,160 --> 00:55:39,437
basically our predictive model of the

1000
00:55:39,520 --> 00:55:42,477
world. [snorts] And that model allows us

1001
00:55:42,560 --> 00:55:45,117
to not do stupid things and to basically

1002
00:55:45,200 --> 00:55:49,597
stay within the context of uh things
we need to do. we still face, you know,

1003
00:55:49,680 --> 00:55:53,837
unpredictable situations and that's how
we learn. But that allows us to learn

1004
00:55:53,920 --> 00:55:57,597
really, really, really quickly.
So that's called model based reinforcement

1005
00:55:57,680 --> 00:56:01,117
learning. There's some
imitation and supervised

1006
00:56:01,200 --> 00:56:04,397
learning because we have a driving
instructor that tells us occasionally

1007
00:56:04,480 --> 00:56:10,157
what to do. Uh but most of the learning >>
is model based >> is learning the model.

1008
00:56:10,240 --> 00:56:12,637
>> Yeah. >> Learning
physics that we've done since

1009
00:56:12,720 --> 00:56:15,357
we were babies. That's where
all almost all the learning

1010
00:56:15,440 --> 00:56:17,277
>> and the physics
is somewhat transferable

1011
00:56:17,360 --> 00:56:21,917
from is transferable from scene to scene.
Stupid things are the same

1012
00:56:22,000 --> 00:56:25,197
everywhere. >> Yeah.
I mean if you you know you have an

1013
00:56:25,280 --> 00:56:29,037
experience of the world you don't need
to be a particularly from a particularly

1014
00:56:29,120 --> 00:56:31,517
intelligent species to know that if you

1015
00:56:31,600 --> 00:56:34,877
spill water from a container the you

1016
00:56:34,960 --> 00:56:37,197
know the the rest is going to get wet.

1017
00:56:37,280 --> 00:56:40,397
You might get wet. Um so you know cats

1018
00:56:40,480 --> 00:56:44,317
know this right? Yeah.
>> So the main problem we need to solve is

1019
00:56:44,400 --> 00:56:47,117
how do we learn models of the world?

1020
00:56:47,200 --> 00:56:48,397
That's and that's what I'm interested.

1021
00:56:48,480 --> 00:56:51,197
That's what self-supervised
learning is all about.

1022
00:56:51,280 --> 00:56:52,877
>> If you were to try to construct a

1023
00:56:52,960 --> 00:56:56,477
benchmark for um

1024
00:56:56,560 --> 00:56:58,877
let's let's look at emnest. I love that

1025
00:56:58,960 --> 00:57:02,077
data set. But if do you think it's

1026
00:57:02,160 --> 00:57:06,957
useful interesting slash possible to uh

1027
00:57:07,040 --> 00:57:11,117
perform well on emnest with just
one example of each digit?

1028
00:57:11,200 --> 00:57:15,597
And how would we solve that problem?

1029
00:57:15,680 --> 00:57:17,037
>> The answer is probably yes.
The question

1030
00:57:17,120 --> 00:57:21,277
is what other type of learning are you
allowed to do? So if what you're allowed

1031
00:57:21,360 --> 00:57:25,117
to do is train on some gigantic data set
of labeled digit, that's called transfer

1032
00:57:25,200 --> 00:57:28,317
learning. >> And we know that works. Okay.

1033
00:57:28,400 --> 00:57:32,397
>> Uh we do this at Facebook like
in production, right? We we train large

1034
00:57:32,480 --> 00:57:35,517
convolutional nest to predict hashtags
that people type on Instagram and we

1035
00:57:35,600 --> 00:57:39,597
train on billions of images, literally
billions. And uh and then we chop off

1036
00:57:39,680 --> 00:57:43,437
the last layer and fine-tune on whatever
task we want. That works really well.

1037
00:57:43,520 --> 00:57:47,117
You can beat you know the imageet record
with this. We actually open source the

1038
00:57:47,200 --> 00:57:48,957
whole thing like a few weeks ago.

1039
00:57:49,040 --> 00:57:50,557
>> Yeah, that's that's still pretty cool.

1040
00:57:50,640 --> 00:57:54,717
But yeah, so what in what would
be impressive and what's useful and

1041
00:57:54,800 --> 00:57:57,437
impressive? What kind of transfer
learning would be useful and impressive?

1042
00:57:57,520 --> 00:58:00,637
Is it Wikipedia? That kind of thing?

