1
00:00:00,835 --> 00:00:02,990
Mark Twain summed up
what I take to be

2
00:00:02,990 --> 00:00:06,110
one of the fundamental problems
of cognitive science

3
00:00:06,110 --> 00:00:07,820
with a single witticism.

4
00:00:08,410 --> 00:00:11,492
He said, "There's something
fascinating about science.

5
00:00:11,492 --> 00:00:14,720
One gets such wholesale
returns of conjecture

6
00:00:14,720 --> 00:00:17,924
out of such a trifling
investment in fact."

7
00:00:17,924 --> 00:00:19,509
(Laughter)

8
00:00:20,199 --> 00:00:22,803
Twain meant it as a joke,
of course, but he's right:

9
00:00:22,803 --> 00:00:25,679
There's something
fascinating about science.

10
00:00:25,679 --> 00:00:29,940
From a few bones, we infer
the existence of dinosuars.

11
00:00:30,910 --> 00:00:34,781
From spectral lines,
the composition of nebulae.

12
00:00:35,471 --> 00:00:38,409
From fruit flies,

13
00:00:38,409 --> 00:00:41,352
the mechanisms of heredity,

14
00:00:41,352 --> 00:00:45,601
and from reconstructed images
of blood flowing through the brain,

15
00:00:45,601 --> 00:00:50,309
or in my case, from the behavior
of very young children,

16
00:00:50,309 --> 00:00:53,138
we try to say something about
the fundamental mechanisms

17
00:00:53,138 --> 00:00:54,756
of human cognition.

18
00:00:55,716 --> 00:01:00,475
In particular, in my lab in the Department
of Brain and Cognitive Sciences at MIT,

19
00:01:00,475 --> 00:01:04,129
I have spent the past decade
trying to understand the mystery

20
00:01:04,129 --> 00:01:08,106
of how children learn so much
from so little so quickly.

21
00:01:08,666 --> 00:01:11,644
Because, it turns out that
the fascinating thing about science

22
00:01:11,644 --> 00:01:15,173
is also a fascinating
thing about children,

23
00:01:15,173 --> 00:01:17,754
which, to put a gentler
spin on Mark Twain,

24
00:01:17,754 --> 00:01:22,404
is precisely their ability
to draw rich, abstract inferences

25
00:01:22,404 --> 00:01:27,065
rapidly and accurately
from sparse, noisy data.

26
00:01:28,355 --> 00:01:30,753
I'm going to give you
just two examples today.

27
00:01:30,753 --> 00:01:33,040
One is about a problem of generalization,

28
00:01:33,040 --> 00:01:35,890
and the other is about a problem
of causal reasoning.

29
00:01:35,890 --> 00:01:38,415
And although I'm going to talk
about work in my lab,

30
00:01:38,415 --> 00:01:41,875
this work is inspired by
and indebted to a field.

31
00:01:41,875 --> 00:01:46,158
I'm grateful to mentors, colleagues,
and collaborators around the world.

32
00:01:47,308 --> 00:01:50,282
Let me start with the problem
of generalization.

33
00:01:50,652 --> 00:01:54,785
Generalizing from small samples of data
is the bread and butter of science.

34
00:01:54,785 --> 00:01:57,339
We poll a tiny fraction of the electorate

35
00:01:57,339 --> 00:01:59,660
and we predict the outcome
of national elections.

36
00:02:00,240 --> 00:02:04,165
We see how a handful of patients
responds to treatment in a clinical trial,

37
00:02:04,165 --> 00:02:07,230
and we bring drugs to a national market.

38
00:02:07,230 --> 00:02:11,595
But this only works if our sample
is randomly drawn from the population.

39
00:02:11,595 --> 00:02:14,330
If our sample is cherry-picked
in some way --

40
00:02:14,330 --> 00:02:16,402
say, we poll only urban voters,

41
00:02:16,402 --> 00:02:20,790
or say, in our clinical trials
for treatments for heart disease,

42
00:02:20,790 --> 00:02:22,671
we include only men --

43
00:02:22,671 --> 00:02:25,829
the results may not generalize
to the broader population.

