1
00:00:00,000 --> 00:00:03,992
[DIGITAL EFFECTS]

2
00:00:10,978 --> 00:00:13,250
NANCY KANWISHER: So
let's start with one

3
00:00:13,250 --> 00:00:16,880
of the deepest questions humans
have ever asked themselves.

4
00:00:16,880 --> 00:00:19,700
We're not messing around in
this class; we're going for it.

5
00:00:19,700 --> 00:00:21,530
And one of the
deepest questions is,

6
00:00:21,530 --> 00:00:23,870
where does knowledge come from?

7
00:00:23,870 --> 00:00:27,230
And as you'll know, if
you've taken even a teeny bit

8
00:00:27,230 --> 00:00:29,600
of philosophy or read
even a teeny bit,

9
00:00:29,600 --> 00:00:32,119
you know that some of the
classic views in Western

10
00:00:32,119 --> 00:00:35,630
philosophy-- especially the
empiricists, Locke and Hume--

11
00:00:35,630 --> 00:00:41,420
argue that all knowledge
comes from experience, right?

12
00:00:41,420 --> 00:00:43,430
On the other hand,
there are a number

13
00:00:43,430 --> 00:00:46,820
of other schools of thought in
Western philosophy, of which

14
00:00:46,820 --> 00:00:49,070
a dominant figure is
Immanuel Kant, who

15
00:00:49,070 --> 00:00:51,990
argued that experience
alone is not enough.

16
00:00:51,990 --> 00:00:55,670
You can't just have experience
and figure out all the stuff

17
00:00:55,670 --> 00:00:57,140
we have figured out.

18
00:00:57,140 --> 00:00:59,960
And so he argued that there
has to be what he called

19
00:00:59,960 --> 00:01:03,260
"a priori conditions"
of cognition, which

20
00:01:03,260 --> 00:01:05,420
can't be derived from
experience themselves,

21
00:01:05,420 --> 00:01:08,930
but have to be given
prior to it, OK?

22
00:01:08,930 --> 00:01:11,330
So you have to have to build
some structure into a mind

23
00:01:11,330 --> 00:01:13,100
or brain to get
it off the ground.

24
00:01:13,100 --> 00:01:15,020
You can't just start
with absolutely nothing

25
00:01:15,020 --> 00:01:16,990
and get anywhere.

26
00:01:16,990 --> 00:01:20,890
OK, and he also argued that
one of the key elements of this

27
00:01:20,890 --> 00:01:23,830
a priori structure that
you have to build in

28
00:01:23,830 --> 00:01:27,240
was space and time--

29
00:01:27,240 --> 00:01:32,090
organizing principles of
cognition and thinking.

30
00:01:32,090 --> 00:01:33,640
And so in his
version of it, space

31
00:01:33,640 --> 00:01:37,240
is nothing but the form of all
appearances of outer sense,

32
00:01:37,240 --> 00:01:41,380
and it can be given prior to
all actual perceptions and so

33
00:01:41,380 --> 00:01:44,800
exist in the mind a
priori, and can contain,

34
00:01:44,800 --> 00:01:47,860
prior to all experience,
principles which determine

35
00:01:47,860 --> 00:01:50,840
the relations of these objects.

36
00:01:50,840 --> 00:01:57,230
OK, well, is that just
empty philosophical hot air?

37
00:01:57,230 --> 00:01:59,480
It's kind of hard to understand
exactly what he means.

38
00:01:59,480 --> 00:02:02,420
You're actually have to go
spend a good deal of time

39
00:02:02,420 --> 00:02:05,390
with reading him to
make any sense of it--

40
00:02:05,390 --> 00:02:08,690
or cheat and get your
friends to tell you, as I do.

41
00:02:08,690 --> 00:02:12,800
But no, I'll argue it's not just
empty philosophical hot air--

42
00:02:12,800 --> 00:02:14,840
that these are, in
some important sense,

43
00:02:14,840 --> 00:02:16,375
empirical questions.

44
00:02:16,375 --> 00:02:17,750
And there are
empirical questions

45
00:02:17,750 --> 00:02:21,770
that our field
addresses very directly.

46
00:02:21,770 --> 00:02:23,450
And so on Wednesday,
we'll talk about

47
00:02:23,450 --> 00:02:27,500
whether your representations of
space in your head are innate

48
00:02:27,500 --> 00:02:28,100
or not.

49
00:02:28,100 --> 00:02:31,340
It's pretty much directly
what Kant is talking about--

50
00:02:31,340 --> 00:02:33,590
or the modern version of
what he was talking about.

51
00:02:33,590 --> 00:02:35,840
And today, we'll talk about
which aspects of the brain

52
00:02:35,840 --> 00:02:38,360
are innate and which
our learned, OK?

53
00:02:38,360 --> 00:02:39,950
That's the agenda.

54
00:02:39,950 --> 00:02:44,728
OK, so this little kind of
Easter egg brain here very

55
00:02:44,728 --> 00:02:46,520
schematically shows
you some of the regions

56
00:02:46,520 --> 00:02:48,312
that we've been talking
about in this class

57
00:02:48,312 --> 00:02:51,020
so far, with regions that
are, to varying degrees,

58
00:02:51,020 --> 00:02:54,290
specialized for processing
things like shape, and color,

59
00:02:54,290 --> 00:02:57,590
and motion, and faces,
and places, and bodies--

60
00:02:57,590 --> 00:03:00,440
visually processing all of
these things in approximately

61
00:03:00,440 --> 00:03:02,420
those locations.

62
00:03:02,420 --> 00:03:05,030
And as I've mentioned,
these regions

63
00:03:05,030 --> 00:03:07,610
are present in approximately
the same location--

64
00:03:07,610 --> 00:03:09,830
with some individual
variability-- in pretty

65
00:03:09,830 --> 00:03:11,958
much every normal person.

66
00:03:11,958 --> 00:03:14,000
One of my lab members
says, you keep saying that,

67
00:03:14,000 --> 00:03:14,860
and it's just not true.

68
00:03:14,860 --> 00:03:16,140
There's some percent
of subjects who

69
00:03:16,140 --> 00:03:17,290
just don't show these things.

70
00:03:17,290 --> 00:03:18,248
He's kind of right, OK.

71
00:03:18,248 --> 00:03:21,307
So maybe, I don't know,
5%, 10% of subjects,

72
00:03:21,307 --> 00:03:22,890
you wouldn't see
some of these things.

73
00:03:22,890 --> 00:03:24,807
And we've never actually
done the serious work

74
00:03:24,807 --> 00:03:27,600
of bringing those subjects back,
scanning the hell out of them,

75
00:03:27,600 --> 00:03:30,100
and finding out whether they
were just asleep in the scanner

76
00:03:30,100 --> 00:03:32,910
or it was a bad scanner
day, or whatever it was.

77
00:03:32,910 --> 00:03:35,370
I bet they all have them,
and it's just sometimes

78
00:03:35,370 --> 00:03:38,430
you don't see it, but I'm trying
to be a little more honest.

79
00:03:38,430 --> 00:03:40,920
OK, but you just look at this.

80
00:03:40,920 --> 00:03:42,890
Given this very
schematic version of it,

81
00:03:42,890 --> 00:03:45,450
you say, how would
you build this system?

82
00:03:45,450 --> 00:03:49,920
How would you start with an
embryo and build into a genome,

83
00:03:49,920 --> 00:03:51,810
or build into
whatever experience

84
00:03:51,810 --> 00:03:54,780
is going to happen to
this developing organism?

85
00:03:54,780 --> 00:03:57,930
How would it end up with this
very particular structure,

86
00:03:57,930 --> 00:04:00,570
with those things in
approximately the same place--

87
00:04:00,570 --> 00:04:03,960
or at least the same relative
positions-- in all subjects?

88
00:04:03,960 --> 00:04:06,840
The face bits are always
lateral to the color bits.

89
00:04:06,840 --> 00:04:09,630
The place bits are
medial to the color bits.

90
00:04:09,630 --> 00:04:12,780
The shape bits are out
on the lateral surface.

91
00:04:12,780 --> 00:04:14,520
It's like always like that.

92
00:04:14,520 --> 00:04:17,459
How do you build a
system like that?

93
00:04:17,459 --> 00:04:19,769
I find it hard not to
immediately think, well,

94
00:04:19,769 --> 00:04:21,899
some aspect of this
must be innate,

95
00:04:21,899 --> 00:04:25,420
or how would it be so damn
similar in each individual,

96
00:04:25,420 --> 00:04:25,920
right?

97
00:04:29,780 --> 00:04:32,930
But it's not the
only hypothesis.

98
00:04:32,930 --> 00:04:37,040
Some big part of it-- even if
some aspect of this is innate,

99
00:04:37,040 --> 00:04:39,440
some big part of it may
also be learned or derived

100
00:04:39,440 --> 00:04:42,030
from experience, OK?

101
00:04:42,030 --> 00:04:43,498
So what do you guys think?

102
00:04:43,498 --> 00:04:45,290
Do you think the fact
that these structures

103
00:04:45,290 --> 00:04:48,080
are in systematically the
same place across subjects

104
00:04:48,080 --> 00:04:50,960
mean you have to build
in all that stuff,

105
00:04:50,960 --> 00:04:53,510
somehow figure out how
to get a bunch of As

106
00:04:53,510 --> 00:04:57,650
and Ts and Gs and Cs in your DNA
to give you a blueprint for how

107
00:04:57,650 --> 00:04:59,620
to build that structure?

108
00:05:02,440 --> 00:05:03,420
What do you think?

109
00:05:08,170 --> 00:05:09,548
Yeah?

110
00:05:09,548 --> 00:05:11,090
AUDIENCE: I mean,
it's a combination,

111
00:05:11,090 --> 00:05:14,210
but it's hard to, then,
think about how that's

112
00:05:14,210 --> 00:05:15,980
involved in
[INAUDIBLE] generation

113
00:05:15,980 --> 00:05:18,710
and then kind of
become more innate?

114
00:05:18,710 --> 00:05:22,000
NANCY KANWISHER: Yeah, so
to some extent, experience--

115
00:05:22,000 --> 00:05:23,800
what I mean here is
learn from experience

116
00:05:23,800 --> 00:05:25,360
within each individual.

117
00:05:25,360 --> 00:05:26,890
You could argue
that "innate" really

118
00:05:26,890 --> 00:05:30,340
means "learned through the
experience of our ancestors,

119
00:05:30,340 --> 00:05:33,850
and hence wired
into the DNA," yeah.

120
00:05:33,850 --> 00:05:35,830
Anyway, I find this not
an obvious question,

121
00:05:35,830 --> 00:05:39,527
and so we'll talk about
what the data say here.

122
00:05:39,527 --> 00:05:41,860
So first of all, we're going
to do some very basic facts

123
00:05:41,860 --> 00:05:44,193
about brain development, just
to get the picture of what

124
00:05:44,193 --> 00:05:47,290
we're talking about physically
with the development of brains.

125
00:05:47,290 --> 00:05:50,980
So we can ask, what
is present at birth?

126
00:05:50,980 --> 00:05:55,030
And so it turns out that most of
the neurons in the adult brain

127
00:05:55,030 --> 00:05:57,880
are generated before birth, OK?

128
00:05:57,880 --> 00:06:00,760
So most of the actual
neurons are generated early.

129
00:06:00,760 --> 00:06:03,400
You're not making a whole
lot more after birth--

130
00:06:03,400 --> 00:06:05,650
a few, but not a lot.

131
00:06:05,650 --> 00:06:08,500
Further, the current view is
that most of the long-range

132
00:06:08,500 --> 00:06:11,000
connections-- that means like
a connection between this part

133
00:06:11,000 --> 00:06:12,250
and that part of the brain--

134
00:06:12,250 --> 00:06:14,770
are also present at birth, OK?

135
00:06:17,630 --> 00:06:20,240
Nonetheless, even though a lot
of stuff is present at birth,

136
00:06:20,240 --> 00:06:23,210
a lot of stuff changes in the
first couple of years of life.

137
00:06:23,210 --> 00:06:25,880
Most obviously, the
brain doubles in volume

138
00:06:25,880 --> 00:06:29,165
in the first year,
from a two-week-old,

139
00:06:29,165 --> 00:06:31,450
to a one-year-old,
to a two-year-old.

140
00:06:34,940 --> 00:06:38,600
The cortical thickness-- you
can see here the dark stuff,

141
00:06:38,600 --> 00:06:41,300
which is the gray
matter out there--

142
00:06:41,300 --> 00:06:46,910
increases sharply between
years one and two.

143
00:06:46,910 --> 00:06:49,610
But also, the complexity
of each individual neuron

144
00:06:49,610 --> 00:06:52,790
increases dramatically in
the first few years of life.

145
00:06:52,790 --> 00:06:56,060
So here's a schematic picture
of a piece of gray matter here.

146
00:06:56,060 --> 00:06:57,650
We have some number
of neurons here

147
00:06:57,650 --> 00:07:01,640
with a few little processes
and a few connections.

148
00:07:01,640 --> 00:07:04,580
And over the first
couple of years of life,

149
00:07:04,580 --> 00:07:07,610
those connections
get much more dense,

150
00:07:07,610 --> 00:07:09,230
and the neurons get
much more complex.

151
00:07:11,940 --> 00:07:13,400
OK, and the final
thing that really

152
00:07:13,400 --> 00:07:18,440
matters early on in development
is that myelination happens

153
00:07:18,440 --> 00:07:19,980
rapidly in the first few years.

154
00:07:19,980 --> 00:07:22,400
And remember, myelin--
this is a little reminder--

155
00:07:22,400 --> 00:07:25,160
neuron with that yellow stuff,
which is a bunch of cells

156
00:07:25,160 --> 00:07:28,970
that wrap around the axons,
the long processes of a neuron.

157
00:07:28,970 --> 00:07:31,938
And that myelin
sheath builds up a lot

158
00:07:31,938 --> 00:07:33,230
over the first couple of years.

159
00:07:33,230 --> 00:07:35,630
And that's important, because
the myelin sheath enables

160
00:07:35,630 --> 00:07:37,940
those neurons to send
their signals faster

161
00:07:37,940 --> 00:07:41,070
down their axons, OK?

162
00:07:41,070 --> 00:07:44,610
OK, and this is just a
picture of different--

163
00:07:44,610 --> 00:07:47,250
of a vertical slice like
this through the anatomy

164
00:07:47,250 --> 00:07:52,530
of infants of different ages,
from 107 days up to about a

165
00:07:52,530 --> 00:07:53,190
year.

166
00:07:53,190 --> 00:07:54,990
And the colored
stuff in the middle

167
00:07:54,990 --> 00:07:57,240
is degree of
myelination, which you

168
00:07:57,240 --> 00:07:59,760
can see with various
kinds of anatomical scans.

169
00:07:59,760 --> 00:08:03,540
You can see it starts at 107
days with a tiny little bit

170
00:08:03,540 --> 00:08:06,690
in the middle, and it gets
more and more myelinated

171
00:08:06,690 --> 00:08:09,300
and moves from
center to periphery

172
00:08:09,300 --> 00:08:11,460
over the first year of life.

173
00:08:11,460 --> 00:08:14,370
So all those fiber
pathways are getting

174
00:08:14,370 --> 00:08:19,920
accelerated as they get wrapped
with myelin and hence sped up.

175
00:08:19,920 --> 00:08:25,020
OK, all right, so bottom
line is most neurons

176
00:08:25,020 --> 00:08:27,210
and long-range connections
are in place at birth,

177
00:08:27,210 --> 00:08:30,640
but development continues
rapidly in the first two years,

178
00:08:30,640 --> 00:08:33,960
especially increasing complexity
of neurons and synapses

179
00:08:33,960 --> 00:08:36,870
and myelination of long-range
connections and white matter,

180
00:08:36,870 --> 00:08:37,409
OK?

181
00:08:37,409 --> 00:08:40,919
So it's just basic anatomy,
nothing functional yet.

182
00:08:40,919 --> 00:08:43,350
OK, now we're going to
consider in some detail

183
00:08:43,350 --> 00:08:46,560
the case of face perception,
not really because that's

184
00:08:46,560 --> 00:08:48,300
what I work on--

185
00:08:48,300 --> 00:08:50,130
or used to work on, mostly--

186
00:08:50,130 --> 00:08:52,140
but just because
there's a very rich set

187
00:08:52,140 --> 00:08:54,480
of data where people have
grappled with this question

188
00:08:54,480 --> 00:08:56,190
in the case of face perception.

189
00:08:56,190 --> 00:08:58,590
Next time, we'll talk about
the navigation network

190
00:08:58,590 --> 00:09:01,950
and reorientation-- what parts
of that system might be innate

191
00:09:01,950 --> 00:09:04,058
and learned.

192
00:09:04,058 --> 00:09:05,850
So I'll just say right
out of the beginning

193
00:09:05,850 --> 00:09:08,880
that this is an extremely
active area, where every time I

194
00:09:08,880 --> 00:09:11,970
turn around, another paper
comes out that contradicts

195
00:09:11,970 --> 00:09:13,740
a previously-published finding.

196
00:09:13,740 --> 00:09:16,320
And so that makes
it fun, but it means

197
00:09:16,320 --> 00:09:18,653
there isn't going to be some
really tight, perfect story

198
00:09:18,653 --> 00:09:19,153
here.

199
00:09:19,153 --> 00:09:21,720
And I'd rather take you guys
straight to the cutting edge,

200
00:09:21,720 --> 00:09:26,310
even though it's kind of a mess,
than give you a nicely packaged

201
00:09:26,310 --> 00:09:28,740
but surely wrong picture, OK?

202
00:09:28,740 --> 00:09:31,228
Because again, I think what
matters most in this area

203
00:09:31,228 --> 00:09:33,270
is how do you go about
answering these questions,

204
00:09:33,270 --> 00:09:36,900
rather than what is the
current state of the thoughts

205
00:09:36,900 --> 00:09:37,680
about the answers.

206
00:09:37,680 --> 00:09:39,472
OK, so how are we going
to think about, how

207
00:09:39,472 --> 00:09:41,472
does face perception develop?

208
00:09:41,472 --> 00:09:42,930
Well just to get
started, I'm going

209
00:09:42,930 --> 00:09:50,658
to show you a very brief
movie of a 72-hour-old monkey,

210
00:09:50,658 --> 00:09:51,910
and see what you think.

211
00:10:09,360 --> 00:10:09,860
He's sleepy.

212
00:10:20,180 --> 00:10:22,070
He's pretty interested
in that face.

213
00:10:22,070 --> 00:10:22,760
And watch now.

214
00:10:29,547 --> 00:10:30,047
Hmm.

215
00:10:41,478 --> 00:10:45,450
[LAUGHS] Pretty cute, huh?

216
00:10:45,450 --> 00:10:46,325
So what do you think?

217
00:10:49,250 --> 00:10:51,590
What does this tell us
about face perception?

218
00:10:51,590 --> 00:10:52,090
Yeah?