1043
00:58:00,720 --> 00:58:04,157
>> I don't think transfer learning
is really where we should focus. We should

1044
00:58:04,240 --> 00:58:07,037
try to do

1045
00:58:07,120 --> 00:58:08,877
you know have a kind of scenario for a

1046
00:58:08,960 --> 00:58:13,037
benchmark where you have unlabelled data

1047
00:58:13,120 --> 00:58:15,597
and you can and it's very large number

1048
00:58:15,680 --> 00:58:18,237
of unlabelled data. Uh it could be video

1049
00:58:18,320 --> 00:58:21,357
clips, it could be where you do you know

1050
00:58:21,440 --> 00:58:23,197
frame prediction, it could be images

1051
00:58:23,280 --> 00:58:28,397
where you could choose
to you know mask a piece of it

1052
00:58:28,480 --> 00:58:30,157
could be whatever but they're unlabeled

1053
00:58:30,240 --> 00:58:32,877
and you're not allowed to label them.

1054
00:58:32,960 --> 00:58:37,517
So you do some training on this and then

1055
00:58:37,600 --> 00:58:41,517
you train on a particular supervised

1056
00:58:41,600 --> 00:58:45,277
task imageet or nest and you measure how

1057
00:58:45,360 --> 00:58:48,797
your test error decreases
or validation error

1058
00:58:48,880 --> 00:58:51,837
decreases as you increase
the number of label training samples.

1059
00:58:51,920 --> 00:58:55,757
>> Mhm. >> Okay. And

1060
00:58:55,840 --> 00:58:58,477
and what what you'd like to see is is

1061
00:58:58,560 --> 00:59:01,357
that you know your your error decreases

1062
00:59:01,440 --> 00:59:05,677
much faster than if you train
from scratch from random weights.

1063
00:59:05,760 --> 00:59:09,197
So that to reach the same level of
performance that a completely supervised

1064
00:59:09,280 --> 00:59:13,757
uh purely supervised system would reach
you would need way fewer samples. So

1065
00:59:13,840 --> 00:59:17,357
that's the crucial question because
it will answer the question to like you

1066
00:59:17,440 --> 00:59:19,197
know people interested in medical image

1067
00:59:19,280 --> 00:59:21,037
analysis. Okay, you know, if I want to

1068
00:59:21,120 --> 00:59:24,637
get to a particular uh level of uh of

1069
00:59:24,720 --> 00:59:27,197
error rate for this task, I know I need

1070
00:59:27,280 --> 00:59:31,197
um a million samples, can I do, you

1071
00:59:31,280 --> 00:59:34,797
know, self-supervised pre-training
to reduce this to about 100 or something?

1072
00:59:34,880 --> 00:59:37,997
>> And you think the answer there
is self-supervised pre-training?

1073
00:59:38,080 --> 00:59:42,237
>> Yeah, some form some
for some form of it.

1074
00:59:42,320 --> 00:59:45,837
>> Telling you active
learning, but you disagree.

1075
00:59:45,920 --> 00:59:49,837
>> No, it's it's not useless. It's just
not going to lead to a quantum leap. It's

1076
00:59:49,920 --> 00:59:51,197
just going to make
things that we already do.

1077
00:59:51,280 --> 00:59:55,437
>> So, you're way smarter than me. I just
disagree with you and but I don't have

1078
00:59:55,520 --> 00:59:59,037
anything to back that. It's just
intuition. So, I worked with a lot of

1079
00:59:59,120 --> 01:00:00,637
large scale data sets and there's

1080
01:00:00,720 --> 01:00:03,437
something there might be magic in active

1081
01:00:03,520 --> 01:00:07,917
learning. But, okay, at least
I said it publicly. [laughter]

1082
01:00:08,000 --> 01:00:09,757
At least I'm being an idiot publicly.

1083
01:00:09,840 --> 01:00:13,437
Okay. It's not being an idiot. It's, you
know, working with the data you have. I

1084
01:00:13,520 --> 01:00:17,757
mean I mean certainly people are doing
things like okay I have 3,000 hours of

1085
01:00:17,840 --> 01:00:21,357
you know imitation running
for self-driving car but most of those are

1086
01:00:21,440 --> 01:00:25,597
incredibly boring. What I like is select
you know 10% of them that are kind of

1087
01:00:25,680 --> 01:00:27,757
the most informative and with just that

1088
01:00:27,840 --> 01:00:33,117
I would probably reach the same. So it's
a weak form of uh of active learning if

1089
01:00:33,200 --> 01:00:37,357
you want. >> Yes. But there might
be a a much stronger version.