44
00:02:26,479 --> 00:02:30,060
So scientists care whether evidence
is randomly sampled or not,

45
00:02:30,060 --> 00:02:32,075
but what does that have to do with babies?

46
00:02:32,585 --> 00:02:37,206
Well, babies have to generalize
from small samples of data all the time.

47
00:02:37,206 --> 00:02:40,364
They see a few rubber ducks
and learn that they float,

48
00:02:40,364 --> 00:02:43,939
or a few balls and learn that they bounce.

49
00:02:43,939 --> 00:02:46,890
And they develop expectations
about ducks and balls

50
00:02:46,890 --> 00:02:49,606
that they're going to extend
to rubber ducks and balls

51
00:02:49,606 --> 00:02:51,485
for the rest of their lives.

52
00:02:51,485 --> 00:02:55,224
And the kinds of generalizations
babies have to make about ducks and balls

53
00:02:55,224 --> 00:02:57,313
they have to make about almost everything:

54
00:02:57,313 --> 00:03:01,230
shoes and ships and ceiling wax
and cabbages and kings.

55
00:03:02,200 --> 00:03:05,161
So do babies care whether
the tiny bit of evidence they see

56
00:03:05,161 --> 00:03:08,853
is plausibly representative
of a larger population?

57
00:03:09,763 --> 00:03:11,663
Let's find out.

58
00:03:11,663 --> 00:03:13,386
I'm going to show you two movies,

59
00:03:13,386 --> 00:03:15,848
one from each of two conditions
of an experiment,

60
00:03:15,848 --> 00:03:18,286
and because you're going to see
just two movies,

61
00:03:18,286 --> 00:03:20,422
you're going to see just two babies,

62
00:03:20,422 --> 00:03:24,369
and any two babies differ from each other
in innumerable ways.

63
00:03:24,369 --> 00:03:27,420
But these babies, of course,
here stand in for groups of babies,

64
00:03:27,420 --> 00:03:29,315
and the differences you're going to see

65
00:03:29,315 --> 00:03:34,510
represent average group differences
in babies' behavior across conditions.

66
00:03:35,160 --> 00:03:37,743
In each movie, you're going to see
a baby doing maybe

67
00:03:37,743 --> 00:03:41,203
just exactly what you might
expect a baby to do,

68
00:03:41,203 --> 00:03:45,220
and we can hardly make babies
more magical than they already are.

69
00:03:46,090 --> 00:03:48,100
But to my mind the magical thing,

70
00:03:48,100 --> 00:03:50,189
and what I want you to pay attention to,

71
00:03:50,189 --> 00:03:53,300
is the contrast between
these two conditions,

72
00:03:53,300 --> 00:03:56,829
because the only thing
that differs between these two movies

73
00:03:56,829 --> 00:04:00,295
is the statistical evidence
the babies are going to observe.

74
00:04:01,425 --> 00:04:04,608
We're going to show babies
a box of blue and yellow balls,

75
00:04:04,608 --> 00:04:09,228
and my then-graduate student,
now colleague at Stanford, Hyowon Gweon,

76
00:04:09,228 --> 00:04:12,305
is going to pull three blue balls
in a row out of this box,

77
00:04:12,305 --> 00:04:15,428
and when she pulls those balls out,
she's going to squeeze them,

78
00:04:15,428 --> 00:04:17,541
and the balls are going to squeak.

79
00:04:17,541 --> 00:04:20,304
And if you're a baby,
that's like a TED Talk.

80
00:04:20,304 --> 00:04:22,208
It doesn't get better than that.

81
00:04:22,208 --> 00:04:24,769
(Laughter)

82
00:04:26,968 --> 00:04:30,627
But the important point is it's really
easy to pull three blue balls in a row

83
00:04:30,627 --> 00:04:32,932
out of a box of mostly blue balls.

84
00:04:32,932 --> 00:04:34,992
You could do that with your eyes closed.

85
00:04:34,992 --> 00:04:37,988
It's plausibly a random sample
from this population.

86
00:04:37,988 --> 00:04:41,720
And if you can reach into a box at random
and pull out things that squeak,

87
00:04:41,720 --> 00:04:44,559
then maybe everything in the box squeaks.

88
00:04:44,559 --> 00:04:48,209
So maybe babies should expect
those yellow balls to squeak as well.