219
00:10:52,090 --> 00:10:53,882
AUDIENCE: Did they try
just moving anything

220
00:10:53,882 --> 00:10:54,610
in front of him?

221
00:10:54,610 --> 00:10:56,020
NANCY KANWISHER: Good question.

222
00:10:56,020 --> 00:10:57,080
Good for you.

223
00:10:57,080 --> 00:10:58,090
Quily, is that right?

224
00:10:58,090 --> 00:10:58,882
AUDIENCE: "Quile-y"

225
00:10:58,882 --> 00:11:02,320
NANCY KANWISHER:
"Quile-y", all right.

226
00:11:02,320 --> 00:11:05,703
Yes, so Quiley asked, did
they try moving just anything

227
00:11:05,703 --> 00:11:06,370
in front of him?

228
00:11:06,370 --> 00:11:07,930
Absolutely the right question.

229
00:11:07,930 --> 00:11:10,150
So that monkey seems pretty
interested in that face,

230
00:11:10,150 --> 00:11:11,920
but a face is a moving thing.

231
00:11:11,920 --> 00:11:15,340
Motion is very salient to young
primates-- humans, and monkeys,

232
00:11:15,340 --> 00:11:17,800
and many others, absolutely.

233
00:11:17,800 --> 00:11:19,180
What else did you see in here?

234
00:11:21,820 --> 00:11:22,543
Yeah.

235
00:11:22,543 --> 00:11:25,740
AUDIENCE: It started
imitating [INAUDIBLE]..

236
00:11:25,740 --> 00:11:27,120
NANCY KANWISHER: Yeah, kind of.

237
00:11:27,120 --> 00:11:29,400
I mean, the person--

238
00:11:29,400 --> 00:11:32,340
the adult human there--
was moving their mouth open

239
00:11:32,340 --> 00:11:35,382
like this, and the monkey
was doing something

240
00:11:35,382 --> 00:11:36,090
with their mouth.

241
00:11:36,090 --> 00:11:37,995
So what would that require?

242
00:11:41,030 --> 00:11:41,530
Sorry?

243
00:11:41,530 --> 00:11:43,340
AUDIENCE: I like,
I have another.

244
00:11:43,340 --> 00:11:48,958
Also, was the monkey allowed
to touch its face before this?

245
00:11:48,958 --> 00:11:50,500
NANCY KANWISHER:
Yeah, good question.

246
00:11:50,500 --> 00:11:51,340
Good question.

247
00:11:51,340 --> 00:11:55,450
72 hours is damned early,
but it's not zero experience,

248
00:11:55,450 --> 00:11:56,320
right?

249
00:11:56,320 --> 00:11:59,350
So who knows what they've
managed to pick up that early.

250
00:11:59,350 --> 00:12:01,250
There are actually
studies in humans,

251
00:12:01,250 --> 00:12:03,340
which I'm hoping Heather
knows better than me.

252
00:12:03,340 --> 00:12:04,720
Those Andy Melzoff things.

253
00:12:04,720 --> 00:12:05,890
How young are those humans?

254
00:12:05,890 --> 00:12:08,532
Those are like first hour.

255
00:12:08,532 --> 00:12:10,135
AUDIENCE: Yeah, [INAUDIBLE].

256
00:12:10,135 --> 00:12:11,510
NANCY KANWISHER:
I think it's a--

257
00:12:11,510 --> 00:12:12,718
AUDIENCE: [INAUDIBLE]

258
00:12:12,718 --> 00:12:14,260
NANCY KANWISHER: So
there are studies

259
00:12:14,260 --> 00:12:16,300
in humans where you can
show versions of that,

260
00:12:16,300 --> 00:12:19,132
with newborn infants copying--

261
00:12:19,132 --> 00:12:21,340
the experimenter comes up
and sticks their tongue out

262
00:12:21,340 --> 00:12:25,210
at the infant, and the infant
does that back, kinda sorta.

263
00:12:25,210 --> 00:12:29,210
Certainly within the first two
days, maybe even earlier, OK?

264
00:12:29,210 --> 00:12:31,090
OK, so it's very suggestive.

265
00:12:31,090 --> 00:12:33,700
It's tantalizing, but we
need controlled conditions.

266
00:12:33,700 --> 00:12:37,210
It doesn't tell us
everything we need to know.

267
00:12:37,210 --> 00:12:39,055
OK, so if we think
about it, there

268
00:12:39,055 --> 00:12:44,020
are ends of the hypothesis space
about how all of this could go.

269
00:12:44,020 --> 00:12:47,228
As Alana mentioned, everything
is both genes and experience.

270
00:12:47,228 --> 00:12:49,270
That's true, but there
are very, very importantly

271
00:12:49,270 --> 00:12:51,790
different ways in which
genes and experience can

272
00:12:51,790 --> 00:12:57,730
act together-- some in which
a big part of the heft of what

273
00:12:57,730 --> 00:12:59,425
the adult form has
might be built in,

274
00:12:59,425 --> 00:13:01,300
and other stories where
most of the structure

275
00:13:01,300 --> 00:13:02,360
comes from experience.

276
00:13:02,360 --> 00:13:04,318
So just because everything
is both doesn't mean

277
00:13:04,318 --> 00:13:07,120
we shouldn't flesh out
exactly what comes from what.

278
00:13:07,120 --> 00:13:11,020
So on one end of
the spectrum, you

279
00:13:11,020 --> 00:13:14,440
might imagine that there's
some very, very rudimentary

280
00:13:14,440 --> 00:13:19,330
precursor that has to be built
in, plus a learning mechanism,

281
00:13:19,330 --> 00:13:20,350
OK?

282
00:13:20,350 --> 00:13:23,080
Or a bunch of
rudimentary precursors,

283
00:13:23,080 --> 00:13:24,820
which are just there
to get the system

284
00:13:24,820 --> 00:13:27,280
to learn in the right way, OK?

285
00:13:27,280 --> 00:13:29,590
And so we'll talk
shortly about the idea

286
00:13:29,590 --> 00:13:33,340
that there might be some kind
of innate template for faces

287
00:13:33,340 --> 00:13:36,490
that gets monkeys and
humans to look at faces.

288
00:13:36,490 --> 00:13:38,920
And then, the idea is once you
get them to look at a face,

289
00:13:38,920 --> 00:13:42,250
then experience can take over
from there and do the rest.

290
00:13:42,250 --> 00:13:46,600
But you've got to get them
to collect the right input.

291
00:13:46,600 --> 00:13:49,450
And there's lots of interesting
computational work going on now

292
00:13:49,450 --> 00:13:53,680
where people are using various
computational models to say,

293
00:13:53,680 --> 00:13:55,930
what do we have to
build into, say,

294
00:13:55,930 --> 00:13:58,570
a convolutional neural
network or some other kind

295
00:13:58,570 --> 00:14:02,650
of computational model to get
it to do some complicated thing?

296
00:14:02,650 --> 00:14:06,160
I just came from a job talk
the last hour-- really amazing

297
00:14:06,160 --> 00:14:08,440
talk-- where the guy is
showing that if you build

298
00:14:08,440 --> 00:14:12,460
in, basically, curiosity
early on in a network,

299
00:14:12,460 --> 00:14:15,610
you get much more general
learners than if you

300
00:14:15,610 --> 00:14:20,890
build in a bunch of goals for
a developing network to seek.

301
00:14:20,890 --> 00:14:23,530
anyway it's a very active
area, and the paper

302
00:14:23,530 --> 00:14:25,570
that I just decided
to assign to you guys,

303
00:14:25,570 --> 00:14:27,760
just kind of skim
it and get the gist.

304
00:14:27,760 --> 00:14:30,290
The basic idea-- this
is from Shimon Ullman,

305
00:14:30,290 --> 00:14:32,140
who is a very deep
thinker in this field.

306
00:14:32,140 --> 00:14:36,130
And he argues that hands are
very important in infants.

307
00:14:36,130 --> 00:14:40,065
Faces are important, but so are
hands, because hands do stuff.

308
00:14:40,065 --> 00:14:41,440
And we're social
primates, and we

309
00:14:41,440 --> 00:14:44,050
want to learn from other social
primates like our parents.

310
00:14:44,050 --> 00:14:46,145
And watching their hands
is extremely informative.

311
00:14:46,145 --> 00:14:47,770
Whatever they're
doing with their hands

312
00:14:47,770 --> 00:14:49,840
is probably stuff we
need to learn about.

313
00:14:49,840 --> 00:14:52,840
And further, we need to know
where they're looking, right?

314
00:14:52,840 --> 00:14:54,295
So gaze perception.

315
00:14:54,295 --> 00:14:55,605
I think I did this demo before.

316
00:14:55,605 --> 00:14:57,730
If I'm talking to you guys,
and I start doing that,

317
00:14:57,730 --> 00:15:00,147
it's really hard, even though
you know I'm just faking you

318
00:15:00,147 --> 00:15:04,150
out, not to have your
attention pulled over there,

319
00:15:04,150 --> 00:15:06,140
and infants need to
learn that as well.

320
00:15:06,140 --> 00:15:09,070
So Shimon Ullman's
basic idea is that you

321
00:15:09,070 --> 00:15:11,920
can start with an extremely
rudimentary system,

322
00:15:11,920 --> 00:15:14,470
and all you have to build
in is this idea that he

323
00:15:14,470 --> 00:15:15,850
calls "mover," right?

324
00:15:15,850 --> 00:15:18,970
So the idea is that if you look
in a whole set of, say, YouTube

325
00:15:18,970 --> 00:15:22,030
videos, and you just look
for patches of the image that

326
00:15:22,030 --> 00:15:23,800
are moving, that's no good.

327
00:15:23,800 --> 00:15:24,800
It won't be a hand.

328
00:15:24,800 --> 00:15:27,770
It might be a whole animal,
or a face, or something else.

329
00:15:27,770 --> 00:15:30,220
But if you look
in YouTube videos,

330
00:15:30,220 --> 00:15:33,370
a proxy for natural experience--
it's OK; it's not perfect,

331
00:15:33,370 --> 00:15:36,730
but it's something- you look
for a patch of the image that

332
00:15:36,730 --> 00:15:41,230
moves over and then causes
another previously-stationary

333
00:15:41,230 --> 00:15:43,270
image patch to move.

334
00:15:43,270 --> 00:15:46,660
That's what happens when
we pick stuff up, OK?

335
00:15:46,660 --> 00:15:49,930
And so his idea is you can
build in this extremely simple

336
00:15:49,930 --> 00:15:50,470
thing--

337
00:15:50,470 --> 00:15:53,620
Mover, which is a very
simple visual algorithm,

338
00:15:53,620 --> 00:15:56,540
can find image patches and move
over and cause another image

339
00:15:56,540 --> 00:15:57,625
patch--

340
00:15:57,625 --> 00:15:59,770
or then the two image
patches move together.

341
00:15:59,770 --> 00:16:02,770
And Mover will enable you
to identify hands in images

342
00:16:02,770 --> 00:16:03,400
pretty well.

343
00:16:03,400 --> 00:16:04,900
He looks in YouTube
videos and shows

344
00:16:04,900 --> 00:16:06,970
that it's really good
at picking out hands.

345
00:16:06,970 --> 00:16:09,760
And then, further, once
you've picked out hands,

346
00:16:09,760 --> 00:16:11,710
that's a really
important teaching signal

347
00:16:11,710 --> 00:16:13,300
in teaching you to read gaze.

348
00:16:13,300 --> 00:16:15,430
Because often, people
look at their hands

349
00:16:15,430 --> 00:16:18,760
before they do things
with them, yeah?

350
00:16:18,760 --> 00:16:21,940
So the idea is there's a
very active ferment now

351
00:16:21,940 --> 00:16:23,570
in computational
modeling saying,

352
00:16:23,570 --> 00:16:26,710
how can we start with just the
most rudimentary, minimalist

353
00:16:26,710 --> 00:16:30,415
stuff that has to be built in,
and then build on experience

354
00:16:30,415 --> 00:16:31,540
to get the rest from there?

355
00:16:31,540 --> 00:16:32,757
Is that idea clear?

356
00:16:32,757 --> 00:16:34,340
It's worth reading
that paper, though.

357
00:16:34,340 --> 00:16:35,382
It's beautifully written.

358
00:16:35,382 --> 00:16:36,430
He's brilliant.

359
00:16:36,430 --> 00:16:38,740
OK, so that's one
end of the spectrum.

360
00:16:38,740 --> 00:16:42,873
Nobody thinks that you
learn absolutely everything

361
00:16:42,873 --> 00:16:43,540
from experience.

362
00:16:43,540 --> 00:16:44,830
You've got to
build in something.

363
00:16:44,830 --> 00:16:46,955
Plus, we know all those
neurons are there at birth.

364
00:16:46,955 --> 00:16:48,800
And so the idea
is some version--

365
00:16:48,800 --> 00:16:52,428
the minimalist
nativist view says

366
00:16:52,428 --> 00:16:54,220
you build in a few very
rudimentary things,

367
00:16:54,220 --> 00:16:56,200
and they're enough to
bootstrap learning.

368
00:16:56,200 --> 00:17:00,052
OK, on the other end of the
spectrum, you might think--

369
00:17:00,052 --> 00:17:01,510
and many have
proposed-- that we're

370
00:17:01,510 --> 00:17:04,390
born with a nearly
adult-like system that

371
00:17:04,390 --> 00:17:07,900
only needs fine-tuning
from experience, right?

372
00:17:07,900 --> 00:17:10,990
Nobody thinks that zero
experience is necessary.

373
00:17:10,990 --> 00:17:14,750
That would be kind of
crazy, or implausible.

374
00:17:14,750 --> 00:17:18,260
But on the other extreme, this
view is that most of the stuff

375
00:17:18,260 --> 00:17:19,250
is built-in.

376
00:17:19,250 --> 00:17:21,260
OK, everybody get the
theoretical space here

377
00:17:21,260 --> 00:17:22,910
that we're considering?

378
00:17:22,910 --> 00:17:26,339
OK, so what kind of data can
constrain these questions?

379
00:17:26,339 --> 00:17:29,180
Well, one obvious question
is, what is present at birth?

380
00:17:29,180 --> 00:17:30,440
What is the initial state--

381
00:17:30,440 --> 00:17:33,740
or as close as we can get to it?

382
00:17:33,740 --> 00:17:36,440
Then we can ask, how does
the system change over time

383
00:17:36,440 --> 00:17:38,240
from birth onward?

384
00:17:38,240 --> 00:17:41,090
And then we can ask, what are
the causal roles of experience

385
00:17:41,090 --> 00:17:44,900
and biological maturation
in that change after birth?

386
00:17:44,900 --> 00:17:46,848
So that's the whole
set of questions

387
00:17:46,848 --> 00:17:48,890
we'd need to answer to
understand how development

388
00:17:48,890 --> 00:17:49,910
works.

389
00:17:49,910 --> 00:17:52,520
And a very central-- if
not the central-- challenge

390
00:17:52,520 --> 00:17:57,080
of development is that
experience and maturation are

391
00:17:57,080 --> 00:18:01,040
deeply confounded as you look
from birth onward, right?

392
00:18:01,040 --> 00:18:03,830
So five-year-olds are
both more mature--

393
00:18:03,830 --> 00:18:06,080
they've had more time for
their biological systems

394
00:18:06,080 --> 00:18:08,750
to wire themselves up, including
their bodies, and their brains,

395
00:18:08,750 --> 00:18:09,960
and the whole bit--

396
00:18:09,960 --> 00:18:12,920
and maybe some of that
is just on a maturation

397
00:18:12,920 --> 00:18:14,090
kind of autopilot.

398
00:18:14,090 --> 00:18:16,560
But they've also had
a lot more experience.

399
00:18:16,560 --> 00:18:18,620
So one of the central
challenges of development

400
00:18:18,620 --> 00:18:21,140
is trying to figure out
how those later stages--

401
00:18:21,140 --> 00:18:24,410
like two months old, one
year old, 10 years old--

402
00:18:24,410 --> 00:18:29,180
how those changes that happen
between birth and those stages

403
00:18:29,180 --> 00:18:29,990
can--

404
00:18:29,990 --> 00:18:33,310
how can we tease apart which of
that came from just maturation

405
00:18:33,310 --> 00:18:34,910
and which came from experience?

406
00:18:34,910 --> 00:18:37,430
All right, OK.

407
00:18:37,430 --> 00:18:41,720
Importantly, things that
happen well after birth

408
00:18:41,720 --> 00:18:46,660
need not be learned, right?

409
00:18:46,660 --> 00:18:48,190
So think about puberty.

410
00:18:48,190 --> 00:18:52,210
Puberty is going to
happen around 10, 11, 12.

411
00:18:52,210 --> 00:18:54,850
And OK, you've got
to eat and have

412
00:18:54,850 --> 00:18:56,630
some basic inputs
to your system,

413
00:18:56,630 --> 00:18:58,660
but it's pretty much
going to happen.

414
00:18:58,660 --> 00:19:01,510
It's not a product of
what you were taught

415
00:19:01,510 --> 00:19:03,670
or the particular
information that landed

416
00:19:03,670 --> 00:19:06,280
on your sensory receptors.

417
00:19:06,280 --> 00:19:08,590
I'm sure there's some
obscure influences that I

418
00:19:08,590 --> 00:19:10,120
don't know about,
but mostly, it's

419
00:19:10,120 --> 00:19:11,980
on a developmental autopilot.

420
00:19:11,980 --> 00:19:13,240
It's just going to happen.

421
00:19:13,240 --> 00:19:15,310
OK, so keep in mind--
this is really important--

422
00:19:15,310 --> 00:19:17,350
that things that
happen well after birth

423
00:19:17,350 --> 00:19:19,030
aren't necessarily learned.

424
00:19:19,030 --> 00:19:22,630
It might be just maturation
that's continuing, right?

425
00:19:22,630 --> 00:19:26,740
OK, just as being 5 feet tall
versus a foot and 1/2 tall

426
00:19:26,740 --> 00:19:27,940
isn't really learned.

427
00:19:27,940 --> 00:19:33,070
It's just a maturation
program that unfolds.

428
00:19:33,070 --> 00:19:37,000
OK, so we can ask these three
questions both behaviorally

429
00:19:37,000 --> 00:19:38,110
and naturally.

430
00:19:38,110 --> 00:19:40,450
And ultimately, we want
them to tell the same story.

431
00:19:40,450 --> 00:19:43,150
When I said there's some
chaos in this field right now,

432
00:19:43,150 --> 00:19:45,850
I mean that basically, they're
not converging very well yet,

433
00:19:45,850 --> 00:19:47,770
but that's fun--

434
00:19:47,770 --> 00:19:48,310
sort of.

435
00:19:48,310 --> 00:19:49,960
[LAUGHS] Sometimes
it's aggravating,

436
00:19:49,960 --> 00:19:51,310
but mostly, it's fun.

437
00:19:51,310 --> 00:19:53,390
OK, so let's start with
some behavioral data.