1090
01:00:37,440 --> 01:00:39,357
>> Yeah that's right. >> That's
what and that's an open question

1091
01:00:39,440 --> 01:00:41,917
if it exists. The the question is how

1092
01:00:42,000 --> 01:00:44,877
much stronger we can get. Elon Musk is

1093
01:00:44,960 --> 01:00:49,437
confident. Talked to him recently. Uh
he's confident that large scale data and

1094
01:00:49,520 --> 01:00:53,197
deep learning can solve
the autonomous driving problem.

1095
01:00:53,280 --> 01:00:57,037
What are your thoughts on the limits
possibilities of deep learning in this

1096
01:00:57,120 --> 01:01:00,237
space? >> It's obviously
part of the solution. I

1097
01:01:00,320 --> 01:01:03,757
mean I don't think we'll ever have
a self-driving system or at least not in

1098
01:01:03,840 --> 01:01:05,517
the foreseeable future that does not use

1099
01:01:05,600 --> 01:01:07,677
deep learning. Let me put it this way.

1100
01:01:07,760 --> 01:01:11,277
Now how much of it? So in the history of

1101
01:01:11,360 --> 01:01:15,837
sort of engineering particularly sort of

1102
01:01:15,920 --> 01:01:19,437
sort of AI like systems there's
generally a first phase where everything

1103
01:01:19,520 --> 01:01:22,477
is built by hand and there is a second
phase and that was the case for

1104
01:01:22,560 --> 01:01:27,357
autonomous driving you know 20 30 years
ago there's a phase where there's a

1105
01:01:27,440 --> 01:01:30,477
little bit of learning is used
but there's a lot of engineering that's

1106
01:01:30,560 --> 01:01:35,437
involved in kind of you know taking care
of corner cases and and putting limits

1107
01:01:35,520 --> 01:01:38,957
etc because the learning system
is not perfect and then as technology

1108
01:01:39,040 --> 01:01:44,157
progresses, we end up relying more and
more on learning. That's the history of

1109
01:01:44,240 --> 01:01:47,277
character recognition, the history of
speech recognition, now computer vision,

1110
01:01:47,360 --> 01:01:49,037
natural language processing. And I think

1111
01:01:49,120 --> 01:01:52,397
the same is going to happen with uh with

1112
01:01:52,480 --> 01:01:55,917
autonomous driving that currently the

1113
01:01:56,000 --> 01:02:00,637
u the methods that are closest
to providing some level of autonomy, some

1114
01:02:00,720 --> 01:02:04,077
you know decent level of autonomy where
you don't expect a driver to kind of do

1115
01:02:04,160 --> 01:02:08,397
anything is where you
constrain the world. So you

1116
01:02:08,480 --> 01:02:12,397
only run within, you know, 100 square
kilometers or square miles in Phoenix,

1117
01:02:12,480 --> 01:02:14,637
but the weather is nice and the roads

1118
01:02:14,720 --> 01:02:17,437
are wide, which is what Whimo is doing.

1119
01:02:17,520 --> 01:02:20,317
You uh completely overengineer the car

1120
01:02:20,400 --> 01:02:24,157
with tons of light and sophisticated

1121
01:02:24,240 --> 01:02:27,517
sensors that are too expensive
for consumer cars, but they're fine if you

1122
01:02:27,600 --> 01:02:32,157
just run a fleet.
And you engineer the thing uh the hell

1123
01:02:32,240 --> 01:02:36,077
out of the everything else. You you map
the entire world, so you have complete

1124
01:02:36,160 --> 01:02:39,597
3D model of everything. So the only
thing that the perception system has to

1125
01:02:39,680 --> 01:02:41,757
take care of is moving objects and and

1126
01:02:41,840 --> 01:02:46,867
and construction and sort of you know
things that that weren't in your map

1127
01:02:46,950 --> 01:02:49,357
[snorts] and you can
engineer a good you know

1128
01:02:49,440 --> 01:02:52,957
slam system and all that stuff right so
so that's kind of the current approach

1129
01:02:53,040 --> 01:02:56,077
that's closest to some level of autonomy
but I think eventually the long-term

1130
01:02:56,160 --> 01:03:00,077
solution is going to rely more and more on

1131
01:03:00,160 --> 01:03:03,197
learning and possibly using
a combination of self-supervised learning

1132
01:03:03,280 --> 01:03:05,837
and model based

1133
01:03:05,920 --> 01:03:11,117
reinforcement or something like that >>
but ultimately learning will be at not

1134
01:03:11,200 --> 01:03:14,397
just at the core but really
the fundamental part of the system.

1135
01:03:14,480 --> 01:03:17,517
>> Yeah, it already is but
it'll become more and more.

1136
01:03:17,600 --> 01:03:21,277
>> What do you think it takes to build
a system with human level intelligence?