89
00:04:48,209 --> 00:04:50,728
Now, those yellow balls
have funny sticks on the end,

90
00:04:50,728 --> 00:04:53,585
so babies could do other things
with them if they wanted to.

91
00:04:53,585 --> 00:04:55,416
They could pound them or whack them.

92
00:04:55,416 --> 00:04:58,002
But let's see what the baby does.

93
00:05:00,548 --> 00:05:03,891
(Video) Hyowon Gweon: See this?
(Ball squeaks)

94
00:05:04,531 --> 00:05:07,576
Did you see that?
(Ball squeaks)

95
00:05:08,036 --> 00:05:11,102
Cool.

96
00:05:12,706 --> 00:05:14,656
See this one?

97
00:05:14,656 --> 00:05:16,537
(Ball squeaks)

98
00:05:16,537 --> 00:05:19,190
Wow.

99
00:05:21,854 --> 00:05:23,967
Laura Schulz: Told you. (Laughs)

100
00:05:23,967 --> 00:05:27,998
(Video) HG: See this one?
(Ball squeaks)

101
00:05:27,998 --> 00:05:32,617
Hey Clara, this one's for you.
You can go ahead and play.

102
00:05:39,854 --> 00:05:44,219
(Laughter)

103
00:05:44,219 --> 00:05:47,214
LS: I don't even have to talk, right?

104
00:05:47,214 --> 00:05:50,113
All right, it's nice that babies
will generalize properties

105
00:05:50,113 --> 00:05:51,641
of blue balls to yellow balls,

106
00:05:51,641 --> 00:05:54,737
and it's impressive that babies
can learn from imitating us,

107
00:05:54,737 --> 00:05:58,406
but we've known those things about babies
for a very long time.

108
00:05:58,406 --> 00:06:00,217
The really interesting question

109
00:06:00,217 --> 00:06:03,069
is what happens when we show babies
exactly the same thing,

110
00:06:03,069 --> 00:06:06,680
and we can ensure it's exactly the same
because we have a secret compartment

111
00:06:06,680 --> 00:06:08,790
and we actually pull the balls from there,

112
00:06:08,790 --> 00:06:12,268
but this time, all we change
is the apparent population

113
00:06:12,268 --> 00:06:15,170
from which that evidence was drawn.

114
00:06:15,170 --> 00:06:18,723
This time, we're going to show babies
three blue balls

115
00:06:18,723 --> 00:06:22,107
pulled out of a box
of mostly yellow balls,

116
00:06:22,107 --> 00:06:23,429
and guess what?

117
00:06:23,429 --> 00:06:26,269
You [probably won't] randomly draw
three blue balls in a row

118
00:06:26,269 --> 00:06:28,753
out of a box of mostly yellow balls.

119
00:06:28,753 --> 00:06:32,500
That is not plausibly
randomly sampled evidence.

120
00:06:32,500 --> 00:06:37,623
That evidence suggests that maybe Hyowon
was deliberately sampling the blue balls.

121
00:06:37,623 --> 00:06:40,206
Maybe there's something special
about the blue balls.

122
00:06:40,846 --> 00:06:43,822
Maybe only the blue balls squeak.

123
00:06:43,822 --> 00:06:45,717
Let's see what the baby does.

124
00:06:45,717 --> 00:06:48,621
(Video) HG: See this?
(Ball squeaks)

125
00:06:50,851 --> 00:06:53,496
See this toy?
(Ball squeaks)

126
00:06:53,496 --> 00:06:58,976
Oh, that was cool. See?
(Ball squeaks)

127
00:06:58,976 --> 00:07:03,370
Now this one's for you to play.
You can go ahead and play.

128
00:07:06,074 --> 00:07:12,421
(Fussing)
(Laughter)

129
00:07:14,901 --> 00:07:17,649
LS: So you just saw
two 15-month-old babies

130
00:07:17,649 --> 00:07:19,591
do entirely different things

131
00:07:19,591 --> 00:07:23,190
based only on the probability
of the sample they observed.

132
00:07:23,190 --> 00:07:25,511
Let me show you the experimental results.