438
00:19:53,390 --> 00:19:54,970
So let's consider
the initial state

439
00:19:54,970 --> 00:19:57,100
of face perception in newborns.

440
00:19:57,100 --> 00:20:01,060
OK, so we can ask, what kind
of perceptual, face perceptual

441
00:20:01,060 --> 00:20:02,650
abilities are
present in newborns?

442
00:20:02,650 --> 00:20:04,990
And we can ask whether they
can detect a face-- that

443
00:20:04,990 --> 00:20:07,690
is, discriminate a
face from a non-face,

444
00:20:07,690 --> 00:20:12,190
whether it's a body, or an
object, or something else.

445
00:20:12,190 --> 00:20:14,500
We can ask about preferred
attention to faces.

446
00:20:14,500 --> 00:20:16,510
Do they, do newborns
want to look at faces

447
00:20:16,510 --> 00:20:18,610
more than non-faces?

448
00:20:18,610 --> 00:20:21,280
We can ask about the
ability to recognize faces,

449
00:20:21,280 --> 00:20:25,150
to discriminate one
face from another, OK?

450
00:20:25,150 --> 00:20:27,460
And we can ask about the
ability to recognize faces

451
00:20:27,460 --> 00:20:28,695
across image changes.

452
00:20:28,695 --> 00:20:30,820
So we spent a lot of time
in the first few lectures

453
00:20:30,820 --> 00:20:32,890
talking about the central
problem of invariance

454
00:20:32,890 --> 00:20:33,880
in vision--

455
00:20:33,880 --> 00:20:36,010
about, how do you know
that this image that you're

456
00:20:36,010 --> 00:20:38,110
looking at here is the
same person as that image,

457
00:20:38,110 --> 00:20:40,480
even though those are
very different images?

458
00:20:40,480 --> 00:20:43,180
And actually, this image
on your retina right

459
00:20:43,180 --> 00:20:46,810
now is more different than
this image on your retina

460
00:20:46,810 --> 00:20:49,360
than if we got one of you and
came up-- had you come up here

461
00:20:49,360 --> 00:20:50,920
and had you look forward.

462
00:20:50,920 --> 00:20:54,253
So the image changes that result
from a change in orientation

463
00:20:54,253 --> 00:20:56,170
are greater than the
image changes that result

464
00:20:56,170 --> 00:20:57,490
from a change in identity.

465
00:20:57,490 --> 00:20:59,410
So it's a big
computational challenge.

466
00:20:59,410 --> 00:21:02,140
When is that solved?

467
00:21:02,140 --> 00:21:04,630
And then, there are these
so-called signatures

468
00:21:04,630 --> 00:21:06,538
of face perception
that we've talked

469
00:21:06,538 --> 00:21:08,830
about a little bit-- for
example, the inversion effect.

470
00:21:08,830 --> 00:21:11,650
Recall the inversion effect
is larger in magnitude

471
00:21:11,650 --> 00:21:12,820
for faces than non-faces.

472
00:21:12,820 --> 00:21:15,940
So we can ask when
those things develop.

473
00:21:15,940 --> 00:21:17,740
OK, so let's start
with face detection

474
00:21:17,740 --> 00:21:20,620
and preferred
attention to faces.

475
00:21:20,620 --> 00:21:22,990
Well, so classic
studies from the early

476
00:21:22,990 --> 00:21:26,110
'90s, and actually, some of
them going back to the '70s,

477
00:21:26,110 --> 00:21:27,910
did the following
very low-tech thing--

478
00:21:27,910 --> 00:21:30,380
a low-tech drawing of
a low-tech experiment.

479
00:21:30,380 --> 00:21:31,870
You take a newborn infant.

480
00:21:31,870 --> 00:21:34,900
In this case, they're less
than an hour old, right?

481
00:21:34,900 --> 00:21:37,210
You've got to set up
in maternity wards.

482
00:21:37,210 --> 00:21:39,850
You want the data,
that's what you do.

483
00:21:39,850 --> 00:21:42,740
Of course, you have to ask
the parents and all of that.

484
00:21:42,740 --> 00:21:44,800
But then, you take this
infant and you sit them

485
00:21:44,800 --> 00:21:47,770
on a person's lap with
a video camera overhead,

486
00:21:47,770 --> 00:21:52,540
and you move different objects
over the infant's head, OK?

487
00:21:52,540 --> 00:21:55,900
And the different objects
that were moved, in this case,

488
00:21:55,900 --> 00:21:58,330
were patterns that were
drawn on this paddle that's

489
00:21:58,330 --> 00:22:00,220
moved over the infant's head.

490
00:22:00,220 --> 00:22:03,730
And the pattern could be a
schematic face like that,

491
00:22:03,730 --> 00:22:06,430
a scrambled schematic
face like that,

492
00:22:06,430 --> 00:22:09,850
and a blank with nothing in it.

493
00:22:09,850 --> 00:22:12,040
And what you measure is,
how far does the infant

494
00:22:12,040 --> 00:22:15,940
turn their head or their eyes
following that paddle as you

495
00:22:15,940 --> 00:22:17,350
move it over them.

496
00:22:17,350 --> 00:22:19,480
OK, nice low-tech measure.

497
00:22:19,480 --> 00:22:23,410
And what you find is they turn
their heads and their eyes

498
00:22:23,410 --> 00:22:26,440
farther when it's an actual
schematic face than when it's

499
00:22:26,440 --> 00:22:32,750
a scrambled schematic face or a
blank, within an hour of birth.

500
00:22:32,750 --> 00:22:35,650
Then you can still say,
well, their parents probably

501
00:22:35,650 --> 00:22:38,020
smiled at them quickly before
they were snatched away

502
00:22:38,020 --> 00:22:42,070
to do the experiment, so they
had some face experience,

503
00:22:42,070 --> 00:22:43,880
but boy, not a whole lot.

504
00:22:43,880 --> 00:22:46,160
And this is a very
abstract face here.

505
00:22:46,160 --> 00:22:49,480
So this has long been taken as
one of the key bits of evidence

506
00:22:49,480 --> 00:22:55,540
that something seems likely
to be innate about faces, OK?

507
00:22:55,540 --> 00:23:00,640
But now, what needs
to be innate for that?

508
00:23:00,640 --> 00:23:03,902
And it's a bizarre thing, where
this happens in the first two

509
00:23:03,902 --> 00:23:05,110
months of life and goes away.

510
00:23:05,110 --> 00:23:06,310
And there's a lot
of consideration

511
00:23:06,310 --> 00:23:07,270
of what that means.

512
00:23:07,270 --> 00:23:09,400
Maybe the first two months
is enough to bootstrap

513
00:23:09,400 --> 00:23:10,983
learning in the way
I was just talking

514
00:23:10,983 --> 00:23:14,860
about-- bootstrapping, getting
attention to the right places.

515
00:23:14,860 --> 00:23:17,470
But there's also a huge
literature on this phenomenon

516
00:23:17,470 --> 00:23:19,870
where there's a big
debate about exactly how

517
00:23:19,870 --> 00:23:21,910
simple those cues need to be.

518
00:23:21,910 --> 00:23:25,120
So people have done
many variations of this

519
00:23:25,120 --> 00:23:29,440
and one dominant story
is that all you need

520
00:23:29,440 --> 00:23:33,820
is a pattern that has more stuff
on the top than on the bottom,

521
00:23:33,820 --> 00:23:34,600
OK?

522
00:23:34,600 --> 00:23:36,280
And that's enough
that infants will

523
00:23:36,280 --> 00:23:39,080
follow this more than that.

524
00:23:39,080 --> 00:23:41,710
And the idea is that in
the visual environment

525
00:23:41,710 --> 00:23:45,920
of an infant, that's
sufficient to pick out faces.

526
00:23:45,920 --> 00:23:48,137
So there's been pushback
against this view as well.

527
00:23:48,137 --> 00:23:50,220
It's probably a little
more complicated than that.

528
00:23:50,220 --> 00:23:53,060
We won't go down the rabbit
hole of all those details,

529
00:23:53,060 --> 00:23:56,090
but whatever it is,
it's pretty simple.

530
00:23:56,090 --> 00:23:58,580
So this is another example of
what I was mentioning before

531
00:23:58,580 --> 00:23:59,750
with the Ullman case.

532
00:23:59,750 --> 00:24:02,930
This is a case where it
may be possible to build

533
00:24:02,930 --> 00:24:04,880
in something pretty basic--

534
00:24:04,880 --> 00:24:06,380
a pretty basic
template-- and then

535
00:24:06,380 --> 00:24:08,210
let learning take it from there.

536
00:24:08,210 --> 00:24:08,820
Make sense?

537
00:24:08,820 --> 00:24:10,320
If the infants are
looking at faces,

538
00:24:10,320 --> 00:24:14,030
then they can use some kind of
synaptic plasticity, whatever,

539
00:24:14,030 --> 00:24:16,370
and learn from their
experience to discriminate

540
00:24:16,370 --> 00:24:17,900
one face from another.

541
00:24:17,900 --> 00:24:21,800
OK, so these things are
present within a day or two.

542
00:24:21,800 --> 00:24:24,758
What about discrimination
of individual identity?

543
00:24:24,758 --> 00:24:26,300
First problem, how
are we going to be

544
00:24:26,300 --> 00:24:29,540
able to tell what
a newborn can see?

545
00:24:29,540 --> 00:24:31,250
And so I didn't
want you guys to be

546
00:24:31,250 --> 00:24:33,350
to thrown by this method
in the last assignment,

547
00:24:33,350 --> 00:24:37,278
so I told you where there's a
version of the explanation I'm

548
00:24:37,278 --> 00:24:38,070
just going to give.

549
00:24:38,070 --> 00:24:39,945
So if you already watched
that, my apologies.

550
00:24:39,945 --> 00:24:42,240
You can read your
email for a minute.

551
00:24:42,240 --> 00:24:45,920
So the classic experiment--

552
00:24:45,920 --> 00:24:48,500
a classic experiment--
that enabled

553
00:24:48,500 --> 00:24:52,910
us to really ask how a newborn,
non-verbal infant, what

554
00:24:52,910 --> 00:24:56,300
they see in the world, is
done by Kellman and Spelke.

555
00:24:56,300 --> 00:24:58,460
Liz Spelke up at Harvard
was at the forefront

556
00:24:58,460 --> 00:25:00,830
of getting this method
to really tell us

557
00:25:00,830 --> 00:25:03,860
a huge great deal about what
infants see and understand

558
00:25:03,860 --> 00:25:04,535
about the world.

559
00:25:04,535 --> 00:25:06,410
And this method that
I'm about to show to you

560
00:25:06,410 --> 00:25:08,990
has been the basis of what's
sometimes called "The Infancy

561
00:25:08,990 --> 00:25:12,140
Revolution," which is basically
the insight that, actually,

562
00:25:12,140 --> 00:25:14,420
infants know a lot.

563
00:25:14,420 --> 00:25:16,800
Their perceptual systems
are really sophisticated.

564
00:25:16,800 --> 00:25:17,930
They know about physics.

565
00:25:17,930 --> 00:25:19,760
They know all kinds
of social stuff.

566
00:25:19,760 --> 00:25:22,070
Within a few months of
life, they know a lot.

567
00:25:22,070 --> 00:25:24,860
And that's been a radical
change in our understanding

568
00:25:24,860 --> 00:25:27,420
of development based on
just behavioral work.

569
00:25:27,420 --> 00:25:29,360
So here's the method.

570
00:25:29,360 --> 00:25:31,430
OK, so what Spelke did--

571
00:25:31,430 --> 00:25:33,140
I always forget
to bring the demo.

572
00:25:33,140 --> 00:25:34,145
Hang on one moment.

573
00:25:36,950 --> 00:25:38,810
We don't need much.

574
00:25:38,810 --> 00:25:43,133
OK, so she showed infants
stuff like this, OK?

575
00:25:43,133 --> 00:25:44,300
The two hands are not there.

576
00:25:44,300 --> 00:25:47,600
You just arranged
to see this, OK?

577
00:25:47,600 --> 00:25:49,490
So even if you hadn't
seen me, imagine

578
00:25:49,490 --> 00:25:52,190
if you hadn't seen me pick
up the phone and the pen,

579
00:25:52,190 --> 00:25:54,500
and you didn't already
know what they were,

580
00:25:54,500 --> 00:25:56,330
and you're seeing this, OK?

581
00:25:56,330 --> 00:25:59,450
That's what they see, OK?

582
00:25:59,450 --> 00:26:01,880
So now, the question is,
when infants see that,

583
00:26:01,880 --> 00:26:04,280
do they think that that's this--

584
00:26:04,280 --> 00:26:06,980
thing behind a rectangle--

585
00:26:06,980 --> 00:26:09,260
or do they think it's
two separate bits moving

586
00:26:09,260 --> 00:26:10,400
behind the rectangle?

587
00:26:10,400 --> 00:26:12,740
It could be two separate
bits moving together, right?

588
00:26:12,740 --> 00:26:14,390
Everybody get the question?

589
00:26:14,390 --> 00:26:16,730
OK, so how would we know
what the infants thought

590
00:26:16,730 --> 00:26:18,410
was back there?

591
00:26:18,410 --> 00:26:22,880
OK, well, we use what's known
as habituation of looking time.

592
00:26:22,880 --> 00:26:24,860
Again, you sit the
infant on a parent's lap,

593
00:26:24,860 --> 00:26:26,480
and you show them
stuff, and you just

594
00:26:26,480 --> 00:26:28,280
measure how long they look.

595
00:26:28,280 --> 00:26:31,880
It's magnificently low-tech
but really profound.

596
00:26:31,880 --> 00:26:34,310
OK, so what we're
going to show here

597
00:26:34,310 --> 00:26:37,760
is how long the infant looks
on each trial as a function

598
00:26:37,760 --> 00:26:40,860
of how many times you do it.

599
00:26:40,860 --> 00:26:43,100
So you show the infant
this the first time,

600
00:26:43,100 --> 00:26:44,630
and they look for 40 seconds.

601
00:26:44,630 --> 00:26:45,650
That's a long time.

602
00:26:45,650 --> 00:26:49,410
You show them again, they look
for 35 seconds, and so forth.

603
00:26:49,410 --> 00:26:53,570
And by the fifth or sixth
time, the infant is bored.

604
00:26:53,570 --> 00:26:57,530
Like been there, done
that, bored, right?

605
00:26:57,530 --> 00:26:59,210
OK, now they're bored.

606
00:26:59,210 --> 00:27:03,540
Now we have a moment to say,
OK, what did you think it was?

607
00:27:03,540 --> 00:27:08,630
And so now, what you can
ask is, what do they think--

608
00:27:08,630 --> 00:27:11,160
you then show them
either this or this,

609
00:27:11,160 --> 00:27:14,570
and you ask them which of
those they're bored to, right?

610
00:27:14,570 --> 00:27:17,390
So the idea is if,
when looking at this,

611
00:27:17,390 --> 00:27:19,190
they thought there
was a continuous line

612
00:27:19,190 --> 00:27:22,880
behind the occluder, then they
should be more bored by this.

613
00:27:22,880 --> 00:27:25,683
But if they thought that was
two separate pieces, then

614
00:27:25,683 --> 00:27:27,100
they should be
more bored by that.

615
00:27:27,100 --> 00:27:28,298
Does that makes sense?

616
00:27:28,298 --> 00:27:30,590
Because it's the same thing
they're already bored with.

617
00:27:30,590 --> 00:27:32,007
I mean, it's not
exactly the same.

618
00:27:32,007 --> 00:27:33,620
The occluder isn't there, right?

619
00:27:33,620 --> 00:27:35,180
But it's more similar.

620
00:27:35,180 --> 00:27:36,950
OK, so here's the data.

621
00:27:36,950 --> 00:27:38,040
Here's what they find.

622
00:27:38,040 --> 00:27:39,390
So what does that mean?

623
00:27:39,390 --> 00:27:41,742
What do the infants see
when you show them this?

624
00:27:41,742 --> 00:27:42,950
It's right there in the data.

625
00:27:42,950 --> 00:27:45,260
Look at the first
test trial here.

626
00:27:45,260 --> 00:27:46,880
This is the first
test trial, when

627
00:27:46,880 --> 00:27:52,190
you show the complete
line or the broken line.

628
00:27:52,190 --> 00:27:54,410
What do they see here?

629
00:27:54,410 --> 00:27:56,090
Yeah, they saw the complete one.

630
00:27:56,090 --> 00:27:59,000
That's why, when you present
the complete one again,

631
00:27:59,000 --> 00:28:00,440
they're still bored--

632
00:28:00,440 --> 00:28:02,000
already saw that.

633
00:28:02,000 --> 00:28:03,870
Make sense?

634
00:28:03,870 --> 00:28:04,850
So isn't that awesome?

635
00:28:04,850 --> 00:28:07,430
It's so low-tech and
so simple, but this

636
00:28:07,430 --> 00:28:10,070
is how you can ask an
infant, what do you see?

637
00:28:10,070 --> 00:28:10,832
Yeah?

638
00:28:10,832 --> 00:28:13,550
AUDIENCE: Why does it switch
positions in the second trial?

639
00:28:13,550 --> 00:28:15,800
NANCY KANWISHER: You know,
frankly, I never understand

640
00:28:15,800 --> 00:28:18,080
why infant and
development people do

641
00:28:18,080 --> 00:28:19,330
a second and third trial.

642
00:28:19,330 --> 00:28:21,303
Seems to me by this
point, the jig is up.

643
00:28:21,303 --> 00:28:23,720
I think it's just because it's
hard to get enough infants,

644
00:28:23,720 --> 00:28:25,070
and you need more
data, and so they

645
00:28:25,070 --> 00:28:26,330
do a second and third trial.

646
00:28:26,330 --> 00:28:28,580
But to me, that's
the diagnostic one.

647
00:28:28,580 --> 00:28:30,710
And that's probably not
a significant switch,

648
00:28:30,710 --> 00:28:32,820
but whatever's going on
out there is obviously

649
00:28:32,820 --> 00:28:34,070
much less important than this.

650
00:28:34,070 --> 00:28:35,870
Heather, do you have a
better answer than that?

651
00:28:35,870 --> 00:28:37,287
Why do they do
those other trials?

652
00:28:37,287 --> 00:28:40,980
They always do, and it
just seems like, what?

653
00:28:40,980 --> 00:28:41,480
[LAUGHS]

654
00:28:41,480 --> 00:28:42,438
AUDIENCE: I don't know.

655
00:28:42,438 --> 00:28:44,480
NANCY KANWISHER:
Yeah, I don't either.

656
00:28:44,480 --> 00:28:45,802
AUDIENCE: [INAUDIBLE]?

657
00:28:45,802 --> 00:28:47,760
NANCY KANWISHER: Oh, you
do it every which way,

658
00:28:47,760 --> 00:28:48,930
but you do it pretty fast.

659
00:28:48,930 --> 00:28:51,420
They get bored, and you don't
want to wait half an hour

660
00:28:51,420 --> 00:28:52,470
and come back, right?