1137
01:03:21,360 --> 01:03:23,517
You talked about the AI system in the

1138
01:03:23,600 --> 01:03:28,477
movie her being way out of reach our
current reach. This might be outdated as

1139
01:03:28,560 --> 01:03:30,557
well but uh >> it's
still way out of reach.

1140
01:03:30,640 --> 01:03:33,197
>> It's still way out of reach. Uh what

1141
01:03:33,280 --> 01:03:38,157
would it take to build her, do you think?
Um, so I can tell you the first

1142
01:03:38,240 --> 01:03:41,357
two obstacles that we have to clear,
but I don't know how many obstacles there

1143
01:03:41,440 --> 01:03:44,637
are after this. So the image I usually
use is that there's a bunch of mountains

1144
01:03:44,720 --> 01:03:47,677
that we have to climb and we can see
the first one, but we don't know if there

1145
01:03:47,760 --> 01:03:50,237
are 50 mountains behind it or not. And

1146
01:03:50,320 --> 01:03:52,077
this might be a good sort of metaphor

1147
01:03:52,160 --> 01:03:55,677
for why AI researchers in the past have

1148
01:03:55,760 --> 01:03:57,517
been overly optimistic about the result

1149
01:03:57,600 --> 01:04:00,557
of of AI.

1150
01:04:00,640 --> 01:04:03,757
You know for example um Newan Simon

1151
01:04:03,840 --> 01:04:08,397
right wrote the general problem solver and
they called it the general problem solver

1152
01:04:08,480 --> 01:04:11,357
>> general problem solver >> okay
and of course the first thing you

1153
01:04:11,440 --> 01:04:13,997
realize is that all the problems you
want to solve are exponential and so you

1154
01:04:14,080 --> 01:04:17,277
can't actually use it
for anything useful >> but you know

1155
01:04:17,360 --> 01:04:21,117
>> yeah so yeah all you see is the first
peak so what are the first couple of

1156
01:04:21,200 --> 01:04:24,477
peaks for her >> um so
the first peak which is precisely

1157
01:04:24,560 --> 01:04:27,837
what I'm working on is uh self-s
supervised learning how how do we get

1158
01:04:27,920 --> 01:04:29,677
machines to learn models of the world by

1159
01:04:29,760 --> 01:04:32,077
observation kind of like babies and like

1160
01:04:32,160 --> 01:04:35,117
young animals.

1161
01:04:35,200 --> 01:04:38,317
So we've been working with um you know

1162
01:04:38,400 --> 01:04:40,797
cognitive scientists.

1163
01:04:40,880 --> 01:04:43,437
So this Emanuel Dup who's at fair in

1164
01:04:43,520 --> 01:04:46,317
Paris is uh

1165
01:04:46,400 --> 01:04:49,037
halftime he's also a researcher in

1166
01:04:49,120 --> 01:04:52,477
French university and he he um he has

1167
01:04:52,560 --> 01:04:55,517
this chart that shows that which how

1168
01:04:55,600 --> 01:04:58,077
many months of of life baby humans can

1169
01:04:58,160 --> 01:05:00,477
learn different concepts and you can you

1170
01:05:00,560 --> 01:05:03,997
can measure this sort of various ways.

1171
01:05:04,080 --> 01:05:08,717
So things like um um
distinguishing animate objects from

1172
01:05:08,800 --> 01:05:12,317
inanimate inanimate object you can you
can tell the difference at age two three

1173
01:05:12,400 --> 01:05:15,837
months whether an object is going to stay

1174
01:05:15,920 --> 01:05:18,797
stable is going to fall you know about

1175
01:05:18,880 --> 01:05:23,517
four months you can tell you know there
various things like this and then things

1176
01:05:23,600 --> 01:05:26,957
like gravity the fact that objects are
not supposed to float in the air but are

1177
01:05:27,040 --> 01:05:30,957
supposed to fall you learn this around
the age of eight or nine months if you

1178
01:05:31,040 --> 01:05:35,037
look at a lot of you know eight-month
old babies You give them a bunch of toys

1179
01:05:35,120 --> 01:05:37,837
on their high chair. First thing they
do is they throw them on the ground and

1180
01:05:37,920 --> 01:05:40,157
they look at them. >> It's because,
you know, they're learning

1181
01:05:40,240 --> 01:05:43,357
about actively learning about gravity.

1182
01:05:43,440 --> 01:05:46,637
>> Gravity. Yeah. >> Okay.
So, they're not trying to annoy

1183
01:05:46,720 --> 01:05:49,677
you, but they, you know,
they need to do the experiment, right?