133
00:07:25,511 --> 00:07:28,275
On the vertical axis, you'll see
the percentage of babies

134
00:07:28,275 --> 00:07:30,805
who squeezed the ball in each condition,

135
00:07:30,805 --> 00:07:34,520
and as you'll see, babies are much
more likely to generalize the evidence

136
00:07:34,520 --> 00:07:37,655
when it's plausibly representative
of the population

137
00:07:37,655 --> 00:07:41,393
than when the evidence
is clearly cherry-picked.

138
00:07:41,393 --> 00:07:43,808
And this leads to a fun prediction:

139
00:07:43,808 --> 00:07:48,676
Suppose you pulled just one blue ball
out of the mostly yellow box.

140
00:07:48,896 --> 00:07:52,765
You [probably won't] pull three blue balls
in a row at random out of a yellow box,

141
00:07:52,765 --> 00:07:55,220
but you could randomly sample
just one blue ball.

142
00:07:55,220 --> 00:07:57,190
That's not an improbable sample.

143
00:07:57,190 --> 00:07:59,414
And if you could reach into
a box at random

144
00:07:59,414 --> 00:08:03,401
and pull out something that squeaks,
maybe everything in the box squeaks.

145
00:08:03,875 --> 00:08:08,320
So even though babies are going to see
much less evidence for squeaking,

146
00:08:08,320 --> 00:08:10,562
and have many fewer actions to imitate

147
00:08:10,562 --> 00:08:13,905
in this one ball condition than in
the condition you just saw,

148
00:08:13,905 --> 00:08:17,797
we predicted that babies themselves
would squeeze more,

149
00:08:17,797 --> 00:08:20,691
and that's exactly what we found.

150
00:08:20,691 --> 00:08:25,102
So 15-month-old babies,
in this respect, like scientists,

151
00:08:25,102 --> 00:08:28,190
care whether evidence
is randomly sampled or not,

152
00:08:28,190 --> 00:08:31,697
and they use this to develop
expectations about the world:

153
00:08:31,697 --> 00:08:33,879
what squeaks and what doesn't,

154
00:08:33,879 --> 00:08:37,024
what to explore and what to ignore.

155
00:08:38,384 --> 00:08:40,450
Let me show you another example now,

156
00:08:40,450 --> 00:08:43,180
this time about a problem
of causal reasoning.

157
00:08:43,180 --> 00:08:45,619
And it starts with a problem
of confounded evidence

158
00:08:45,619 --> 00:08:47,291
that all of us have,

159
00:08:47,291 --> 00:08:49,311
which is that we are part of the world.

160
00:08:49,311 --> 00:08:52,747
And this might not seem like a problem
to you, but like most problems,

161
00:08:52,747 --> 00:08:55,084
it's only a problem when things go wrong.

162
00:08:55,464 --> 00:08:57,275
Take this baby, for instance.

163
00:08:57,275 --> 00:08:58,980
Things are going wrong for him.

164
00:08:58,980 --> 00:09:01,251
He would like to make
this toy go, and he can't.

165
00:09:01,251 --> 00:09:03,780
I'll show you a few-second clip.

166
00:09:09,340 --> 00:09:11,260
And there's two possibilities, broadly:

167
00:09:11,260 --> 00:09:13,894
Maybe he's doing something wrong,

168
00:09:13,894 --> 00:09:18,110
or maybe there's something
wrong with the toy.

169
00:09:18,110 --> 00:09:20,221
So in this next experiment,

170
00:09:20,221 --> 00:09:23,518
we're going to give babies
just a tiny bit of statistical data

171
00:09:23,518 --> 00:09:26,100
supporting one hypothesis over the other,

172
00:09:26,100 --> 00:09:29,555
and we're going to see if babies
can use that to make different decisions

173
00:09:29,555 --> 00:09:31,389
about what to do.

174
00:09:31,389 --> 00:09:33,411
Here's the setup.

175
00:09:34,071 --> 00:09:37,101
Hyowon is going to try to make
the toy go and succeed.

176
00:09:37,101 --> 00:09:40,421
I am then going to try twice
and fail both times,

177
00:09:40,421 --> 00:09:43,533
and then Hyowon is going
to try again and succeed,

178
00:09:43,533 --> 00:09:46,705
and this roughly sums up my relationship
to my graduate students

179
00:09:46,705 --> 00:09:49,540
in technology across the board.