661
00:28:52,470 --> 00:28:53,553
I mean, you could do that.

662
00:28:53,553 --> 00:28:56,130
Then that would be a
memory question, right?

663
00:28:56,130 --> 00:28:57,900
Yeah, Jimmy.

664
00:28:57,900 --> 00:29:00,315
AUDIENCE: Just curious,
is this conserved

665
00:29:00,315 --> 00:29:03,640
between [INAUDIBLE] do they
all see complete lines, where

666
00:29:03,640 --> 00:29:04,140
[INAUDIBLE]?

667
00:29:04,140 --> 00:29:05,220
NANCY KANWISHER:
It's pretty robust.

668
00:29:05,220 --> 00:29:06,887
Well, OK, so first
of all, these methods

669
00:29:06,887 --> 00:29:09,060
are awesome, that you can
learn these deep things

670
00:29:09,060 --> 00:29:11,220
about perception in infants.

671
00:29:11,220 --> 00:29:12,720
But these data
are noisy as hell.

672
00:29:12,720 --> 00:29:14,220
There's no error
bars on this plot,

673
00:29:14,220 --> 00:29:16,678
but I bet if there were, you'd
have to run a lot of infants

674
00:29:16,678 --> 00:29:18,810
to get to the point where
you reach significance.

675
00:29:18,810 --> 00:29:21,270
Because a lot of times, the
infants will just throw up,

676
00:29:21,270 --> 00:29:24,160
or they'll just do what-- they
do all kinds of random things.

677
00:29:24,160 --> 00:29:26,820
So the data are extremely
noisy, and it's very hard

678
00:29:26,820 --> 00:29:28,210
to get enough data
with an infant

679
00:29:28,210 --> 00:29:30,210
to say anything about the
difference between one

680
00:29:30,210 --> 00:29:31,380
infant and another.

681
00:29:31,380 --> 00:29:34,230
By the way, there's a very
exciting development going on

682
00:29:34,230 --> 00:29:36,000
in this department
right now, where

683
00:29:36,000 --> 00:29:38,700
Kim Scott, who's a former grad
student of this department,

684
00:29:38,700 --> 00:29:40,950
has figured out how to do
looking time experiments

685
00:29:40,950 --> 00:29:43,560
like this online, OK?

686
00:29:43,560 --> 00:29:45,960
And that's hugely important,
because the number one

687
00:29:45,960 --> 00:29:48,300
bottleneck in this kind
of developmental research

688
00:29:48,300 --> 00:29:51,330
has been finding enough
infants, or getting

689
00:29:51,330 --> 00:29:52,810
enough data per infant.

690
00:29:52,810 --> 00:29:55,350
And so I think that she's going
to just crack it wide open.

691
00:29:55,350 --> 00:29:55,850
Talia?

692
00:29:55,850 --> 00:29:58,590
AUDIENCE: I guess I'm
a little bit confused

693
00:29:58,590 --> 00:30:03,270
how we know what the infant
really saw based on how long it

694
00:30:03,270 --> 00:30:05,670
looked at something.

695
00:30:05,670 --> 00:30:10,830
Could it be that maybe
they look at like--

696
00:30:10,830 --> 00:30:13,080
maybe they look at the
broken sticks longer,

697
00:30:13,080 --> 00:30:15,460
because it's like what
they thought was behind it,

698
00:30:15,460 --> 00:30:18,030
so they're now excited that
they get to see what's--

699
00:30:18,030 --> 00:30:20,730
NANCY KANWISHER: Maybe, but
then, why would you get this?

700
00:30:20,730 --> 00:30:24,030
So we know from this that
the more familiar it looks,

701
00:30:24,030 --> 00:30:26,130
the less time they look.

702
00:30:26,130 --> 00:30:27,840
So you would have to
come up with-- yeah,

703
00:30:27,840 --> 00:30:29,170
there's wiggle
room in these data,

704
00:30:29,170 --> 00:30:30,660
but you'd have to come--
your account would have

705
00:30:30,660 --> 00:30:32,520
to say, why would they
look less, and less,

706
00:30:32,520 --> 00:30:35,640
and less long when we repeat
the exact same thing, right?

707
00:30:35,640 --> 00:30:38,695
And you could tell a
story like, OK, it's

708
00:30:38,695 --> 00:30:41,070
a little bit different, because
the occluder isn't there.

709
00:30:41,070 --> 00:30:42,300
But it's a little bit
the same, and that's

710
00:30:42,300 --> 00:30:43,175
kind of edgy and fun.

711
00:30:43,175 --> 00:30:45,570
Or you could tell
another story, but I

712
00:30:45,570 --> 00:30:47,490
think the bulk of the
developmental literature

713
00:30:47,490 --> 00:30:49,410
shows that when you
do this kind of stuff,

714
00:30:49,410 --> 00:30:51,915
it's a change that
makes infants look more.

715
00:30:51,915 --> 00:30:54,540
I'm going to go on unless there
are questions of clarification,

716
00:30:54,540 --> 00:30:56,460
just because there's so
much other cool stuff.

717
00:31:00,120 --> 00:31:02,760
OK, so how can we use this
to study face recognition?

718
00:31:02,760 --> 00:31:05,130
That was just a
sidebar on the method.

719
00:31:05,130 --> 00:31:07,200
OK, so there's a
lab in Italy where

720
00:31:07,200 --> 00:31:10,230
they have an infant psychology
lab next to a maternity ward,

721
00:31:10,230 --> 00:31:13,470
and they've been doing
all these awesome studies.

722
00:31:13,470 --> 00:31:16,475
OK, and they test
1-3-day-old infants.

723
00:31:16,475 --> 00:31:17,850
And so one of the
things they did

724
00:31:17,850 --> 00:31:21,323
is show infants, just like the
paradigm I just showed you.

725
00:31:21,323 --> 00:31:23,490
They show the infant the
same face again, and again,

726
00:31:23,490 --> 00:31:24,240
and again.

727
00:31:24,240 --> 00:31:26,460
That's the habituation phase.

728
00:31:26,460 --> 00:31:28,950
And then, this is a
slightly different one.

729
00:31:28,950 --> 00:31:31,615
You give them a choice
of whether they--

730
00:31:31,615 --> 00:31:33,240
actually, you don't
give them a choice.

731
00:31:33,240 --> 00:31:34,110
I take it back.

732
00:31:34,110 --> 00:31:37,200
Yeah, you show this
condition or that condition,

733
00:31:37,200 --> 00:31:39,720
and you see how long
they look at each

734
00:31:39,720 --> 00:31:41,040
across different infants.

735
00:31:41,040 --> 00:31:43,620
And so this is the same person
from a different viewpoint.

736
00:31:43,620 --> 00:31:46,140
Actually, pretty subtle, as
we discussed with the Jenkins

737
00:31:46,140 --> 00:31:47,640
study way back.

738
00:31:47,640 --> 00:31:50,130
And that's a different
person from that viewpoint.

739
00:31:50,130 --> 00:31:53,640
And what they found is
that-- it's hard to see,

740
00:31:53,640 --> 00:31:55,740
but a very low P level
means that there's

741
00:31:55,740 --> 00:31:57,240
a significant
difference in how much

742
00:31:57,240 --> 00:32:00,250
the infants looked at those two.

743
00:32:00,250 --> 00:32:02,070
So that's pretty amazing.

744
00:32:02,070 --> 00:32:06,570
1-3-day-old infants can
apparently recognize

745
00:32:06,570 --> 00:32:10,710
the identity of a face, a novel
individual they don't already

746
00:32:10,710 --> 00:32:14,850
know, with similar-looking
faces, without hair,

747
00:32:14,850 --> 00:32:16,650
and across view changes.

748
00:32:16,650 --> 00:32:18,360
Wow, right?

749
00:32:18,360 --> 00:32:21,313
So that's pretty impressive.

750
00:32:21,313 --> 00:32:23,730
OK, and so then, they've done
all kinds of other variants.

751
00:32:23,730 --> 00:32:26,420
If you have them rotate all
the way from front profile,

752
00:32:26,420 --> 00:32:28,310
there's no longer a
significant difference.

753
00:32:28,310 --> 00:32:30,435
Infants can't do that.

754
00:32:30,435 --> 00:32:32,310
And then they do all
kinds of other variants.

755
00:32:32,310 --> 00:32:34,512
If you show them the
same individual and then

756
00:32:34,512 --> 00:32:36,470
habituate to that, they
can tell the difference

757
00:32:36,470 --> 00:32:37,310
between viewpoint.

758
00:32:37,310 --> 00:32:39,542
That's the same, and
that's different,

759
00:32:39,542 --> 00:32:41,000
even though it's
the same identity.

760
00:32:41,000 --> 00:32:42,650
So you can use this
to test what they

761
00:32:42,650 --> 00:32:45,510
think is same or different,
which is a deep question

762
00:32:45,510 --> 00:32:46,010
to ask.

763
00:32:46,010 --> 00:32:48,630
If you're interested in
representations and cognition,

764
00:32:48,630 --> 00:32:51,680
the question of what an
infant, or an animal,

765
00:32:51,680 --> 00:32:54,230
or a bunch of neurons thinks
is the same or different

766
00:32:54,230 --> 00:32:56,840
is the essence of characterizing
what it represents.

767
00:32:56,840 --> 00:32:57,680
Yeah, Quiley?

768
00:32:57,680 --> 00:33:01,385
AUDIENCE: [INAUDIBLE] the
rotated face [INAUDIBLE]??

769
00:33:01,385 --> 00:33:02,510
NANCY KANWISHER: Down here?

770
00:33:02,510 --> 00:33:03,290
Yeah.

771
00:33:03,290 --> 00:33:04,070
Yeah, they do.

772
00:33:04,070 --> 00:33:06,800
So here, basically,
it's either identical,

773
00:33:06,800 --> 00:33:08,670
or it's different
in some respect.

774
00:33:08,670 --> 00:33:13,940
So given a choice, when
it's rotated anyway,

775
00:33:13,940 --> 00:33:16,310
the familiar one
is more similar.

776
00:33:16,310 --> 00:33:18,980
But down here, this one is
more similar in viewpoint.

777
00:33:18,980 --> 00:33:20,210
Yeah?

778
00:33:20,210 --> 00:33:22,850
AUDIENCE: And these are
not like the [INAUDIBLE]

779
00:33:22,850 --> 00:33:26,338
in such [INAUDIBLE] the
student, the [INAUDIBLE]

780
00:33:26,338 --> 00:33:27,880
NANCY KANWISHER:
Sorry, say it again?

781
00:33:27,880 --> 00:33:28,630
They're not like--

782
00:33:28,630 --> 00:33:31,285
AUDIENCE: The children have
seen faces before this.

783
00:33:31,285 --> 00:33:33,160
NANCY KANWISHER: Well,
as little as possible.

784
00:33:33,160 --> 00:33:35,960
As I say, I mean, they've
seen some, but not very many,

785
00:33:35,960 --> 00:33:38,650
and they haven't
seen these faces.

786
00:33:38,650 --> 00:33:42,160
So when you're trying to get
those innateness questions,

787
00:33:42,160 --> 00:33:43,870
you go as close to
birth as you can,

788
00:33:43,870 --> 00:33:46,240
but you can't usually
go into the very moment

789
00:33:46,240 --> 00:33:47,260
of birth itself, right?

790
00:33:47,260 --> 00:33:49,270
And so there's usually
some experience,

791
00:33:49,270 --> 00:33:51,310
and it's a challenge,
but this is pretty early.

792
00:33:51,310 --> 00:33:52,270
Yeah?

793
00:33:52,270 --> 00:33:55,850
AUDIENCE: So couldn't that just
meant that the face perception

794
00:33:55,850 --> 00:33:57,240
network is just like--

795
00:33:57,240 --> 00:34:00,100
it develops really quickly,
right after [INAUDIBLE]..

796
00:34:00,100 --> 00:34:01,600
NANCY KANWISHER:
It could, it could.

797
00:34:01,600 --> 00:34:04,480
Based on these data
alone, it could.

798
00:34:04,480 --> 00:34:05,950
That's considered
kind of unlikely,

799
00:34:05,950 --> 00:34:08,440
but I agree that that's
consistent with these data.

800
00:34:08,440 --> 00:34:10,719
In the first two days
of life, the whole thing

801
00:34:10,719 --> 00:34:11,830
wires itself up.

802
00:34:11,830 --> 00:34:14,290
That's be pretty unusual.

803
00:34:14,290 --> 00:34:17,727
It's not really consistent
with those samples of neurons

804
00:34:17,727 --> 00:34:19,810
that people have looked
at elsewhere in the brain,

805
00:34:19,810 --> 00:34:22,090
but maybe there's a special
little circuit that just

806
00:34:22,090 --> 00:34:23,659
wires itself up really fast.

807
00:34:23,659 --> 00:34:27,139
So not likely, but possible, OK?

808
00:34:27,139 --> 00:34:32,090
All right, now, you might
say, well, maybe there's

809
00:34:32,090 --> 00:34:34,699
some kind of simple
visual features

810
00:34:34,699 --> 00:34:37,628
that are short of an actual
face representation here.

811
00:34:37,628 --> 00:34:39,920
This doesn't show us that
this is something about faces

812
00:34:39,920 --> 00:34:42,500
per se, even though it can
generalize across viewpoints.

813
00:34:42,500 --> 00:34:46,670
So it's not just pixel
intensity, right?

814
00:34:46,670 --> 00:34:48,620
So what is the
classic way we asked

815
00:34:48,620 --> 00:34:51,050
this question in face
perception, where we ask,

816
00:34:51,050 --> 00:34:53,420
is this really something
about faces, or is it

817
00:34:53,420 --> 00:34:56,040
something about the low-level
perceptual properties

818
00:34:56,040 --> 00:34:56,540
of the face?

819
00:34:59,670 --> 00:35:00,920
AUDIENCE: Turn it upside down?

820
00:35:00,920 --> 00:35:03,500
NANCY KANWISHER: Yeah,
turn it upside down.

821
00:35:03,500 --> 00:35:08,370
God's gift to the face
researcher, right?

822
00:35:08,370 --> 00:35:11,990
So-- oh, I guess that
was not on this slide.

823
00:35:11,990 --> 00:35:12,710
OK, right?

824
00:35:12,710 --> 00:35:14,840
OK, so now, in the
next experiment,

825
00:35:14,840 --> 00:35:18,560
they present whole faces, or
just the internal features

826
00:35:18,560 --> 00:35:22,980
without hair, or just the
external features without hair.

827
00:35:22,980 --> 00:35:25,963
So the infants can
do that at the top.

828
00:35:25,963 --> 00:35:27,380
They know those
two are different.

829
00:35:27,380 --> 00:35:29,463
They can do this here, and
they can do that there.

830
00:35:29,463 --> 00:35:30,680
OK, not too shocking yet.

831
00:35:30,680 --> 00:35:33,950
Just tells you any of those
cues can support performance.

832
00:35:33,950 --> 00:35:39,110
But now, we can ask, is
that just pattern-matching?

833
00:35:39,110 --> 00:35:39,950
No, it's not.

834
00:35:39,950 --> 00:35:41,630
Because when you turn
them upside down,

835
00:35:41,630 --> 00:35:45,290
you find that only--

836
00:35:45,290 --> 00:35:49,280
let's see, it's only
performance in this case that

837
00:35:49,280 --> 00:35:51,380
suffers when you turn
them upside-down,

838
00:35:51,380 --> 00:35:54,680
not this case or that case.

839
00:35:54,680 --> 00:35:57,410
OK, so that shows that there
are a variety of cues here that

840
00:35:57,410 --> 00:36:00,200
infants could be using, but when
you show them just the internal

841
00:36:00,200 --> 00:36:03,560
features-- the
actual face proper--

842
00:36:03,560 --> 00:36:05,930
that part, the ability to
do this discrimination,

843
00:36:05,930 --> 00:36:08,060
goes away when you
turn it upside down.

844
00:36:08,060 --> 00:36:11,600
So that part, at least,
seems to be at least somewhat

845
00:36:11,600 --> 00:36:13,910
face-specific, or
has the signature

846
00:36:13,910 --> 00:36:15,920
of face-specific processing.

847
00:36:15,920 --> 00:36:17,690
Make sense?

848
00:36:17,690 --> 00:36:20,840
OK, I mean, as a
pattern, it'd be just as

849
00:36:20,840 --> 00:36:22,490
easy to recognize
this upside-down

850
00:36:22,490 --> 00:36:24,530
and distinguish it
from that upside-down,

851
00:36:24,530 --> 00:36:26,780
if it was just the pixels
you were registering.

852
00:36:26,780 --> 00:36:28,640
But if you were doing
face processing that's

853
00:36:28,640 --> 00:36:30,182
something like adult
face processing,

854
00:36:30,182 --> 00:36:32,480
you'd expect that
inversion effect.

855
00:36:32,480 --> 00:36:37,010
OK, all right, so where are we?

856
00:36:37,010 --> 00:36:40,130
And I should just say, even
this is actively debated.

857
00:36:40,130 --> 00:36:43,580
In fact, the author
of this study

858
00:36:43,580 --> 00:36:45,530
considers this
not to be evidence

859
00:36:45,530 --> 00:36:48,020
that that processing
is face-specific.

860
00:36:48,020 --> 00:36:50,690
I think she's got some of
the strongest evidence ever,

861
00:36:50,690 --> 00:36:52,350
but she's got some
counterargument

862
00:36:52,350 --> 00:36:54,350
about how in the inverted
faces, they don't look

863
00:36:54,350 --> 00:36:55,880
as long in the situation phase.

864
00:36:55,880 --> 00:36:58,250
And so it's like I'm telling
you these cool methods,

865
00:36:58,250 --> 00:37:01,340
but boy, every one of
them can be fought over.

866
00:37:01,340 --> 00:37:03,530
OK, so where are?

867
00:37:03,530 --> 00:37:05,060
We've just shown
that discrimination

868
00:37:05,060 --> 00:37:08,450
of individual identity is
present in very young newborns,

869
00:37:08,450 --> 00:37:12,110
recognition across viewpoints,
and inversion effects

870
00:37:12,110 --> 00:37:14,960
are all present within the
first few days of life.

871
00:37:14,960 --> 00:37:18,890
OK, so newborns have very
impressive face perception

872
00:37:18,890 --> 00:37:21,980
abilities, and that's
particularly surprising

873
00:37:21,980 --> 00:37:24,690
given that their acuity
is terrible, right?

874
00:37:24,690 --> 00:37:28,280
the vision is really
blurry for young infants,

875
00:37:28,280 --> 00:37:30,750
so it's amazing that
they can do these things.

876
00:37:30,750 --> 00:37:33,740
But now, there's room for
quibbling about whether this is

877
00:37:33,740 --> 00:37:36,320
really a face-specific system.

878
00:37:36,320 --> 00:37:38,270
So the inversion
effect is suggestive,

879
00:37:38,270 --> 00:37:40,940
but they haven't totally
nailed the case about what's

880
00:37:40,940 --> 00:37:42,200
being tapped into here.