1184
01:05:49,760 --> 01:05:51,917
>> Yeah. >> Uh, so, you know,
how do we get machines

1185
01:05:52,000 --> 01:05:55,277
to learn like babies?
mostly by observation with a little bit of

1186
01:05:55,360 --> 01:05:58,877
interaction and learning those those
those models of the world because I

1187
01:05:58,960 --> 01:06:03,677
think that's really a crucial piece
of an intelligent autonomous system. So if

1188
01:06:03,760 --> 01:06:07,037
you think about the architecture of an
intelligent autonomous system, it needs

1189
01:06:07,120 --> 01:06:08,557
to have a predictive model of the world.

1190
01:06:08,640 --> 01:06:12,237
So something that says here is a world
at time t here is a set of world at time

1191
01:06:12,320 --> 01:06:16,797
t plus one if I take this action >>
and it's not a single answer it can be a

1192
01:06:16,880 --> 01:06:18,477
>> yeah it can be distribution. Yeah.

1193
01:06:18,560 --> 01:06:21,037
>> Yeah. Well we don't know how
to represent distributions in high

1194
01:06:21,120 --> 01:06:24,397
dimension spaces. So it's got to be
something weaker than that. Okay.

1195
01:06:24,480 --> 01:06:28,157
>> But with some some
representation of uncertainty.

1196
01:06:28,240 --> 01:06:31,917
>> If you have that then you can do
what optimal control theorist call model

1197
01:06:32,000 --> 01:06:36,157
predictive control which means that you
can run your model with a hypothesis for

1198
01:06:36,240 --> 01:06:39,997
a sequence of action and then see
the result. Now what you need the other

1199
01:06:40,080 --> 01:06:43,917
thing you need is some sort of objective
that you want to optimize. Am I reaching

1200
01:06:44,000 --> 01:06:48,397
the goal of grabbing this object? Am I
minimizing energy? Am I whatever? Right?

1201
01:06:48,480 --> 01:06:53,037
So there is some sort of objective that
you have to minimize and so in your head

1202
01:06:53,120 --> 01:06:55,757
if you have this model you can figure
out the sequence of action that will

1203
01:06:55,840 --> 01:07:00,077
optimize your objective.
That objective is something that

1204
01:07:00,160 --> 01:07:03,757
ultimately is rooted in your basel
ganglia at least in the human brain.

1205
01:07:03,840 --> 01:07:08,797
That's that's what it is. Basel ganglia
computes your level of contentment or or

1206
01:07:08,880 --> 01:07:12,717
miscontentment. I don't know if
that's a word. Uh unhappiness. Okay.

1207
01:07:12,800 --> 01:07:14,477
>> Yeah. Yeah. >> Discontentment.

1208
01:07:14,560 --> 01:07:17,117
>> Discontentment maybe. Um and so your

1209
01:07:17,200 --> 01:07:20,397
entire behavior is driven towards

1210
01:07:20,480 --> 01:07:22,317
kind of minimizing that objective which

1211
01:07:22,400 --> 01:07:25,037
is maximizing your contentment computed

1212
01:07:25,120 --> 01:07:30,077
by your your bas ganglia.
>> And what you have is an objective

1213
01:07:30,160 --> 01:07:33,197
function which is basically a predictor
of what your bas ganglia is going to

1214
01:07:33,280 --> 01:07:35,917
tell you. >> So you're not
going to put your hand on

1215
01:07:36,000 --> 01:07:40,237
fire because you know it's going to you
know it's going to burn and you're going

1216
01:07:40,320 --> 01:07:43,517
to get hurt. And you're predicting this
because of your model of the world and

1217
01:07:43,600 --> 01:07:45,757
your your sort of predictor of this

1218
01:07:45,840 --> 01:07:50,557
objective, right? So you if you have
those you have those three components,

1219
01:07:50,640 --> 01:07:52,637
you have the four components. You have

1220
01:07:52,720 --> 01:07:54,797
the the hardwired

1221
01:07:54,880 --> 01:07:59,037
uh contentment objective um computer if

1222
01:07:59,120 --> 01:08:02,637
you want calculator [snorts] and then
you have the three components. One is

1223
01:08:02,720 --> 01:08:04,397
the objective predictor which basically

1224
01:08:04,480 --> 01:08:06,317
predicts your level of contentment. one

1225
01:08:06,400 --> 01:08:10,077
is the model of the world and there's a

1226
01:08:10,160 --> 01:08:12,317
third module I didn't mention which is a

1227
01:08:12,400 --> 01:08:14,877
module that will figure out the best

1228
01:08:14,960 --> 01:08:19,917
course of action to optimize
an objective given your model

1229
01:08:20,000 --> 01:08:21,517
>> okay

1230
01:08:21,600 --> 01:08:24,557
>> call this a policy policy network or

1231
01:08:24,640 --> 01:08:29,677
something like that right [snorts] now
you need those three components to act

1232
01:08:29,760 --> 01:08:33,357
autonomously intelligently and you can
be stupid in three different ways you

1233
01:08:33,440 --> 01:08:36,157
can be stupid because your model
of your of the world is wrong.