180
00:09:50,030 --> 00:09:53,322
But the important point here is
it provides a little bit of evidence

181
00:09:53,322 --> 00:09:56,990
that the problem isn't with the toy,
it's with the person.

182
00:09:56,990 --> 00:09:59,340
Some people can make this toy go,

183
00:09:59,340 --> 00:10:00,299
and some can't.

184
00:10:00,799 --> 00:10:04,212
Now, when the baby gets the toy,
he's going to have a choice.

185
00:10:04,212 --> 00:10:06,400
His mom is right there,

186
00:10:06,400 --> 00:10:09,715
so he can go ahead and hand off the toy
and change the person,

187
00:10:09,715 --> 00:10:12,873
but there's also going to be
another toy at the end of that cloth,

188
00:10:12,873 --> 00:10:16,425
and he can pull the cloth towards him
and change the toy.

189
00:10:16,425 --> 00:10:18,515
So let's see what the baby does.

190
00:10:18,515 --> 00:10:22,698
(Video) HG: Two, three. Go!
(Music)

191
00:10:22,698 --> 00:10:25,829
LS: One, two, three, go!

192
00:10:25,829 --> 00:10:33,211
Arthur, I'm going to try again.
One, two, three, go!

193
00:10:33,677 --> 00:10:36,277
YG: Arthur, let me try again, okay?

194
00:10:36,277 --> 00:10:40,827
One, two, three, go!
(Music)

195
00:10:41,583 --> 00:10:43,466
Look at that. Remember these toys?

196
00:10:43,466 --> 00:10:46,730
See these toys? Yeah, I'm going
to put this one over here,

197
00:10:46,730 --> 00:10:48,792
and I'm going to give this one to you.

198
00:10:48,792 --> 00:10:51,127
You can go ahead and play.

199
00:11:11,213 --> 00:11:15,950
LS: Okay, Laura, but of course,
babies love their mommies.

200
00:11:15,950 --> 00:11:18,132
Of course babies give toys
to their mommies

201
00:11:18,132 --> 00:11:20,162
when they can't make them work.

202
00:11:20,162 --> 00:11:23,755
So again, the really important question
is what happens when we change

203
00:11:23,755 --> 00:11:26,909
the statistical data ever so slightly.

204
00:11:26,909 --> 00:11:30,996
This time, babies are going to see the toy
work and fail in exactly the same order,

205
00:11:30,996 --> 00:11:33,411
but we're changing
the distribution of evidence.

206
00:11:33,411 --> 00:11:37,822
This time, Hyowon is going to succeed
once and fail once, and so am I.

207
00:11:37,822 --> 00:11:43,459
And this suggests it doesn't matter
who tries this toy, the toy is broken.

208
00:11:43,459 --> 00:11:45,345
It doesn't work all the time.

209
00:11:45,345 --> 00:11:47,310
Again, the baby's going to have a choice.

210
00:11:47,310 --> 00:11:50,706
Her mom is right next to her,
so she can change the person,

211
00:11:50,706 --> 00:11:53,910
and there's going to be another toy
at the end of the cloth.

212
00:11:53,910 --> 00:11:55,288
Let's watch what she does.

213
00:11:55,288 --> 00:11:59,636
(Video) HG: Two, three, go!
(Music)

214
00:11:59,636 --> 00:12:04,620
Let me try one more time.
One, two, three, go!

215
00:12:05,460 --> 00:12:07,157
Hmm.

216
00:12:07,950 --> 00:12:10,642
LS: Let me try, Clara.

217
00:12:10,642 --> 00:12:14,587
One, two, three, go!

218
00:12:15,265 --> 00:12:17,200
Hmm, let me try again.

219
00:12:17,200 --> 00:12:22,870
One, two, three, go!
(Music)

220
00:12:23,009 --> 00:12:25,242
HG: I'm going
to put this one over here,

221
00:12:25,242 --> 00:12:27,243
and I'm going to give this one to you.

222
00:12:27,243 --> 00:12:30,840
You can go ahead and play.