881
00:37:42,200 --> 00:37:44,480
Is it really face
perception per se--

882
00:37:44,480 --> 00:37:46,160
something specific
to face perception--

883
00:37:46,160 --> 00:37:50,420
or is it some more generic
kind of object perception?

884
00:37:50,420 --> 00:37:54,350
OK, and further, we want to
know what happens after that.

885
00:37:54,350 --> 00:37:57,985
OK, so you don't need
to memorize this table.

886
00:37:57,985 --> 00:38:00,110
I'm just going to make a
few simple points with it.

887
00:38:00,110 --> 00:38:02,030
There are lots and
lots of studies

888
00:38:02,030 --> 00:38:05,030
where people have tested
behaviorally all kinds

889
00:38:05,030 --> 00:38:07,370
of different aspects
of face perception,

890
00:38:07,370 --> 00:38:09,950
and the basic story
is that by age four,

891
00:38:09,950 --> 00:38:12,290
you see the little
smiley face means

892
00:38:12,290 --> 00:38:16,730
that this adult-like property
of the face perception system

893
00:38:16,730 --> 00:38:18,740
is present by age four.

894
00:38:18,740 --> 00:38:21,830
So all of those signatures
of face perception

895
00:38:21,830 --> 00:38:25,580
that are present in adults
are present by age four, OK?

896
00:38:25,580 --> 00:38:29,600
And in fact, much of the
action is much before that.

897
00:38:29,600 --> 00:38:31,700
You can see that
all of these things

898
00:38:31,700 --> 00:38:34,280
are present at the earliest
age they've ever been tested.

899
00:38:34,280 --> 00:38:37,520
The little square means
nobody's tested it at that age.

900
00:38:37,520 --> 00:38:43,220
So all this stuff is
developing very fast, right?

901
00:38:43,220 --> 00:38:47,067
OK, one particularly
important thing

902
00:38:47,067 --> 00:38:48,650
here that you read
about a little bit,

903
00:38:48,650 --> 00:38:51,150
but that I want to take a moment
to make sure you understand

904
00:38:51,150 --> 00:38:52,880
because it's so
interesting and cool,

905
00:38:52,880 --> 00:38:55,700
is the phenomenon of
perceptual narrowing, OK?

906
00:38:55,700 --> 00:38:57,260
And this happens
in face perception,

907
00:38:57,260 --> 00:39:00,710
and it happens in phoneme
perception in speech.

908
00:39:00,710 --> 00:39:02,550
And I'm going to do a demo here.

909
00:39:02,550 --> 00:39:04,527
So I'm going to show you
a monkey face briefly.

910
00:39:04,527 --> 00:39:06,110
OK, it's going to
come on in a second,

911
00:39:06,110 --> 00:39:07,110
and you just look at It.

912
00:39:07,110 --> 00:39:07,850
Here we go.

913
00:39:07,850 --> 00:39:10,280
Boom, there it is, OK?

914
00:39:10,280 --> 00:39:13,430
OK, in a moment, I'm going to
show you another monkey face,

915
00:39:13,430 --> 00:39:15,622
and you're going
to shout out same

916
00:39:15,622 --> 00:39:17,330
if you think it's the
same, and different

917
00:39:17,330 --> 00:39:22,250
if you think it's different,
and, huh, if you don't know.

918
00:39:22,250 --> 00:39:24,800
How many people don't know?

919
00:39:24,800 --> 00:39:27,560
Yeah, it's different, right?

920
00:39:27,560 --> 00:39:29,990
OK, well, OK, maybe
that was too hard.

921
00:39:29,990 --> 00:39:31,370
Let's try it with a human, OK?

922
00:39:31,370 --> 00:39:32,660
Remember how hard that was?

923
00:39:32,660 --> 00:39:34,382
Now let's try it
with a human face.

924
00:39:34,382 --> 00:39:35,840
I'm going to show
you a human face.

925
00:39:35,840 --> 00:39:36,590
Everybody ready?

926
00:39:36,590 --> 00:39:39,420
Here we go.

927
00:39:39,420 --> 00:39:41,010
OK?

928
00:39:41,010 --> 00:39:43,260
OK, and I'm going to
show you another human,

929
00:39:43,260 --> 00:39:45,540
and you're going to say,
is it same or different?

930
00:39:45,540 --> 00:39:48,010
Here we go.

931
00:39:48,010 --> 00:39:49,330
Duh!

932
00:39:49,330 --> 00:39:51,250
Easy, right?

933
00:39:51,250 --> 00:39:55,450
OK, so here's the amazing thing.

934
00:39:55,450 --> 00:39:57,730
You were better at
that monkey face task

935
00:39:57,730 --> 00:39:59,740
when you were six months old.

936
00:39:59,740 --> 00:40:01,570
You could do that
monkey face task

937
00:40:01,570 --> 00:40:03,340
when you were six months old.

938
00:40:03,340 --> 00:40:06,040
One of the things that you
have learned from experience

939
00:40:06,040 --> 00:40:07,960
is that you don't
need that information,

940
00:40:07,960 --> 00:40:10,340
and you threw away your
ability to do that,

941
00:40:10,340 --> 00:40:13,450
but you had it when you
were six months old.

942
00:40:13,450 --> 00:40:16,300
Isn't that awesome
and interesting?

943
00:40:16,300 --> 00:40:19,540
That's called
perceptual narrowing.

944
00:40:19,540 --> 00:40:22,790
So the experiments, in
particular, do the following.

945
00:40:22,790 --> 00:40:25,660
You use that preferential
looking paradigm--

946
00:40:25,660 --> 00:40:29,110
the preferential looking to
the novel face in infants--

947
00:40:29,110 --> 00:40:31,150
as your measure of
discrimination ability.

948
00:40:31,150 --> 00:40:33,100
What can they discriminate?

949
00:40:33,100 --> 00:40:34,960
And so you show
two human faces--

950
00:40:34,960 --> 00:40:37,070
two different
individuals, like this.

951
00:40:37,070 --> 00:40:41,140
And so now, what you see is
that at six months, nine months,

952
00:40:41,140 --> 00:40:43,750
and adulthood,
people preferentially

953
00:40:43,750 --> 00:40:47,505
look to the novel face more
than the familiar face, OK?

954
00:40:47,505 --> 00:40:48,880
That's just what
we've just done.

955
00:40:48,880 --> 00:40:54,040
People like to look at the new
thing, not the old thing, OK?

956
00:40:54,040 --> 00:40:57,288
However, if we do six months,
nine months-- oh, yeah,

957
00:40:57,288 --> 00:40:58,330
that's what we just said.

958
00:40:58,330 --> 00:40:59,450
OK, they can do that.

959
00:40:59,450 --> 00:41:03,010
So now, if you try
this on monkey faces,

960
00:41:03,010 --> 00:41:07,060
you find that
adults are like us.

961
00:41:07,060 --> 00:41:09,640
We're barely able to tell
the familiar from the novel.

962
00:41:09,640 --> 00:41:12,070
We're not so good at
monkey face discrimination.

963
00:41:12,070 --> 00:41:15,220
Nine-months-old are the same.

964
00:41:15,220 --> 00:41:20,020
But at six months,
infants can discriminate

965
00:41:20,020 --> 00:41:25,310
the monkey faces, and you could,
too, if somebody had asked you.

966
00:41:25,310 --> 00:41:30,080
So there's a very similar
phenomena with phonemes.

967
00:41:30,080 --> 00:41:32,930
Those of you who are not
native speakers of English

968
00:41:32,930 --> 00:41:35,330
maybe aware of some
phonemes in English,

969
00:41:35,330 --> 00:41:37,760
if you learned it relatively
late, that are hard for you

970
00:41:37,760 --> 00:41:39,380
to discriminate.

971
00:41:39,380 --> 00:41:41,180
There are sounds in Hindi--

972
00:41:41,180 --> 00:41:43,100
I forget, it's like
a "da" and a "ta,"

973
00:41:43,100 --> 00:41:46,700
that sound identical to me, but
that are just like completely

974
00:41:46,700 --> 00:41:49,730
obviously different to
native Hindi speakers.

975
00:41:49,730 --> 00:41:50,960
And all languages have this.

976
00:41:50,960 --> 00:41:53,930
So of the kinds of phonemes
that are discriminated

977
00:41:53,930 --> 00:41:55,700
in any language
in the world, you

978
00:41:55,700 --> 00:41:58,795
could discriminate all of those
when you were six months old.

979
00:41:58,795 --> 00:42:00,170
And one of the
things you do when

980
00:42:00,170 --> 00:42:03,080
you learn a language
is just throw together

981
00:42:03,080 --> 00:42:05,150
in the same bag things
that are actually

982
00:42:05,150 --> 00:42:07,250
different that other
people can discriminate

983
00:42:07,250 --> 00:42:11,150
if your language doesn't
discriminate it, OK?

984
00:42:11,150 --> 00:42:12,650
And so you get
that with phonemes,

985
00:42:12,650 --> 00:42:14,450
and you get it with faces.

986
00:42:14,450 --> 00:42:16,880
OK, everybody get what
perceptual narrowing is?

987
00:42:16,880 --> 00:42:18,140
OK.

988
00:42:18,140 --> 00:42:19,850
OK, you also get this--

989
00:42:19,850 --> 00:42:21,470
I mentioned this way back--

990
00:42:21,470 --> 00:42:24,080
with perceiving faces
of other races, right?

991
00:42:24,080 --> 00:42:26,600
Not just faces of other
species, but if you grow up

992
00:42:26,600 --> 00:42:30,920
in an environment
where you're only

993
00:42:30,920 --> 00:42:34,730
exposed to races A, B,
and C, and you later

994
00:42:34,730 --> 00:42:36,610
have to discriminate
faces of races

995
00:42:36,610 --> 00:42:40,190
D, E, and F, you're not
so good at it, right?

996
00:42:40,190 --> 00:42:41,300
All the same deal.

997
00:42:41,300 --> 00:42:44,360
OK, all right.

998
00:42:44,360 --> 00:42:50,420
So how would we know whether
this change between six months

999
00:42:50,420 --> 00:42:53,570
and older is just
maturation-- it's just

1000
00:42:53,570 --> 00:42:55,250
some kind of
developmental program

1001
00:42:55,250 --> 00:42:58,880
that's going on autopilot
independent of what you see,

1002
00:42:58,880 --> 00:43:00,620
or whether it's learned
from experience?

1003
00:43:04,070 --> 00:43:04,648
Josh?

1004
00:43:04,648 --> 00:43:06,190
AUDIENCE: You control
for experience.

1005
00:43:06,190 --> 00:43:09,040
NANCY KANWISHER: You control
for experience, absolutely,

1006
00:43:09,040 --> 00:43:10,870
like the Sugita paper.

1007
00:43:10,870 --> 00:43:12,710
OK, so we'll get to
that in a second.

1008
00:43:12,710 --> 00:43:14,800
So we started with these
key questions-- what

1009
00:43:14,800 --> 00:43:16,420
is the initial state
at birth, and we

1010
00:43:16,420 --> 00:43:20,620
showed impressive perceptual
abilities within a few days,

1011
00:43:20,620 --> 00:43:23,530
although people dispute
whether those abilities are

1012
00:43:23,530 --> 00:43:26,110
a face-specific system.

1013
00:43:26,110 --> 00:43:28,600
And we don't know much about
what that system is, other

1014
00:43:28,600 --> 00:43:31,510
than it works surprisingly
well given the low acuity.

1015
00:43:31,510 --> 00:43:33,430
And we showed that how
it changes after that,

1016
00:43:33,430 --> 00:43:36,940
there's perceptual narrowing
between six and 12 months,

1017
00:43:36,940 --> 00:43:39,910
but a great deal is not known
about what happens then.

1018
00:43:39,910 --> 00:43:41,710
And so now, we're
onto this question

1019
00:43:41,710 --> 00:43:45,220
of how are we going
to un-confound

1020
00:43:45,220 --> 00:43:47,140
what changes after
birth, whether it's

1021
00:43:47,140 --> 00:43:48,718
maturation or experience.

1022
00:43:48,718 --> 00:43:50,260
And I'm not going
to have time to get

1023
00:43:50,260 --> 00:43:51,552
to these other awesome methods.

1024
00:43:51,552 --> 00:43:53,920
We're going to focus on
controlled rearing, of what you

1025
00:43:53,920 --> 00:43:55,540
read the Sugita paper.

1026
00:43:55,540 --> 00:43:58,750
OK, so just to remind you
of the basics, most of you

1027
00:43:58,750 --> 00:44:00,260
seemed to get the
paper just fine.

1028
00:44:00,260 --> 00:44:02,980
The big idea was again, using
this preferential looking

1029
00:44:02,980 --> 00:44:04,870
method, what Sugita et al.

1030
00:44:04,870 --> 00:44:08,350
Showed is that when they
reared monkeys for six, 12,

1031
00:44:08,350 --> 00:44:11,080
or 24 months without ever
letting them see a face,

1032
00:44:11,080 --> 00:44:13,600
and then tested them on
the very first session

1033
00:44:13,600 --> 00:44:19,180
that they ever saw faces
with preferential looking,

1034
00:44:19,180 --> 00:44:21,730
they found that on the very
first exposure to faces,

1035
00:44:21,730 --> 00:44:25,370
the monkeys looked more at
faces compared to novel objects,

1036
00:44:25,370 --> 00:44:25,870
right?

1037
00:44:25,870 --> 00:44:29,050
They showed that face
preference, sort of akin

1038
00:44:29,050 --> 00:44:32,770
to infants looking
at the paddle,

1039
00:44:32,770 --> 00:44:35,740
and they discriminated
between faces--

1040
00:44:35,740 --> 00:44:39,250
very similar faces--
with adult-like accuracy.

1041
00:44:39,250 --> 00:44:41,570
And this part, I don't know
if you found it surprising,

1042
00:44:41,570 --> 00:44:44,500
but when this paper came
out I, was like, whoa,

1043
00:44:44,500 --> 00:44:46,270
that is crazy, right?

1044
00:44:46,270 --> 00:44:49,660
Because as I said, the whole
space of sensible hypotheses

1045
00:44:49,660 --> 00:44:52,090
is, OK, maybe a lot
of stuff is innate,

1046
00:44:52,090 --> 00:44:54,580
but you're still going to
need experience to tone it up,

1047
00:44:54,580 --> 00:44:55,750
for God's sake, right?

1048
00:44:55,750 --> 00:44:59,560
Who would think the entire
adult ability could exist

1049
00:44:59,560 --> 00:45:01,122
without any experience at all?

1050
00:45:01,122 --> 00:45:02,830
So I don't know if
you had that reaction,

1051
00:45:02,830 --> 00:45:04,455
but I think that's
a sensible reaction.

1052
00:45:04,455 --> 00:45:07,900
It's a pretty astonishing
finding in that paper.

1053
00:45:07,900 --> 00:45:10,390
Unfortunately, there's
one author on that paper.

1054
00:45:10,390 --> 00:45:13,750
It was done once, and it's
such a labor-intensive study

1055
00:45:13,750 --> 00:45:16,340
that probably nobody will
ever try to replicate it.

1056
00:45:16,340 --> 00:45:19,690
So in the back of many
people's minds is like, really?

1057
00:45:19,690 --> 00:45:23,060
Can that really be true, or
is there something funny here?

1058
00:45:23,060 --> 00:45:26,780
So I hope somebody
replicates it someday,

1059
00:45:26,780 --> 00:45:28,780
but it hasn't been done yet.

1060
00:45:28,780 --> 00:45:31,240
OK, the other thing that
you guys presumably noticed

1061
00:45:31,240 --> 00:45:33,430
is there was perceptual
narrowing in that study.

1062
00:45:33,430 --> 00:45:34,835
There were many interesting
things in there.

1063
00:45:34,835 --> 00:45:36,340
It's actually
quite a rich paper.

1064
00:45:36,340 --> 00:45:39,710
But after the initial
testing session,

1065
00:45:39,710 --> 00:45:41,380
no matter how long
the deprivation,

1066
00:45:41,380 --> 00:45:45,220
the monkeys were then housed in
either an environment with just

1067
00:45:45,220 --> 00:45:47,470
humans or just monkeys.

1068
00:45:47,470 --> 00:45:49,810
And so whether that
was 6, 12, or 24 months

1069
00:45:49,810 --> 00:45:53,860
after birth of face
deprivation, they then

1070
00:45:53,860 --> 00:45:56,230
lost their ability,
at that point,

1071
00:45:56,230 --> 00:45:59,640
to discriminate the
unexperienced faces, OK?

1072
00:45:59,640 --> 00:46:01,390
So they went through
perceptual narrowing.

1073
00:46:01,390 --> 00:46:02,890
Does that all make
sense to you guys?

1074
00:46:02,890 --> 00:46:03,460
You got that?

1075
00:46:03,460 --> 00:46:04,510
Good.

1076
00:46:04,510 --> 00:46:06,940
OK, all right.

1077
00:46:06,940 --> 00:46:09,820
So anyway, that suggests
that an awful lot of the face

1078
00:46:09,820 --> 00:46:13,630
perception system is present
without any exposure to faces,

1079
00:46:13,630 --> 00:46:15,940
and that's pretty astonishing.

1080
00:46:15,940 --> 00:46:17,800
What experience
seems to do there

1081
00:46:17,800 --> 00:46:20,620
is not create abilities,
but eliminate them

1082
00:46:20,620 --> 00:46:24,070
right for the species
that you don't see.

1083
00:46:24,070 --> 00:46:27,610
OK, so first
reaction is, really?

1084
00:46:27,610 --> 00:46:29,050
Second reaction,
is there any way

1085
00:46:29,050 --> 00:46:32,680
to account for this in terms of
some non-face-specific system?

1086
00:46:32,680 --> 00:46:34,600
I think you can, but
it takes some work,

1087
00:46:34,600 --> 00:46:37,797
and the counter-explanations are
really difficult. You can say,

1088
00:46:37,797 --> 00:46:39,880
well, maybe this is all
being carried by some more

1089
00:46:39,880 --> 00:46:42,640
generic object system.

1090
00:46:42,640 --> 00:46:45,790
They didn't test inverted
faces, unfortunately,

1091
00:46:45,790 --> 00:46:48,500
but if it was carried by
a generic object system,

1092
00:46:48,500 --> 00:46:50,500
why would you find the
perceptual narrowing?

1093
00:46:50,500 --> 00:46:52,120
Why would they have
lost their ability

1094
00:46:52,120 --> 00:46:54,410
for the unexperienced species?

1095
00:46:54,410 --> 00:46:56,793
So I think that story
is hard to tell.

1096
00:46:56,793 --> 00:46:58,210
And, of course,
the other question

1097
00:46:58,210 --> 00:46:59,668
I'm sure you guys
are wondering is,

1098
00:46:59,668 --> 00:47:01,630
what is going on in
those monkeys' brains?