1234
01:08:36,240 --> 01:08:38,397
>> Mhm. >> You can be
stupid because your objective

1235
01:08:38,480 --> 01:08:42,237
is not aligned with what
you actually want to achieve.

1236
01:08:42,320 --> 01:08:46,317
>> Okay. Uh >> in humans
that would be a psychopath,

1237
01:08:46,400 --> 01:08:49,757
>> right? >> And then
the the third thing you the

1238
01:08:49,840 --> 01:08:52,797
third way you can be stupid is that you
have the right model, you have the right

1239
01:08:52,880 --> 01:08:56,717
objective, but you're unable to figure
out a course of action to optimize your

1240
01:08:56,800 --> 01:08:58,477
objective given your model. >> Right?

1241
01:08:58,560 --> 01:09:02,637
>> Okay. Um some people
who are in charge of big

1242
01:09:02,720 --> 01:09:04,157
countries actually have all three that

1243
01:09:04,240 --> 01:09:07,357
are wrong. All right. [laughter]

1244
01:09:07,440 --> 01:09:09,757
Uh, which countries? I don't know. Okay.

1245
01:09:09,840 --> 01:09:12,317
So, uh, if we think about this this

1246
01:09:12,400 --> 01:09:16,957
agent, if we think about
the movie her, you've

1247
01:09:17,040 --> 01:09:20,957
criticized the art
project that is Sophia the

1248
01:09:21,040 --> 01:09:23,837
robot. And what that project essentially

1249
01:09:23,920 --> 01:09:26,797
does is uses our natural inclination to

1250
01:09:26,880 --> 01:09:30,877
anthropomorphize things
that look like human and give

1251
01:09:30,960 --> 01:09:32,957
them more. Do you think that could be

1252
01:09:33,040 --> 01:09:35,677
used by AI systems like in the movie

1253
01:09:35,760 --> 01:09:40,797
Her? So, do you think
that body is needed to

1254
01:09:40,880 --> 01:09:43,997
create a feeling of intelligence?

1255
01:09:44,080 --> 01:09:47,437
>> Uh, well, if Sophia was just an art
piece, I would have no problem with it,

1256
01:09:47,520 --> 01:09:50,157
but it's presented as something else.

1257
01:09:50,240 --> 01:09:55,117
>> Let me add that comment real quick.
If creators of Sophia could change

1258
01:09:55,200 --> 01:09:58,637
something about their marketing
or behavior in general, what would it be?

1259
01:09:58,720 --> 01:10:01,787
What What's >> Oh, just about everything.

1260
01:10:01,870 --> 01:10:04,877
[laughter] I mean,

1261
01:10:04,960 --> 01:10:07,197
don't you think here's a tough question.

1262
01:10:07,280 --> 01:10:09,837
Let me So, I agree with you. So, Sophia

1263
01:10:09,920 --> 01:10:15,517
is not the general public feels
that Sophia can do way more than

1264
01:10:15,600 --> 01:10:17,437
she actually can. >> That's right.

1265
01:10:17,520 --> 01:10:20,237
>> And the people who created Sophia are

1266
01:10:20,320 --> 01:10:22,077
not

1267
01:10:22,160 --> 01:10:25,517
honestly publicly communicating uh

1268
01:10:25,600 --> 01:10:28,717
trying to teach the public. But

1269
01:10:28,800 --> 01:10:31,917
>> here's a tough question.
Don't you think

1270
01:10:32,000 --> 01:10:36,957
uh this the same thing is uh scientists

1271
01:10:37,040 --> 01:10:41,037
in industry and research are
taking advantage of the same mis

1272
01:10:41,120 --> 01:10:43,117
misunderstanding in the public when they

1273
01:10:43,200 --> 01:10:47,117
create AI companies or publish stuff. um

1274
01:10:47,200 --> 01:10:49,197
>> some companies yes I mean there is no

1275
01:10:49,280 --> 01:10:54,317
sense of there's no desire to delude
there's no desire to to kind of

1276
01:10:54,400 --> 01:10:57,757
overclaim what something is done right
you publish a paper on AI that you know

1277
01:10:57,840 --> 01:11:01,517
has this result on imageet you know it's
pretty clear I mean it's not not even