223
00:12:46,376 --> 00:12:51,273
(Applause)

224
00:12:52,993 --> 00:12:55,385
LS: Let me show you
the experimental results.

225
00:12:55,385 --> 00:12:57,860
On the vertical axis,
you'll see the distribution

226
00:12:57,860 --> 00:13:00,437
of children's choices in each condition,

227
00:13:00,437 --> 00:13:04,988
and you'll see that the distribution
of the choices children make

228
00:13:04,988 --> 00:13:07,775
depends on the evidence they observe.

229
00:13:07,775 --> 00:13:09,632
So in the second year of life,

230
00:13:09,632 --> 00:13:12,209
babies can use a tiny bit
of statistical data

231
00:13:12,209 --> 00:13:15,576
to decide between two
fundamentally different strategies

232
00:13:15,576 --> 00:13:17,457
for acting in the world:

233
00:13:17,457 --> 00:13:20,200
asking for help and exploring.

234
00:13:21,700 --> 00:13:25,134
I've just shown you
two laboratory experiments

235
00:13:25,134 --> 00:13:28,825
out of literally hundreds in the field
that make similar points,

236
00:13:28,825 --> 00:13:31,217
because the really critical point

237
00:13:31,217 --> 00:13:36,325
is that children's ability
to make rich inferences from sparse data

238
00:13:36,325 --> 00:13:41,666
underlies all the species-specific
cultural learning that we do.

239
00:13:41,666 --> 00:13:46,263
Children learn about new tools
from just a few examples.

240
00:13:46,263 --> 00:13:50,980
They learn new causal relationships
from just a few examples.

241
00:13:51,928 --> 00:13:56,799
They even learn new words,
in this case in American Sign Language.

242
00:13:56,799 --> 00:13:59,110
I want to close with just two points.

243
00:14:00,050 --> 00:14:03,738
If you've been following my world,
the field of brain and cognitive sciences,

244
00:14:03,738 --> 00:14:05,665
for the past few years,

245
00:14:05,665 --> 00:14:08,080
three big ideas will have come
to your attention.

246
00:14:08,080 --> 00:14:11,516
The first is that this is
the era of the brain.

247
00:14:11,516 --> 00:14:15,185
And indeed, there have been
staggering discoveries in neuroscience:

248
00:14:15,185 --> 00:14:18,621
localizing functionally specialized
regions of cortex,

249
00:14:18,621 --> 00:14:21,222
turning mouse brains transparent,

250
00:14:21,222 --> 00:14:24,998
activating neurons with light.

251
00:14:24,998 --> 00:14:26,994
A second big idea

252
00:14:26,994 --> 00:14:31,098
is that this is the era of big data
and machine learning,

253
00:14:31,098 --> 00:14:34,239
and machine learning promises
to revolutionize our understanding

254
00:14:34,239 --> 00:14:38,906
of everything from social networks
to epidemiology.

255
00:14:38,906 --> 00:14:41,599
And maybe, as it tackles problems
of scene understanding

256
00:14:41,599 --> 00:14:43,592
and natural language processing,

257
00:14:43,592 --> 00:14:46,916
to tell us something
about human cognition.

258
00:14:47,756 --> 00:14:49,693
And the final big idea you'll have heard

259
00:14:49,693 --> 00:14:53,080
is that maybe it's a good idea we're going
to know so much about brains

260
00:14:53,080 --> 00:14:54,997
and have so much access to big data,

261
00:14:54,997 --> 00:14:57,504
because left to our own devices,

262
00:14:57,504 --> 00:15:01,335
humans are fallible, we take shortcuts,

263
00:15:01,335 --> 00:15:04,772
we err, we make mistakes,

264
00:15:04,772 --> 00:15:08,456
we're biased, and in innumerable ways,

265
00:15:08,456 --> 00:15:11,425
we get the world wrong.

266
00:15:12,843 --> 00:15:15,792
I think these are all important stories,

267
00:15:15,792 --> 00:15:19,577
and they have a lot to tell us
about what it means to be human,

268
00:15:19,577 --> 00:15:23,106
but I want you to note that today
I told you a very different story.