1099
00:47:01,630 --> 00:47:04,810
Yeah, OK, so let's get to that.

1100
00:47:04,810 --> 00:47:07,080
Let's talk about what we
know about development

1101
00:47:07,080 --> 00:47:10,268
of this system by
looking at brains.

1102
00:47:10,268 --> 00:47:12,060
And first of all,
there's been lots of work

1103
00:47:12,060 --> 00:47:16,840
on this in older kids, age 5 and
up, going back over a decade.

1104
00:47:16,840 --> 00:47:19,380
And it's now clear that
all of that basic machinery

1105
00:47:19,380 --> 00:47:23,550
I showed you is present by age
five, in most kids age five.

1106
00:47:23,550 --> 00:47:25,870
It's continuing to
change after that,

1107
00:47:25,870 --> 00:47:29,460
but you can detect most of
that stuff by age five, or six,

1108
00:47:29,460 --> 00:47:31,090
or seven-- something like that.

1109
00:47:31,090 --> 00:47:34,470
OK, trouble is, that's
cool, but age five

1110
00:47:34,470 --> 00:47:38,070
is late with respect to
experience and with respect

1111
00:47:38,070 --> 00:47:40,830
to all those behavioral
abilities that I showed you.

1112
00:47:40,830 --> 00:47:43,800
So we need to go earlier.

1113
00:47:43,800 --> 00:47:46,020
And so a couple of years
ago, Rebecca Sachs--

1114
00:47:46,020 --> 00:47:50,790
who's straight up
there, two floors up--

1115
00:47:50,790 --> 00:47:52,770
started scanning infants, OK?

1116
00:47:52,770 --> 00:47:56,400
And this is-- as Heather can
tell you-- almost impossible.

1117
00:47:56,400 --> 00:47:59,040
It is right on the edge.

1118
00:47:59,040 --> 00:48:02,010
It took Rebecca and her lab
many, many years of work

1119
00:48:02,010 --> 00:48:05,498
over five years just to
get the system going.

1120
00:48:05,498 --> 00:48:07,290
There were all kinds
of technical advances,

1121
00:48:07,290 --> 00:48:10,440
like making scanning coils
that were optimized for infants

1122
00:48:10,440 --> 00:48:13,020
and comfortable for infants.

1123
00:48:13,020 --> 00:48:14,790
Rebecca herself went
to great lengths,

1124
00:48:14,790 --> 00:48:18,390
including producing some
of her own subjects.

1125
00:48:18,390 --> 00:48:21,930
That's her son Arthur
there and her two--

1126
00:48:21,930 --> 00:48:25,470
her grad student and postdoc
who were working with her.

1127
00:48:25,470 --> 00:48:27,750
But all of this massive
effort was worth it,

1128
00:48:27,750 --> 00:48:31,050
because what they found
was, first, for comparison,

1129
00:48:31,050 --> 00:48:35,010
this is adults with a contrast
of faces versus scenes, OK?

1130
00:48:35,010 --> 00:48:39,180
So this is basically the PPA in
blue responding more to scenes,

1131
00:48:39,180 --> 00:48:42,480
and the FFA in here and some
other face-selective bits

1132
00:48:42,480 --> 00:48:45,100
responding more to
faces in adults.

1133
00:48:45,100 --> 00:48:47,580
What do you see in
six-month-old infants?

1134
00:48:47,580 --> 00:48:49,990
It's astonishingly
similar, right?

1135
00:48:49,990 --> 00:48:53,430
You can really see a
very similar layout

1136
00:48:53,430 --> 00:48:55,410
of the functional
organization of the brain

1137
00:48:55,410 --> 00:48:58,740
already by six months.

1138
00:48:58,740 --> 00:49:00,180
So that's a huge advance.

1139
00:49:00,180 --> 00:49:02,895
That pushes way back
the timeline by which

1140
00:49:02,895 --> 00:49:04,020
these things had developed.

1141
00:49:04,020 --> 00:49:05,812
Previously, everybody
is talking about, oh,

1142
00:49:05,812 --> 00:49:07,560
what changes after age five?

1143
00:49:07,560 --> 00:49:09,340
Age five, come on?

1144
00:49:09,340 --> 00:49:11,670
OK, it's mostly
there by age six.

1145
00:49:11,670 --> 00:49:17,160
OK, now, importantly, these
systems are not adult-like.

1146
00:49:17,160 --> 00:49:18,810
Their selectivities
are very different.

1147
00:49:18,810 --> 00:49:20,960
Those regions are less
selective in infants

1148
00:49:20,960 --> 00:49:21,960
than they are in adults.

1149
00:49:21,960 --> 00:49:26,982
But the spatial layout is
there already by six months,

1150
00:49:26,982 --> 00:49:28,440
and that, importantly,
constrains--

1151
00:49:28,440 --> 00:49:30,090
whatever our model
is of development

1152
00:49:30,090 --> 00:49:31,860
that pushes it way back.

1153
00:49:31,860 --> 00:49:34,530
OK, so now, the
next questions are,

1154
00:49:34,530 --> 00:49:36,330
what is it about that region--

1155
00:49:36,330 --> 00:49:37,830
or those particular
regions-- that

1156
00:49:37,830 --> 00:49:43,230
makes them become face-specific
already by six months?

1157
00:49:43,230 --> 00:49:45,060
How does the face
system know to take up

1158
00:49:45,060 --> 00:49:48,370
residence in that systematic
location in the brain,

1159
00:49:48,370 --> 00:49:49,840
and what is the
role of experience

1160
00:49:49,840 --> 00:49:51,400
in their construction?

1161
00:49:51,400 --> 00:49:53,920
And how could we
ever answer this?

1162
00:49:53,920 --> 00:49:57,520
One way to answer that is
to use an animal model, OK?

1163
00:49:57,520 --> 00:50:00,380
So there's been-- yes.

1164
00:50:00,380 --> 00:50:02,750
AUDIENCE: OK, yeah,
similar question about--

1165
00:50:02,750 --> 00:50:04,500
NANCY KANWISHER: I'm
sorry, I didn't hear.

1166
00:50:04,500 --> 00:50:04,950
About what?

1167
00:50:04,950 --> 00:50:06,408
AUDIENCE: General
physical layout--

1168
00:50:06,408 --> 00:50:10,260
like why does your stomach
always come in the same place,

1169
00:50:10,260 --> 00:50:13,920
and would it maybe be
the same mechanism that

1170
00:50:13,920 --> 00:50:17,670
guides development of any organs
and the layout of the body,

1171
00:50:17,670 --> 00:50:19,450
[INAUDIBLE]?

1172
00:50:19,450 --> 00:50:20,440
NANCY KANWISHER: Yes.

1173
00:50:20,440 --> 00:50:23,770
Now, I don't know much about how
hearts, and kidneys, and livers

1174
00:50:23,770 --> 00:50:28,270
develop, but my understanding
is that's pretty much wired in.

1175
00:50:28,270 --> 00:50:33,280
There's some chunks of DNA that
tell you how to build a kidney

1176
00:50:33,280 --> 00:50:35,380
and where to put it
in your body, right?

1177
00:50:35,380 --> 00:50:37,600
And so that is one of
the hypotheses here.

1178
00:50:37,600 --> 00:50:39,400
It's a tempting
hypothesis, right?

1179
00:50:39,400 --> 00:50:40,750
There's all that structure.

1180
00:50:40,750 --> 00:50:43,690
It's a very tempting
hypothesis, but that doesn't

1181
00:50:43,690 --> 00:50:45,340
mean it's necessarily right.

1182
00:50:45,340 --> 00:50:46,705
Yeah, it absolutely is.

1183
00:50:46,705 --> 00:50:50,140
It's a hypothesis we should
consider and take seriously,

1184
00:50:50,140 --> 00:50:50,980
yeah.

1185
00:50:50,980 --> 00:50:52,930
OK, so but we want data.

1186
00:50:52,930 --> 00:50:53,890
We want to find out.

1187
00:50:53,890 --> 00:50:55,900
OK, so animal models.

1188
00:50:55,900 --> 00:50:58,803
So starting a few years
ago, Marge Livingstone over

1189
00:50:58,803 --> 00:51:00,220
at Harvard Med
School over there--

1190
00:51:00,220 --> 00:51:02,140
a couple of miles over there--

1191
00:51:02,140 --> 00:51:04,270
started doing these
also really amazingly

1192
00:51:04,270 --> 00:51:07,388
heroic studies where she
was scanning infant monkeys.

1193
00:51:07,388 --> 00:51:08,930
OK, now, this is
really hard to read,

1194
00:51:08,930 --> 00:51:10,540
so let me tell you
what we got here.

1195
00:51:10,540 --> 00:51:12,320
We have the cortex.

1196
00:51:12,320 --> 00:51:14,740
This is all the same animal
at different time points,

1197
00:51:14,740 --> 00:51:16,720
and each of these
things is the cortex

1198
00:51:16,720 --> 00:51:18,670
unfolded mathematically
and flattened

1199
00:51:18,670 --> 00:51:20,240
so you can see the whole thing.

1200
00:51:20,240 --> 00:51:21,907
I don't expect you
to know what's where.

1201
00:51:21,907 --> 00:51:23,710
I can barely tell myself.

1202
00:51:23,710 --> 00:51:27,970
But if you look at it, what
you see is at 81 days of age,

1203
00:51:27,970 --> 00:51:29,200
there's just blue stuff.

1204
00:51:29,200 --> 00:51:31,030
There's no orange stuff.

1205
00:51:31,030 --> 00:51:34,030
The orange stuff is the
face-selective response.

1206
00:51:34,030 --> 00:51:35,530
In fact, if you
look down, you start

1207
00:51:35,530 --> 00:51:38,400
to see, oh, that looks--
yeah, yeah, OK, that

1208
00:51:38,400 --> 00:51:39,400
looks pretty systematic.

1209
00:51:39,400 --> 00:51:41,770
It starts replicating
after that.

1210
00:51:41,770 --> 00:51:45,490
And so the claim is you
don't see face selectivity

1211
00:51:45,490 --> 00:51:49,120
until about 170 days
after birth in monkeys.

1212
00:51:49,120 --> 00:51:50,410
OK, that's about here.

1213
00:51:50,410 --> 00:51:52,390
Here's another monkey
for comparison.

1214
00:51:52,390 --> 00:51:53,920
If you stare at
it, you'll see, OK,

1215
00:51:53,920 --> 00:51:56,710
there's these systematic
bits-- boom, boom, boom, boom--

1216
00:51:56,710 --> 00:52:00,130
and maybe a little hint
at 170, but-- there's

1217
00:52:00,130 --> 00:52:03,700
some garbage up there, but
nothing systematic before that.

1218
00:52:03,700 --> 00:52:04,330
Yeah?

1219
00:52:04,330 --> 00:52:06,970
AUDIENCE: So there's no
control of the environment?

1220
00:52:06,970 --> 00:52:08,950
This is like monkeys--

1221
00:52:08,950 --> 00:52:10,420
NANCY KANWISHER:
Normal monkeys who

1222
00:52:10,420 --> 00:52:13,690
have exposure to human faces
and monkey faces hanging out

1223
00:52:13,690 --> 00:52:14,412
in the lab, yeah.

1224
00:52:14,412 --> 00:52:16,120
We haven't gotten to
control rearing yet.

1225
00:52:16,120 --> 00:52:16,930
It's coming.

1226
00:52:16,930 --> 00:52:20,310
OK, first thing is just, when
does it develop in monkeys?

1227
00:52:20,310 --> 00:52:21,520
OK, all right.

1228
00:52:21,520 --> 00:52:24,760
So are you surprised by this?

1229
00:52:24,760 --> 00:52:27,550
It's not there here,
and it is there there.

1230
00:52:27,550 --> 00:52:29,120
You should be surprised.

1231
00:52:29,120 --> 00:52:30,130
Why are you surprised?

1232
00:52:33,417 --> 00:52:34,750
This is what you guys predicted.

1233
00:52:34,750 --> 00:52:36,910
Quiley?

1234
00:52:36,910 --> 00:52:39,160
AUDIENCE: I guess I'm
surprised because they

1235
00:52:39,160 --> 00:52:40,520
were able to discriminate.

1236
00:52:40,520 --> 00:52:43,800
NANCY KANWISHER: Yeah,
what is up with that?

1237
00:52:43,800 --> 00:52:45,180
Absolutely!

1238
00:52:45,180 --> 00:52:47,730
The Sugida paper really made
it look like that system was

1239
00:52:47,730 --> 00:52:49,020
innate, right?

1240
00:52:49,020 --> 00:52:50,490
No experience-- boom!

1241
00:52:50,490 --> 00:52:51,070
They're fine.

1242
00:52:51,070 --> 00:52:53,790
It was just behavior, but it
was a good behavioral study.

1243
00:52:53,790 --> 00:52:57,640
So why the hell isn't it here?

1244
00:52:57,640 --> 00:53:01,570
Everybody with the program
on how surprising that is?

1245
00:53:01,570 --> 00:53:04,250
OK, so a bunch of things.

1246
00:53:04,250 --> 00:53:06,850
First of all-- and it
gets stable after that,

1247
00:53:06,850 --> 00:53:08,350
and replicable.

1248
00:53:08,350 --> 00:53:11,380
Well, the first thing is
one's a behavioral measure,

1249
00:53:11,380 --> 00:53:12,970
and one's a neural measure.

1250
00:53:12,970 --> 00:53:15,100
Maybe those fabulous
behavioral measures

1251
00:53:15,100 --> 00:53:18,520
weren't actually being driven
by some face-specific system.

1252
00:53:18,520 --> 00:53:20,140
Wouldn't that be sad, right?

1253
00:53:20,140 --> 00:53:21,640
I mean, they did
lots of controls.

1254
00:53:21,640 --> 00:53:22,810
It was a nice idea.

1255
00:53:22,810 --> 00:53:25,360
I thought they did as well
as they could, but who knows?

1256
00:53:25,360 --> 00:53:28,188
Maybe those monkeys
could do that task

1257
00:53:28,188 --> 00:53:30,730
with some other system and they
didn't need their face system

1258
00:53:30,730 --> 00:53:31,670
for it.

1259
00:53:31,670 --> 00:53:33,250
That's one possibility, right?

1260
00:53:33,250 --> 00:53:35,980
Then, you could have the face
system not develop till later,

1261
00:53:35,980 --> 00:53:38,440
but the monkeys
could do it before.

1262
00:53:38,440 --> 00:53:42,640
But the other thing
is, notice that Sugita

1263
00:53:42,640 --> 00:53:46,450
didn't test their monkeys
until, with the youngest ones,

1264
00:53:46,450 --> 00:53:49,310
six months of age.

1265
00:53:49,310 --> 00:53:53,740
So maybe it just got
wired up just before--

1266
00:53:53,740 --> 00:53:56,440
right there-- they
were tested, OK?

1267
00:53:56,440 --> 00:53:57,940
So it seemed
contradictory at first,

1268
00:53:57,940 --> 00:54:02,930
but it's not completely,
literally contradictory, yeah?

1269
00:54:02,930 --> 00:54:10,130
OK, all right, so now, the fact
that this stuff doesn't show up

1270
00:54:10,130 --> 00:54:14,930
until here, does that mean
that this face system requires

1271
00:54:14,930 --> 00:54:15,905
experience to develop?

1272
00:54:19,990 --> 00:54:22,770
You know the answer, because
whenever I ask that question,

1273
00:54:22,770 --> 00:54:24,660
the answer is always no.

1274
00:54:24,660 --> 00:54:28,530
Why does that not imply
that you need experience

1275
00:54:28,530 --> 00:54:30,000
with faces to wire up?

1276
00:54:30,000 --> 00:54:30,690
It's tempting.

1277
00:54:30,690 --> 00:54:31,680
You look at it,
and it's like, OK,

1278
00:54:31,680 --> 00:54:33,240
you had to look at
faces all this time

1279
00:54:33,240 --> 00:54:34,198
before you wired it up.

1280
00:54:34,198 --> 00:54:35,190
Boom, there it is--

1281
00:54:35,190 --> 00:54:37,300
very tempting.

1282
00:54:37,300 --> 00:54:39,210
But-- is it Jessica, no?

1283
00:54:39,210 --> 00:54:40,770
Sorry, what's your name?

1284
00:54:40,770 --> 00:54:41,270
Yeah.

1285
00:54:41,270 --> 00:54:42,320
AUDIENCE: Bele.

1286
00:54:42,320 --> 00:54:42,672
NANCY KANWISHER: Bele.

1287
00:54:42,672 --> 00:54:44,600
Oh, sorry, you told me
that like six times.

1288
00:54:44,600 --> 00:54:48,465
AUDIENCE: I could be merely
due to mature, physical.

1289
00:54:48,465 --> 00:54:50,590
NANCY KANWISHER: Yeah, it
could be just maturation.

1290
00:54:50,590 --> 00:54:53,090
I keep making the same point,
because it's important, right?

1291
00:54:53,090 --> 00:54:54,700
Just because it shows
up later doesn't

1292
00:54:54,700 --> 00:54:56,500
mean it's learned, right?

1293
00:54:56,500 --> 00:54:58,660
Maybe it's like puberty,
or height, or something

1294
00:54:58,660 --> 00:55:00,868
like that that's on some
developmental program that's

1295
00:55:00,868 --> 00:55:04,450
just going to unfold
independent of what you see, OK?

1296
00:55:04,450 --> 00:55:07,330
So how would we find out?

1297
00:55:07,330 --> 00:55:08,860
We would do controlled rearing.

1298
00:55:08,860 --> 00:55:12,520
And that's exactly what
these guys did, OK?

1299
00:55:12,520 --> 00:55:16,180
So in another paper that just
came out a couple of years ago,

1300
00:55:16,180 --> 00:55:19,030
they raised baby monkeys without
ever letting them see a face.

1301
00:55:19,030 --> 00:55:21,760
Much like Sugita did,
they use welder's masks

1302
00:55:21,760 --> 00:55:24,100
every time they were in the
lab, so the monkeys never

1303
00:55:24,100 --> 00:55:25,300
got to see faces.

1304
00:55:25,300 --> 00:55:27,520
And like Sugita,
they went to lengths

1305
00:55:27,520 --> 00:55:29,260
to treat the monkeys nicely.

1306
00:55:29,260 --> 00:55:31,720
They heard the calls
of their com-specifics,

1307
00:55:31,720 --> 00:55:35,080
they got lots of attention,
they had rich visual experience.

1308
00:55:35,080 --> 00:55:36,610
They just didn't see faces.

1309
00:55:36,610 --> 00:55:38,828
So it sounds kind of tragic
and horrible at first,

1310
00:55:38,828 --> 00:55:40,120
but it's actually not that bad.

1311
00:55:40,120 --> 00:55:42,490
They had social contact
and visual experience.

1312
00:55:42,490 --> 00:55:45,520
They just never saw faces--

1313
00:55:45,520 --> 00:55:47,620
both this study and
the Sugita study.