1278
01:11:01,600 --> 01:11:06,237
interesting anymore but >> you
know I I don't think there is that

1279
01:11:06,320 --> 01:11:08,317
I mean the reviewers are generally not

1280
01:11:08,400 --> 01:11:10,877
very forgiving of

1281
01:11:10,960 --> 01:11:13,357
>> of you know unsupported claims of of

1282
01:11:13,440 --> 01:11:15,757
this type and but there are certainly

1283
01:11:15,840 --> 01:11:18,557
quite a few startups that have had a

1284
01:11:18,640 --> 01:11:21,197
huge amount of hype around this that uh

1285
01:11:21,280 --> 01:11:25,597
I find extremely damaging and I've been
calling it out when I've seen it. Um so

1286
01:11:25,680 --> 01:11:28,637
yeah, but to go back
to your original question like

1287
01:11:28,720 --> 01:11:32,957
>> the necessity of embodiment. I think
I don't think embodiment is necessary. I

1288
01:11:33,040 --> 01:11:36,637
think grounding is necessary. So I don't
think we're going to get machines that

1289
01:11:36,720 --> 01:11:39,517
really understand language without some
level of grounding in the real world.

1290
01:11:39,600 --> 01:11:43,517
And it's not clear to me that language
is a high enough bandwidth medium to

1291
01:11:43,600 --> 01:11:46,077
communicate how the real world works.

1292
01:11:46,160 --> 01:11:49,357
Um I think for this >> can you
talk to ground what grounding means?

1293
01:11:49,440 --> 01:11:51,837
>> So grounding means
that so there is this

1294
01:11:51,920 --> 01:11:54,237
classic problem of common sense

1295
01:11:54,320 --> 01:11:56,717
reasoning you know the the winner winner

1296
01:11:56,800 --> 01:12:02,077
schema right and so I tell you
the the trophy doesn't fit in the suitcase

1297
01:12:02,160 --> 01:12:04,877
because it's too it's too big
or the trophy doesn't fit in the suitcase

1298
01:12:04,960 --> 01:12:09,117
because it's too small and the it in the
first case refers to the trophy in the

1299
01:12:09,200 --> 01:12:12,397
second case to the suitcase.
And the reason you can figure this out is

1300
01:12:12,480 --> 01:12:14,877
because you know what the trophy and the
suitcase are. you know one is supposed

1301
01:12:14,960 --> 01:12:18,957
to fit in the other one and you know the
notion of size and a big object doesn't

1302
01:12:19,040 --> 01:12:22,477
fit in a small object unless it's
a TARDIS you know things like that right

1303
01:12:22,560 --> 01:12:25,117
so you have this this knowledge of how

1304
01:12:25,200 --> 01:12:27,357
the world works of geometry and things

1305
01:12:27,440 --> 01:12:30,477
like that um I don't believe you can

1306
01:12:30,560 --> 01:12:32,157
learn everything about the world by just

1307
01:12:32,240 --> 01:12:34,877
being told in language how the world

1308
01:12:34,960 --> 01:12:39,437
works I think you need some uh low-level
perception of the world you know be it

1309
01:12:39,520 --> 01:12:41,437
visual touch you know whatever but some

1310
01:12:41,520 --> 01:12:46,557
higher bandwidth perception of the So
by reading all the world's text, you still

1311
01:12:46,640 --> 01:12:48,317
may not have enough information.

1312
01:12:48,400 --> 01:12:50,557
>> That's right. There's a lot of things

1313
01:12:50,640 --> 01:12:52,717
that just will never appear in text and

1314
01:12:52,800 --> 01:12:55,757
that you can't really infer. So I think

1315
01:12:55,840 --> 01:13:00,077
>> uh common sense will emerge from >>
you know certainly a lot of language

1316
01:13:00,160 --> 01:13:02,317
interaction but also with watching

1317
01:13:02,400 --> 01:13:07,357
videos or perhaps even interacting in
in virtual environments and possibly you

1318
01:13:07,440 --> 01:13:10,717
know robot interacting in the real world
but I don't actually believe necessarily

1319
01:13:10,800 --> 01:13:14,637
that this last one is absolutely
necessary but I think there's a the need

1320
01:13:14,720 --> 01:13:19,677
for some grounding >>
but the final product doesn't

1321
01:13:19,760 --> 01:13:23,197
necessarily need to be embodied
you're saying it just needs to have an

1322
01:13:23,280 --> 01:13:25,117
awareness a grounding to

1323
01:13:25,200 --> 01:13:28,157
>> but he needs to know
how the world works

1324
01:13:28,240 --> 01:13:30,317
to have you know to not be frustrated

1325
01:13:30,400 --> 01:13:35,197
frustrating to talk to >> and uh
you talked about emotions being