269
00:15:23,966 --> 00:15:27,773
It's a story about minds and not brains,

270
00:15:27,773 --> 00:15:30,779
and in particular, it's a story
about the kinds of computations

271
00:15:30,779 --> 00:15:33,369
that uniquely human minds can perform,

272
00:15:33,369 --> 00:15:37,313
which involve rich, structured knowledge
and the ability to learn

273
00:15:37,313 --> 00:15:42,581
from small amounts of data,
the evidence of just a few examples.

274
00:15:44,301 --> 00:15:48,600
And fundamentally, it's a story
about how starting as very small children

275
00:15:48,600 --> 00:15:52,780
and continuing out all the way
to the greatest accomplishments

276
00:15:52,780 --> 00:15:56,623
of our culture,

277
00:15:56,623 --> 00:15:58,620
we get the world right.

278
00:16:00,433 --> 00:16:05,700
Folks, human minds do not only learn
from small amounts of data.

279
00:16:06,285 --> 00:16:08,386
Human minds think
of altogether new ideas.

280
00:16:08,746 --> 00:16:11,787
Human minds generate
research and discovery,

281
00:16:11,787 --> 00:16:17,060
and human minds generate
art and literature and poetry and theater,

282
00:16:17,070 --> 00:16:20,830
and human minds take care of other humans:

283
00:16:20,830 --> 00:16:24,257
our old, our young, our sick.

284
00:16:24,517 --> 00:16:26,884
We even heal them.

285
00:16:27,564 --> 00:16:30,667
In the years to come, we're going
to see technological innovations

286
00:16:30,667 --> 00:16:34,464
beyond anything I can even envision,

287
00:16:34,464 --> 00:16:36,614
but we are very unlikely

288
00:16:36,614 --> 00:16:42,323
to see anything even approximating
the computational power of a human child

289
00:16:42,323 --> 00:16:46,621
in my lifetime or in yours.

290
00:16:46,621 --> 00:16:51,668
If we invest in these most powerful
learners and their development,

291
00:16:51,668 --> 00:16:54,585
in babies and children

292
00:16:54,585 --> 00:16:56,411
and mothers and fathers

293
00:16:56,411 --> 00:16:59,110
and caregivers and teachers

294
00:16:59,110 --> 00:17:03,280
the ways we invest in our other
most powerful and elegant forms

295
00:17:03,280 --> 00:17:06,498
of technology, engineering and design,

296
00:17:06,498 --> 00:17:09,437
we will not just be dreaming
of a better future,

297
00:17:09,437 --> 00:17:11,564
we will be planning for one.

298
00:17:11,564 --> 00:17:13,909
Thank you very much.

299
00:17:13,909 --> 00:17:17,330
(Applause)

300
00:17:17,810 --> 00:17:22,236
Chris Anderson: Laura, thank you.
I do actually have a question for you.

301
00:17:22,236 --> 00:17:24,595
First of all, the research is insane.

302
00:17:24,595 --> 00:17:28,320
I mean, who would design
an experiment like that? (Laughter)

303
00:17:29,150 --> 00:17:30,940
I've seen that a couple of times,

304
00:17:30,940 --> 00:17:34,162
and I still don't honestly believe
that that can truly be happening,

305
00:17:34,162 --> 00:17:37,320
but other people have done
similar experiments; it checks out.

306
00:17:37,320 --> 00:17:38,953
The babies really are that genius.

307
00:17:38,953 --> 00:17:41,960
LS: You know, they look really impressive
in our experiments,

308
00:17:41,960 --> 00:17:44,612
but think about what they
look like in real life, right?

309
00:17:44,612 --> 00:17:45,762
It starts out as a baby.

310
00:17:45,762 --> 00:17:47,769
Eighteen months later,
it's talking to you,

311
00:17:47,769 --> 00:17:50,810
and babies' first words aren't just
things like balls and ducks,

312
00:17:50,810 --> 00:17:53,691
they're things like "all gone,"
which refer to disappearance,

313
00:17:53,691 --> 00:17:55,974
or "uh-oh," which refer
to unintentional actions.

314
00:17:55,974 --> 00:17:57,536
It has to be that powerful.

315
00:17:57,536 --> 00:18:00,311
It has to be much more powerful
than anything I showed you.