1314
00:55:47,620 --> 00:55:53,200
All right, OK, so they could
hear and smell other monkeys.

1315
00:55:53,200 --> 00:55:56,050
So the face-deprived
monkeys saw no faces at all

1316
00:55:56,050 --> 00:55:58,250
until 90 days old.

1317
00:55:58,250 --> 00:56:01,420
And at that point, they went
straight into the scanner, OK?

1318
00:56:01,420 --> 00:56:03,820
And the first time they
saw faces was inside an MRI

1319
00:56:03,820 --> 00:56:07,480
machine getting scanned, OK?

1320
00:56:07,480 --> 00:56:08,610
So what do you think?

1321
00:56:08,610 --> 00:56:11,395
Are the face-deprived monkeys
going to show face patches?

1322
00:56:16,513 --> 00:56:18,055
So there's no way
to tell, because we

1323
00:56:18,055 --> 00:56:23,350
have all these contradictory
bits of evidence here, right?

1324
00:56:23,350 --> 00:56:25,255
From Sugita, you
might think yes.

1325
00:56:29,830 --> 00:56:30,430
Hard to tell.

1326
00:56:30,430 --> 00:56:32,030
So let's just look at the data.

1327
00:56:32,030 --> 00:56:36,520
OK So here first is a normally
normally reared monkey

1328
00:56:36,520 --> 00:56:39,370
260 days old just
for comparison.

1329
00:56:39,370 --> 00:56:44,230
And those face patches in yellow
in two different monkeys here,

1330
00:56:44,230 --> 00:56:46,720
B4 and B5, left and
right hemisphere.

1331
00:56:46,720 --> 00:56:48,910
OK, so those yellow bits
are the face patches.

1332
00:56:48,910 --> 00:56:51,910
OK, normal 260-day-old monkey.

1333
00:56:51,910 --> 00:56:54,230
Now we're going to see
a face-deprived monkey,

1334
00:56:54,230 --> 00:56:55,370
260 days old.

1335
00:56:55,370 --> 00:56:58,180
This monkey was face-deprived
that entire time up

1336
00:56:58,180 --> 00:56:59,680
until scanning.

1337
00:56:59,680 --> 00:57:02,630
No face patches.

1338
00:57:02,630 --> 00:57:08,460
The plot thickens-- no
face patches at all.

1339
00:57:08,460 --> 00:57:11,670
So these guys published this
paper in a very high-profile

1340
00:57:11,670 --> 00:57:13,200
journal and said--

1341
00:57:13,200 --> 00:57:14,520
this is the title of paper--

1342
00:57:14,520 --> 00:57:18,990
"Seeing faces is necessary for
face-domain formation," OK?

1343
00:57:18,990 --> 00:57:21,860
Face domain just means
face-selective patch.

1344
00:57:21,860 --> 00:57:24,860
OK, everybody see?

1345
00:57:24,860 --> 00:57:28,880
You deprive them of face
experience, you don't see it.

1346
00:57:28,880 --> 00:57:32,600
OK, that's pretty
interesting, and it strongly

1347
00:57:32,600 --> 00:57:35,270
suggests that the face system
is not innate but depends

1348
00:57:35,270 --> 00:57:39,620
on face experience, doesn't it?

1349
00:57:39,620 --> 00:57:43,760
Rare case where the
answer is, yes, it does.

1350
00:57:43,760 --> 00:57:48,210
And it feels like it contradicts
the Sugita finding, right?

1351
00:57:48,210 --> 00:57:49,800
But not exactly.

1352
00:57:49,800 --> 00:57:51,810
You could still wiggle
out of it, right?

1353
00:57:51,810 --> 00:57:54,630
You could say, OK,
the thing that Sugita

1354
00:57:54,630 --> 00:57:57,250
was studying doesn't
use those patches,

1355
00:57:57,250 --> 00:57:58,800
so it's not flat
out contradictory.

1356
00:57:58,800 --> 00:58:00,778
Sugita was measuring
behavior; these guys

1357
00:58:00,778 --> 00:58:01,695
are looking at brains.

1358
00:58:01,695 --> 00:58:04,230
So it's kind of unsatisfying,
but it's, in principle,

1359
00:58:04,230 --> 00:58:06,600
possible.

1360
00:58:06,600 --> 00:58:08,850
Me and everyone else has
been nudging these guys

1361
00:58:08,850 --> 00:58:12,030
to run the Sugita behavioral
experiment on your monkeys,

1362
00:58:12,030 --> 00:58:13,080
please!

1363
00:58:13,080 --> 00:58:14,520
And I gather that's
getting going,

1364
00:58:14,520 --> 00:58:16,935
but I haven't seen
any of the data yet.

1365
00:58:16,935 --> 00:58:18,810
So we don't know how
that's going to resolve.

1366
00:58:18,810 --> 00:58:20,937
OK, so let's take stock.

1367
00:58:20,937 --> 00:58:22,020
What is the initial state?

1368
00:58:22,020 --> 00:58:24,120
We show with behavior
that there is

1369
00:58:24,120 --> 00:58:27,090
both attention to faces and--

1370
00:58:27,090 --> 00:58:31,230
present in newborn humans, and
face specificity seems like it,

1371
00:58:31,230 --> 00:58:34,890
but it's not totally nailed,
whereas functional MRI says

1372
00:58:34,890 --> 00:58:37,830
there's no evidence for
face specificity at birth--

1373
00:58:37,830 --> 00:58:39,720
at least in monkeys, right?

1374
00:58:39,720 --> 00:58:40,740
That's other.

1375
00:58:40,740 --> 00:58:42,870
Yeah, OK, so how are we
going to reconcile this

1376
00:58:42,870 --> 00:58:45,090
with all the behavioral
results I showed you,

1377
00:58:45,090 --> 00:58:47,340
that there seems to be
a lot of face abilities

1378
00:58:47,340 --> 00:58:49,050
present in newborns?

1379
00:58:49,050 --> 00:58:53,280
Well, one possibility is
that face specificity exists

1380
00:58:53,280 --> 00:58:57,510
behaviorally, but MRI fails--
oh, sorry, face specificity

1381
00:58:57,510 --> 00:59:00,850
exists in the brain, but
MRI fails to detect it.

1382
00:59:00,850 --> 00:59:03,480
There's a whole rigmarole about
whether functional MRI works

1383
00:59:03,480 --> 00:59:04,260
well in infants.

1384
00:59:04,260 --> 00:59:06,000
It's barely possible,
as I mentioned.

1385
00:59:06,000 --> 00:59:07,890
It's also hard with
infant monkeys.

1386
00:59:07,890 --> 00:59:10,020
Their blood flow
regulation is different.

1387
00:59:10,020 --> 00:59:12,270
They're squirming and wiggling.

1388
00:59:12,270 --> 00:59:14,430
There are a million issues
with scanning babies,

1389
00:59:14,430 --> 00:59:15,580
whether human or monkey.

1390
00:59:15,580 --> 00:59:17,580
And so you could always
say, well, it was there,

1391
00:59:17,580 --> 00:59:19,740
and just the MRI data
are just kind of crappy,

1392
00:59:19,740 --> 00:59:21,840
or blood flow
regulation to the brain

1393
00:59:21,840 --> 00:59:25,290
develops later-- an argument
many people have made.

1394
00:59:25,290 --> 00:59:27,030
However, a paper was
published last week

1395
00:59:27,030 --> 00:59:29,730
that argues against
that hypothesis.

1396
00:59:29,730 --> 00:59:33,750
The same group just showed
that the somatosensory touch

1397
00:59:33,750 --> 00:59:39,180
system is totally in place
by 11 days in baby monkeys.

1398
00:59:39,180 --> 00:59:41,910
So that suggests that you can
get really nice functional MRI

1399
00:59:41,910 --> 00:59:44,280
data at 11 days of
age in baby monkeys,

1400
00:59:44,280 --> 00:59:46,650
and it makes it less
likely that this

1401
00:59:46,650 --> 00:59:49,290
is some kind of spurious
failure to detect something

1402
00:59:49,290 --> 00:59:51,810
that was actually there.

1403
00:59:51,810 --> 00:59:54,060
I'm not going to test you
on every little detail here.

1404
00:59:54,060 --> 00:59:55,810
I want you to think
about the logic of how

1405
00:59:55,810 --> 00:59:58,770
you can ask these questions.

1406
00:59:58,770 --> 01:00:01,677
OK, the other possibility
is that the face abilities

1407
01:00:01,677 --> 01:00:04,260
that we showed behaviorally are
using some more generic object

1408
01:00:04,260 --> 01:00:07,770
recognition system, not using
this face-selective system

1409
01:00:07,770 --> 01:00:09,460
in the brain.

1410
01:00:09,460 --> 01:00:12,950
OK, so how does it
change over time?

1411
01:00:12,950 --> 01:00:14,650
Well, we showed
that behaviorally--

1412
01:00:14,650 --> 01:00:16,510
in humans, at least--
all the hallmarks

1413
01:00:16,510 --> 01:00:19,300
of face-specific processing
or present by age four,

1414
01:00:19,300 --> 01:00:21,670
and we get this perceptual
narrowing between six and 12

1415
01:00:21,670 --> 01:00:23,485
months.

1416
01:00:23,485 --> 01:00:25,360
But then we showed that
with functional MRI--

1417
01:00:25,360 --> 01:00:26,350
at least in monkeys--

1418
01:00:26,350 --> 01:00:30,140
there's no evidence for face
specificity before 200 days,

1419
01:00:30,140 --> 01:00:30,640
right?

1420
01:00:30,640 --> 01:00:33,460
AUDIENCE: [INAUDIBLE]?

1421
01:00:33,460 --> 01:00:36,280
NANCY KANWISHER: I gather
they're working on it,

1422
01:00:36,280 --> 01:00:38,740
but I haven't seen any
of the data yet, yeah.

1423
01:00:42,350 --> 01:00:45,980
OK, so that lack
of face specificity

1424
01:00:45,980 --> 01:00:47,960
is consistent with
the idea that all

1425
01:00:47,960 --> 01:00:50,690
that human early face
recognition behavior

1426
01:00:50,690 --> 01:00:52,430
is driven by a
different system--

1427
01:00:52,430 --> 01:00:55,610
because they don't have their
face system yet, presumably.

1428
01:00:55,610 --> 01:00:58,490
But it's also consistent
with this idea

1429
01:00:58,490 --> 01:01:01,340
that it's just failing
to be detected.

1430
01:01:01,340 --> 01:01:03,530
Even though I said
that's probably not true,

1431
01:01:03,530 --> 01:01:07,250
given you can detect other
stuff, it might be true here.

1432
01:01:07,250 --> 01:01:09,190
The ability to see
things with MRI

1433
01:01:09,190 --> 01:01:11,750
depends where you're
looking at the brain.

1434
01:01:11,750 --> 01:01:15,320
OK, so what about
these causal roles

1435
01:01:15,320 --> 01:01:20,210
of structured experience
and biological maturation?

1436
01:01:20,210 --> 01:01:22,970
OK, so we argued that
early face experience

1437
01:01:22,970 --> 01:01:25,160
isn't crucial for the
face recognition system.

1438
01:01:25,160 --> 01:01:27,080
That was the Sugita
paper you read.

1439
01:01:27,080 --> 01:01:30,530
But now, functional MRI is
showing that face experience

1440
01:01:30,530 --> 01:01:34,070
is necessary for the development
of face patches, at least

1441
01:01:34,070 --> 01:01:35,540
in monkeys.

1442
01:01:35,540 --> 01:01:38,475
And so a very sensible
reaction, is what, what, what?

1443
01:01:38,475 --> 01:01:40,100
How are we going to
make sense of this?

1444
01:01:40,100 --> 01:01:41,983
This is a big conundrum.

1445
01:01:41,983 --> 01:01:44,150
It's going to get worse on
Monday, where there's yet

1446
01:01:44,150 --> 01:01:47,480
more contradictory data.

1447
01:01:47,480 --> 01:01:52,750
And further, if that
face system isn't innate,

1448
01:01:52,750 --> 01:01:57,670
then what, if anything, is
innate about face perception,

1449
01:01:57,670 --> 01:01:58,210
right?

1450
01:01:58,210 --> 01:02:04,070
So maybe what all these data are
telling us is, not that much.

1451
01:02:04,070 --> 01:02:07,220
Maybe just a biased
look at faces,

1452
01:02:07,220 --> 01:02:09,440
or some very simple
image template

1453
01:02:09,440 --> 01:02:11,750
that's sufficient in the
environment of infants

1454
01:02:11,750 --> 01:02:14,330
to get them to look at faces.

1455
01:02:14,330 --> 01:02:16,280
So there's a lot of
studies I didn't have time

1456
01:02:16,280 --> 01:02:18,680
to work into this lecture,
where people stick cameras

1457
01:02:18,680 --> 01:02:21,560
on the foreheads of
newborns, and they

1458
01:02:21,560 --> 01:02:24,560
collect, what is the typical
visual experience of a newborn?

1459
01:02:24,560 --> 01:02:26,030
And then, you can take that--

1460
01:02:26,030 --> 01:02:29,345
you can take that experience
and ask, what kind of--

1461
01:02:29,345 --> 01:02:31,790
you can write
machine learning code

1462
01:02:31,790 --> 01:02:34,700
to say, what would you have to
build in to reliably pick out

1463
01:02:34,700 --> 01:02:37,820
the faces in typical
infant input?

1464
01:02:37,820 --> 01:02:39,710
And it's probably
not that complicated,

1465
01:02:39,710 --> 01:02:42,350
because infants don't see that
many different kinds of things,

1466
01:02:42,350 --> 01:02:43,220
right?

1467
01:02:43,220 --> 01:02:44,840
OK.

1468
01:02:44,840 --> 01:02:47,930
We showed early visual
discrimination abilities

1469
01:02:47,930 --> 01:02:50,030
of faces in newborn infants.

1470
01:02:50,030 --> 01:02:51,770
But again, it's not
clear that's part

1471
01:02:51,770 --> 01:02:54,218
of the face-specific system.

1472
01:02:54,218 --> 01:02:55,760
And we showed that
the face patches--

1473
01:02:55,760 --> 01:02:56,870
at least in monkeys--

1474
01:02:56,870 --> 01:02:59,930
seem to require experience, OK?

1475
01:02:59,930 --> 01:03:02,060
I'm just recapping here.

1476
01:03:02,060 --> 01:03:03,980
But now, there's
this big question of,

1477
01:03:03,980 --> 01:03:08,120
how do those face patches know
where to develop in the brain?

1478
01:03:08,120 --> 01:03:11,030
Like here they are in humans,
these little purple blobs.

1479
01:03:11,030 --> 01:03:14,270
The occipital face area I've
got two different fusiform face

1480
01:03:14,270 --> 01:03:17,030
areas, because various
people think there's two.

1481
01:03:17,030 --> 01:03:17,760
I'm not sure.

1482
01:03:17,760 --> 01:03:19,260
I don't really care;
doesn't matter.

1483
01:03:19,260 --> 01:03:22,180
Anyway, how do they know
to land right there?

1484
01:03:22,180 --> 01:03:24,680
OK, we keep bringing up this
question and dancing around it,

1485
01:03:24,680 --> 01:03:29,690
but so far, I've given no
basis for thinking about this.

1486
01:03:29,690 --> 01:03:33,920
One possibility
is that infants--

1487
01:03:33,920 --> 01:03:37,220
monkey and humans-- are
born with some earlier

1488
01:03:37,220 --> 01:03:40,110
kind of selectivity of
that patch of brain.

1489
01:03:40,110 --> 01:03:41,870
It's not a whole face template.

1490
01:03:41,870 --> 01:03:43,700
It's not a whole face system.

1491
01:03:43,700 --> 01:03:47,480
Maybe it's a bias for
curvy things, right?

1492
01:03:47,480 --> 01:03:50,750
And then, somehow, that
makes the faces land there,

1493
01:03:50,750 --> 01:03:52,340
and the system wires itself up.

1494
01:03:52,340 --> 01:03:54,380
It's not exactly clear
how that would go.

1495
01:03:54,380 --> 01:03:57,590
But that's one kind of story.

1496
01:03:57,590 --> 01:03:59,208
Another story is
based on this fact

1497
01:03:59,208 --> 01:04:01,250
I told you at the beginning
of the lecture, which

1498
01:04:01,250 --> 01:04:04,670
is most of the long-range
connectivity of the brain is

1499
01:04:04,670 --> 01:04:06,050
present at birth.

1500
01:04:06,050 --> 01:04:08,270
And so maybe the
particular connections

1501
01:04:08,270 --> 01:04:11,840
of that patch of brain are
already there at birth,

1502
01:04:11,840 --> 01:04:13,760
and maybe that
patch of connections

1503
01:04:13,760 --> 01:04:18,410
are sufficient to somehow
gate the input to that system

1504
01:04:18,410 --> 01:04:24,130
and arrange for it to end
up being face-specific, OK?

1505
01:04:24,130 --> 01:04:27,760
So this is a very active
area of investigation,

1506
01:04:27,760 --> 01:04:31,482
and there's other very active,
ongoing kinds of investigation

1507
01:04:31,482 --> 01:04:33,190
where people are trying
to understand how

1508
01:04:33,190 --> 01:04:34,970
this development might work.

1509
01:04:34,970 --> 01:04:36,580
One way people are
looking at this--

1510
01:04:36,580 --> 01:04:38,122
I mentioned this
briefly, but I think

1511
01:04:38,122 --> 01:04:40,210
it's super exciting--
is people are asking

1512
01:04:40,210 --> 01:04:42,370
with deep nets and other
kinds of modeling, what

1513
01:04:42,370 --> 01:04:45,940
do you have to build into a
system to get it to produce

1514
01:04:45,940 --> 01:04:48,100
face recognition abilities?

1515
01:04:48,100 --> 01:04:49,660
If you're trying
to make a deep net,

1516
01:04:49,660 --> 01:04:52,060
you're trying to make it really
good at face recognition,

1517
01:04:52,060 --> 01:04:54,970
do you need to give it
a template of faces?

1518
01:04:54,970 --> 01:04:57,100
Do you need to give it
only experience with faces?

1519
01:04:57,100 --> 01:04:58,990
What do you need to
build into it to get

1520
01:04:58,990 --> 01:05:00,580
it to be really good, right?

1521
01:05:00,580 --> 01:05:04,420
And so that's a very active
area of investigation.

1522
01:05:04,420 --> 01:05:07,690
And you can actually-- with some
ongoing work with Jim DiCarlo's

1523
01:05:07,690 --> 01:05:10,420
lab, we're asking, OK, deep
nets don't have topography.

1524
01:05:10,420 --> 01:05:12,940
Next door units in a deep net
doesn't mean anything, what's

1525
01:05:12,940 --> 01:05:14,320
next door versus far apart.

1526
01:05:14,320 --> 01:05:16,282
Location doesn't mean
anything in a deep net,

1527
01:05:16,282 --> 01:05:17,740
but you can make
it mean something.