1326
01:13:35,280 --> 01:13:38,957
important that's that's
a whole another topic

1327
01:13:39,040 --> 01:13:41,437
>> well so you know I talked about this uh

1328
01:13:41,520 --> 01:13:45,197
the basil ganglia ganglia as the

1329
01:13:45,280 --> 01:13:48,557
you your your you know the thing that
calc you know calculates your level of

1330
01:13:48,640 --> 01:13:52,557
mix contentment contentment and then
there is this other module that sort of

1331
01:13:52,640 --> 01:13:55,677
tries to do a prediction of whether
you're going to be content or not.

1332
01:13:55,760 --> 01:13:57,597
That's the source of some emotion. So

1333
01:13:57,680 --> 01:14:03,117
fear, for example, is an anticipation
of bad things that can happen to you,

1334
01:14:03,200 --> 01:14:05,917
right? >> Uh you have
this inkling that there is

1335
01:14:06,000 --> 01:14:08,637
some chance that something really bad
is going to happen to you and that creates

1336
01:14:08,720 --> 01:14:10,637
fear. >> When you know
for sure that something

1337
01:14:10,720 --> 01:14:14,557
bad is going to happen to you, you kind
of give up, right? It's not fear anymore.

1338
01:14:14,640 --> 01:14:16,637
>> It's uncertainty that creates fear.

1339
01:14:16,720 --> 01:14:19,597
>> So So the punch line is yes, we're
not going to have autonomous intelligence

1340
01:14:19,680 --> 01:14:24,157
without emotions. Okay. [sighs and gasps]

1341
01:14:24,240 --> 01:14:28,157
>> Whatever the heck emotions are. So you
mentioned very practical things of fear,

1342
01:14:28,240 --> 01:14:30,557
but there's a lot of other mess around it.

1343
01:14:30,640 --> 01:14:33,597
>> But there are kind of the
results of, you know, drives.

1344
01:14:33,680 --> 01:14:36,077
>> Yeah. There's deeper
uh biological stuff

1345
01:14:36,160 --> 01:14:40,237
going on and I've talked to a few folks
on this. There's uh there's fascinating

1346
01:14:40,320 --> 01:14:42,877
stuff that ultimately connects to our to

1347
01:14:42,960 --> 01:14:48,077
to our brain.
If we create an AGI system, sorry,

1348
01:14:48,160 --> 01:14:51,517
>> human level intelligence,
>> human level intelligence system

1349
01:14:51,600 --> 01:14:54,237
and you get to ask her one question,

1350
01:14:54,320 --> 01:14:56,957
what would that question be?

1351
01:14:57,040 --> 01:15:01,277
>> You know, I think the the first one
we'll create will probably not be not be

1352
01:15:01,360 --> 01:15:04,637
that smart. They be like
a four-year-old. Okay.

1353
01:15:04,720 --> 01:15:09,997
>> So, you would have to ask her
a question to know she's not that smart.

1354
01:15:10,080 --> 01:15:13,757
>> Yeah. Well, what's
a good question to ask, you

1355
01:15:13,840 --> 01:15:17,917
know, to be >> the cause of wind?

1356
01:15:18,000 --> 01:15:20,637
And if she answers, oh, it's because the

1357
01:15:20,720 --> 01:15:25,677
leaves of the tree are moving and that
creates wind, she's on to something.

1358
01:15:25,760 --> 01:15:29,837
>> And if she says, that's a stupid
question, she's really on to something.

1359
01:15:29,920 --> 01:15:32,317
>> No. And then you tell her actually, you

1360
01:15:32,400 --> 01:15:35,197
know, here is the the real thing. And

1361
01:15:35,280 --> 01:15:37,677
she says, oh yeah, that makes sense.

1362
01:15:37,760 --> 01:15:42,877
So questions that that uh reveal
the ability to do common sense reasoning

1363
01:15:42,960 --> 01:15:47,197
about the physical world. >> Yeah. And
you'll sum it up with a causal inference.

1364
01:15:47,280 --> 01:15:51,597
>> Causal inference. Well, >> it
was a huge honor. Congratulations on

1365
01:15:51,680 --> 01:15:54,397
the touring award, Yan.
Thank you so much for talking today.

1366
01:15:54,480 --> 01:15:56,807
>> Thank you. >> Appreciate it.

1367
01:15:56,890 --> 01:15:58,910
[music]

1368
01:16:04,090 --> 01:16:06,110
>> [music]