316
00:18:00,311 --> 00:18:02,285
They're figuring out the entire world.

317
00:18:02,285 --> 00:18:05,429
A four-year-old can talk to you
about almost anything.

318
00:18:05,429 --> 00:18:07,030
(Applause)

319
00:18:07,030 --> 00:18:10,444
CA: And if I understand you right,
the other key point you're making is,

320
00:18:10,444 --> 00:18:13,198
we've been through these years
where there's all this talk

321
00:18:13,198 --> 00:18:15,130
of how quirky and buggy our minds are,

322
00:18:15,130 --> 00:18:17,997
that behavioral economics
and the whole theories behind that

323
00:18:17,997 --> 00:18:19,600
that we're not rational agents.

324
00:18:19,600 --> 00:18:23,816
You're really saying that the bigger
story is how extraordinary,

325
00:18:23,816 --> 00:18:28,760
and there really is genius there
that is underappreciated.

326
00:18:28,760 --> 00:18:30,830
LS: One of my favorite
quotes in psychology

327
00:18:30,830 --> 00:18:33,120
comes from the social
psychologist Solomon Asch,

328
00:18:33,120 --> 00:18:35,927
and he said the fundamental task
of psychology is to remove

329
00:18:35,927 --> 00:18:38,553
the veil of self-evidence from things.

330
00:18:38,553 --> 00:18:43,104
There are orders of magnitude
more decisions you make every day

331
00:18:43,104 --> 00:18:44,451
that get the world right.

332
00:18:44,451 --> 00:18:46,583
You know about objects
and their properties.

333
00:18:46,583 --> 00:18:49,612
You know them when they're occluded.
You know them in the dark.

334
00:18:49,612 --> 00:18:50,920
You can walk through rooms.

335
00:18:50,920 --> 00:18:54,452
You can figure out what other people
are thinking. You can talk to them.

336
00:18:54,452 --> 00:18:56,682
You can navigate space.
You know about numbers.

337
00:18:56,682 --> 00:18:59,704
You know causal relationships.
You know about moral reasoning.

338
00:18:59,704 --> 00:19:02,060
You do this effortlessly,
so we don't see it,

339
00:19:02,060 --> 00:19:04,972
but that is how we get the world right,
and it's a remarkable

340
00:19:04,972 --> 00:19:07,290
and very difficult-to-understand
accomplishment.

341
00:19:07,290 --> 00:19:09,918
CA: I suspect there are people
in the audience who have

342
00:19:09,918 --> 00:19:12,156
this view of accelerating
technological power

343
00:19:12,156 --> 00:19:15,114
who might dispute your statement
that never in our lifetimes

344
00:19:15,114 --> 00:19:17,732
will a computer do what
a three-year-old child can do,

345
00:19:17,732 --> 00:19:20,980
but what's clear is that in any scenario,

346
00:19:20,980 --> 00:19:24,750
our machines have so much to learn
from our toddlers.

347
00:19:26,230 --> 00:19:29,446
LS: I think so. You'll have some
machine learning folks up here.

348
00:19:29,446 --> 00:19:33,649
I mean, you should never bet
against babies or chimpanzees

349
00:19:33,649 --> 00:19:37,294
or technology as a matter of practice,

350
00:19:37,294 --> 00:19:41,822
but it's not just
a difference in quantity,

351
00:19:41,822 --> 00:19:43,586
it's a difference in kind.

352
00:19:43,586 --> 00:19:45,746
We have incredibly powerful computers,

353
00:19:45,746 --> 00:19:48,137
and they do do amazingly
sophisticated things,

354
00:19:48,137 --> 00:19:51,341
often with very big amounts of data.

355
00:19:51,341 --> 00:19:53,948
Human minds do, I think,
something quite different,

356
00:19:53,948 --> 00:19:57,843
and I think it's the structured,
hierarchical nature of human knowledge

357
00:19:57,843 --> 00:19:59,875
that remains a real challenge.

358
00:19:59,875 --> 00:20:02,936
CA: Laura Schulz, wonderful
food for thought. Thank you so much.

359
00:20:02,936 --> 00:20:05,858
LS: Thank you.
(Applause)