1528
01:05:17,740 --> 01:05:20,470
And then you can ask when,
and whether, and how,

1529
01:05:20,470 --> 01:05:25,030
and why you get face patches
in a deep net and what

1530
01:05:25,030 --> 01:05:26,980
computational role they serve.

1531
01:05:26,980 --> 01:05:29,680
Well, totally weirdly,
I'm finishing early,

1532
01:05:29,680 --> 01:05:31,000
but I'm not going to finish.

1533
01:05:31,000 --> 01:05:32,417
I'll take questions,
and then I'll

1534
01:05:32,417 --> 01:05:34,420
maybe add a little bit more.

1535
01:05:34,420 --> 01:05:36,970
I think that was all
I had here, right.

1536
01:05:36,970 --> 01:05:38,823
Any questions about all this?

1537
01:05:38,823 --> 01:05:40,240
If it feels a
little bit chaotic--

1538
01:05:40,240 --> 01:05:42,340
I've sort of said
x and not x, and x,

1539
01:05:42,340 --> 01:05:44,620
although they're not
exactly x and not x.

1540
01:05:44,620 --> 01:05:46,210
They're just-- yeah, Sirdul.

1541
01:05:46,210 --> 01:05:49,600
AUDIENCE: So the fMRI tends
to [INAUDIBLE] activity

1542
01:05:49,600 --> 01:05:51,100
in boxes, right?

1543
01:05:51,100 --> 01:05:54,800
[INAUDIBLE] you said
contain millions of neurons.

1544
01:05:54,800 --> 01:05:57,700
So is it possible
that the neurons

1545
01:05:57,700 --> 01:06:00,730
that are specific to
faces are distributed

1546
01:06:00,730 --> 01:06:03,430
at an early age throughout
the brain, and somehow

1547
01:06:03,430 --> 01:06:05,323
the function for them--

1548
01:06:05,323 --> 01:06:07,240
NANCY KANWISHER: They
get spatially clustered.

1549
01:06:07,240 --> 01:06:08,990
AUDIENCE: Yeah, but
the neurons themselves

1550
01:06:08,990 --> 01:06:10,483
already exist at birth?

1551
01:06:10,483 --> 01:06:11,650
NANCY KANWISHER: Absolutely.

1552
01:06:11,650 --> 01:06:12,850
That's a great hypothesis.

1553
01:06:12,850 --> 01:06:13,930
It's absolutely possible.

1554
01:06:13,930 --> 01:06:15,130
Everybody get the idea?

1555
01:06:15,130 --> 01:06:17,560
You have all those
face neurons at birth,

1556
01:06:17,560 --> 01:06:19,270
and maybe they're
face-specific at birth,

1557
01:06:19,270 --> 01:06:21,250
but they're
spatially spread out.

1558
01:06:21,250 --> 01:06:23,650
And then they have
to find each other

1559
01:06:23,650 --> 01:06:25,390
and hang out together
next to each other

1560
01:06:25,390 --> 01:06:27,130
before you ever
get an MRI signal.

1561
01:06:27,130 --> 01:06:29,410
It's totally possible logically.

1562
01:06:29,410 --> 01:06:32,290
It seems to be quite
unlikely actually,

1563
01:06:32,290 --> 01:06:34,750
because it would be very
hard for all those neurons,

1564
01:06:34,750 --> 01:06:36,792
with their necessary
connections-- which is,

1565
01:06:36,792 --> 01:06:38,500
after all, how they
become face-specific,

1566
01:06:38,500 --> 01:06:40,833
is what their inputs are and
what they're connected to--

1567
01:06:40,833 --> 01:06:43,000
it'd be very hard for
them to migrate spatially

1568
01:06:43,000 --> 01:06:45,428
across the brain maintaining
their connections.

1569
01:06:45,428 --> 01:06:46,720
Yes, you're going to push back?

1570
01:06:46,720 --> 01:06:47,220
Go for it.

1571
01:06:47,220 --> 01:06:50,200
AUDIENCE: Well, I
think [INAUDIBLE]

1572
01:06:50,200 --> 01:06:52,680
But since you said
[INAUDIBLE],, they

1573
01:06:52,680 --> 01:06:54,430
care about what their
neighbors are doing.

1574
01:06:54,430 --> 01:06:57,100
So maybe it's just like
a neighboring neuron's

1575
01:06:57,100 --> 01:07:00,820
properties, but the
[INAUDIBLE] in this chain

1576
01:07:00,820 --> 01:07:03,610
moves it back until
that brief [INAUDIBLE]..

1577
01:07:03,610 --> 01:07:06,910
But that progression is the
most efficient way to pop up.

1578
01:07:06,910 --> 01:07:10,210
NANCY KANWISHER: It's totally
possible, totally possible,

1579
01:07:10,210 --> 01:07:11,320
absolutely.

1580
01:07:11,320 --> 01:07:13,870
Yep, other questions?

1581
01:07:13,870 --> 01:07:15,010
And this is wide open.

1582
01:07:15,010 --> 01:07:16,450
Nobody knows, right?

1583
01:07:19,310 --> 01:07:21,635
Let me just see what else
I have time for briefly.

1584
01:07:27,010 --> 01:07:28,600
So funny, I took
out all these slides

1585
01:07:28,600 --> 01:07:31,240
because I just thought I'm
not going to run out of time,

1586
01:07:31,240 --> 01:07:32,920
and go over, and
drive everyone crazy.

1587
01:07:38,400 --> 01:07:41,820
I moved all this stuff
to the other lecture.

1588
01:07:41,820 --> 01:07:42,765
Maybe I will just--

1589
01:07:47,197 --> 01:07:48,780
All right, hang on,
let me just glance

1590
01:07:48,780 --> 01:07:52,290
at the lineup for Wednesday.

1591
01:07:54,900 --> 01:07:55,610
Yeah?

1592
01:07:55,610 --> 01:07:58,650
AUDIENCE: Is there--
the perceptual narrowing

1593
01:07:58,650 --> 01:08:02,430
is really surprising
and fascinating.

1594
01:08:02,430 --> 01:08:09,240
Does anybody have a model for
how that processing might work

1595
01:08:09,240 --> 01:08:10,950
or what it might be for?

1596
01:08:10,950 --> 01:08:13,710
I mean, it feels
like a lot of it--

1597
01:08:13,710 --> 01:08:17,850
assumptions, or the
common sense assumptions

1598
01:08:17,850 --> 01:08:22,184
when we look at fMRI, and
when we look at neural signals

1599
01:08:22,184 --> 01:08:25,470
is that they all
mean positive things.

1600
01:08:25,470 --> 01:08:28,200
But maybe a lot of that
signals, a lot of activity

1601
01:08:28,200 --> 01:08:29,939
might be inhibitory--

1602
01:08:29,939 --> 01:08:31,828
might be the opposite.

1603
01:08:31,828 --> 01:08:33,120
NANCY KANWISHER: Totally, yeah.

1604
01:08:33,120 --> 01:08:35,760
But how would that explain
perceptual narrowing?

1605
01:08:35,760 --> 01:08:38,880
AUDIENCE: Well, if what you're
learning is what to ignore,

1606
01:08:38,880 --> 01:08:42,840
then maybe it takes a lot
of effort to ignore things.

1607
01:08:42,840 --> 01:08:46,889
And not really sure.

1608
01:08:46,889 --> 01:08:48,420
I'm not sure exactly, yeah.

1609
01:08:48,420 --> 01:08:49,350
NANCY KANWISHER: No,
it's a good point.

1610
01:08:49,350 --> 01:08:51,010
Like I mentioned
at the beginning,

1611
01:08:51,010 --> 01:08:54,689
one of the limitations
of functional MRI

1612
01:08:54,689 --> 01:08:57,120
is we don't know what the
actual neurophysiological basis

1613
01:08:57,120 --> 01:08:58,450
of the bold signal is.

1614
01:08:58,450 --> 01:09:01,649
It could be anything that
increases your metabolic costs,

1615
01:09:01,649 --> 01:09:03,390
and hence changes blood flow.

1616
01:09:03,390 --> 01:09:05,609
But one of the things that
increases metabolic costs

1617
01:09:05,609 --> 01:09:07,979
is inhibiting other neurons.

1618
01:09:07,979 --> 01:09:10,290
And so way back in the
early days of, actually,

1619
01:09:10,290 --> 01:09:12,840
PET imaging, before
functional MRI came along,

1620
01:09:12,840 --> 01:09:16,529
there was an early proto version
of a face-specific paper.

1621
01:09:16,529 --> 01:09:20,760
It didn't nail everything,
but it was not bad for 1981,

1622
01:09:20,760 --> 01:09:22,740
when I think it was published.

1623
01:09:22,740 --> 01:09:24,270
And the person who
did that paper,

1624
01:09:24,270 --> 01:09:29,310
Justine Sergent argued that
it's very, very ambiguous

1625
01:09:29,310 --> 01:09:31,740
what it means to find
a hotspot in the brain

1626
01:09:31,740 --> 01:09:33,930
where the activity--
the metabolic activity--

1627
01:09:33,930 --> 01:09:36,571
is higher, say, when you
look at faces than objects.

1628
01:09:36,571 --> 01:09:39,029
And her point was, that could
be the part of the brain that

1629
01:09:39,029 --> 01:09:41,290
really sucks at
face recognition.

1630
01:09:41,290 --> 01:09:42,750
That's the part
that's going, ah,

1631
01:09:42,750 --> 01:09:44,160
I can't deal with this thing!

1632
01:09:44,160 --> 01:09:45,700
What is this thing, right!

1633
01:09:45,700 --> 01:09:48,359
That's really bad at it, and
the neurons are firing a lot.

1634
01:09:48,359 --> 01:09:50,100
It's sort of facetious,
but sort of not.

1635
01:09:50,100 --> 01:09:51,870
And it's probably not
the right account,

1636
01:09:51,870 --> 01:09:54,180
but it is an important
reminder that we actually

1637
01:09:54,180 --> 01:09:57,240
don't know what actual
kind of neural activity

1638
01:09:57,240 --> 01:10:00,660
is driving those
things and whether it's

1639
01:10:00,660 --> 01:10:03,258
excitatory or
inhibitory, absolutely.

1640
01:10:03,258 --> 01:10:04,050
Hang on one second.

1641
01:10:04,050 --> 01:10:05,580
I feel like there was
another part of what you said

1642
01:10:05,580 --> 01:10:07,010
that I was going to engage on.

1643
01:10:07,010 --> 01:10:09,120
AUDIENCE: No, It feels
like somehow, possibly,

1644
01:10:09,120 --> 01:10:11,910
connected to the
perception [INAUDIBLE]..

1645
01:10:11,910 --> 01:10:13,990
NANCY KANWISHER: Yeah.

1646
01:10:13,990 --> 01:10:15,040
Yeah, possibly.

1647
01:10:15,040 --> 01:10:17,800
We'd have to work it out.

1648
01:10:17,800 --> 01:10:21,423
AUDIENCE: In one of the
lectures [INAUDIBLE],,

1649
01:10:21,423 --> 01:10:22,340
NANCY KANWISHER: Yeah.

1650
01:10:22,340 --> 01:10:28,290
AUDIENCE: And then, [INAUDIBLE]

1651
01:10:28,290 --> 01:10:29,165
NANCY KANWISHER: Yes.

1652
01:10:29,165 --> 01:10:30,855
AUDIENCE: [INAUDIBLE]

1653
01:10:30,855 --> 01:10:31,730
NANCY KANWISHER: Yes.

1654
01:10:31,730 --> 01:10:32,940
AUDIENCE: Then,
I'm a bit confused,

1655
01:10:32,940 --> 01:10:35,720
because, like, you said before,
almost like all the wiring is

1656
01:10:35,720 --> 01:10:36,980
[INAUDIBLE].

1657
01:10:36,980 --> 01:10:38,750
NANCY KANWISHER: OK,
long-range wiring.

1658
01:10:38,750 --> 01:10:39,410
AUDIENCE: Oh.

1659
01:10:39,410 --> 01:10:40,243
NANCY KANWISHER: OK?

1660
01:10:40,243 --> 01:10:42,260
Which is very different
than all the circuits

1661
01:10:42,260 --> 01:10:44,330
that live in each
little patch of cortex.

1662
01:10:44,330 --> 01:10:47,840
Remember, I showed
you this big change

1663
01:10:47,840 --> 01:10:53,180
in the complexity of neurons
and the number of connections.

1664
01:10:53,180 --> 01:10:54,950
Oops, looks like
we've lost it now.

1665
01:11:00,590 --> 01:11:04,130
So they're changing a lot within
each patch of cortex, right?

1666
01:11:04,130 --> 01:11:06,530
So those local circuits
that are doing computations

1667
01:11:06,530 --> 01:11:09,590
are surely changing a lot over
the first couple of years.

1668
01:11:09,590 --> 01:11:11,960
It's just the long-range
connections between that patch

1669
01:11:11,960 --> 01:11:13,490
and some remote region--

1670
01:11:13,490 --> 01:11:17,390
where it gets its inputs and
where it sends its outputs to.

1671
01:11:17,390 --> 01:11:18,290
But hang on a second.

1672
01:11:18,290 --> 01:11:22,043
You asked something-- there's
also very interesting stuff

1673
01:11:22,043 --> 01:11:23,210
about the other race effect.

1674
01:11:23,210 --> 01:11:26,060
I did mention that a
month ago or so, didn't I?

1675
01:11:26,060 --> 01:11:30,110
Which is another version of
this perceptual narrowing.

1676
01:11:30,110 --> 01:11:33,415
And in fact, a friend of mine
who's a great face researcher

1677
01:11:33,415 --> 01:11:34,790
has not yet
published this paper,

1678
01:11:34,790 --> 01:11:35,957
but she found the following.

1679
01:11:35,957 --> 01:11:38,780
Totally, that's right-- you
mentioned the adoption studies.

1680
01:11:38,780 --> 01:11:40,845
So what she has done is ask--

1681
01:11:40,845 --> 01:11:42,470
did I tell you guys
about this already?

1682
01:11:42,470 --> 01:11:43,850
I feel like I did,
but maybe not.

1683
01:11:43,850 --> 01:11:48,360
Anyway, so what you find
is that people are--

1684
01:11:48,360 --> 01:11:49,670
they all look alike.

1685
01:11:49,670 --> 01:11:52,340
Whoever they are, if
you've seen fewer of them

1686
01:11:52,340 --> 01:11:55,580
than whoever we
are, you are less

1687
01:11:55,580 --> 01:11:56,750
good at discriminating them.

1688
01:11:56,750 --> 01:11:57,920
That's just what it is.

1689
01:11:57,920 --> 01:12:02,120
But so Elinor McKone
asked if there's

1690
01:12:02,120 --> 01:12:05,000
a developmental timeline
for getting your way out

1691
01:12:05,000 --> 01:12:06,960
of the other race effect.

1692
01:12:06,960 --> 01:12:09,380
And so what she did
was-- she's in Australia,

1693
01:12:09,380 --> 01:12:11,270
and she got various
communities of people

1694
01:12:11,270 --> 01:12:14,210
who move from dominant
racial composition

1695
01:12:14,210 --> 01:12:16,700
x to dominant
racial composition y

1696
01:12:16,700 --> 01:12:20,880
and who made that move
at different ages.

1697
01:12:20,880 --> 01:12:22,790
And so what she finds
is that, actually,

1698
01:12:22,790 --> 01:12:25,580
much like learning the
phonemes of a language-- which,

1699
01:12:25,580 --> 01:12:27,556
even if you--

1700
01:12:27,556 --> 01:12:29,202
hey, let me back up a second.

1701
01:12:29,202 --> 01:12:31,160
I said that with phonemes,
you can discriminate

1702
01:12:31,160 --> 01:12:33,758
all those phonemes of all the
world's languages at birth,

1703
01:12:33,758 --> 01:12:35,300
and by six months,
you've thrown away

1704
01:12:35,300 --> 01:12:36,802
the abilities for
all the phonemes

1705
01:12:36,802 --> 01:12:37,760
you can't discriminate.

1706
01:12:37,760 --> 01:12:41,480
However, if you then go learn
a foreign language sometime

1707
01:12:41,480 --> 01:12:44,360
between six months
and, say, 12, you

1708
01:12:44,360 --> 01:12:45,540
can become a native speaker.

1709
01:12:45,540 --> 01:12:46,880
So you can learn
them back, right?

1710
01:12:46,880 --> 01:12:48,338
So there's another
window-- it gets

1711
01:12:48,338 --> 01:12:51,950
narrowed-- but you still have a
window to learn them back, OK?

1712
01:12:51,950 --> 01:12:55,460
After you're like 12,
15, whatever, forget it.

1713
01:12:55,460 --> 01:12:57,380
You won't be a native
speaker, right?

1714
01:12:57,380 --> 01:12:59,012
Same deal with the
other race effect.

1715
01:12:59,012 --> 01:13:01,220
This is exactly what McKone
found with the other race

1716
01:13:01,220 --> 01:13:02,240
effect.

1717
01:13:02,240 --> 01:13:08,330
People who moved to a different
dominant racial community

1718
01:13:08,330 --> 01:13:11,420
learned the ability to
natively discriminate people

1719
01:13:11,420 --> 01:13:15,020
in that other race if
they moved before age 12.

1720
01:13:15,020 --> 01:13:17,277
So it really seems like
there's some general ability.

1721
01:13:17,277 --> 01:13:18,860
Oh, I remember David's
other question.

1722
01:13:18,860 --> 01:13:20,130
Why does this make sense?

1723
01:13:20,130 --> 01:13:21,797
I don't know exactly
why it makes sense,

1724
01:13:21,797 --> 01:13:26,570
but certainly, neural activity
is expensive metabolically,

1725
01:13:26,570 --> 01:13:28,310
and we don't want to
make discriminations

1726
01:13:28,310 --> 01:13:29,750
we don't have to.

1727
01:13:29,750 --> 01:13:32,990
And so it can be just that
the nervous system is learning

1728
01:13:32,990 --> 01:13:34,790
what kinds of
discriminations it needs

1729
01:13:34,790 --> 01:13:37,487
to make in its environment and
what kind it doesn't, right?

1730
01:13:37,487 --> 01:13:39,320
And with the case of
phonemes, it's actually

1731
01:13:39,320 --> 01:13:41,360
part of what you're doing
in speech perception,

1732
01:13:41,360 --> 01:13:45,110
is you want to know,
every time I say "ba,"

1733
01:13:45,110 --> 01:13:47,130
it sounds different in
all different contexts.

1734
01:13:47,130 --> 01:13:50,030
And so part of the essence
of the difficulty in speech

1735
01:13:50,030 --> 01:13:52,520
recognition is understanding
that all those different "ba"s

1736
01:13:52,520 --> 01:13:54,980
are the same sound, right?

1737
01:13:54,980 --> 01:13:57,650
And so part of what perceptual
narrowing might be doing

1738
01:13:57,650 --> 01:13:59,210
is saying all those things--

1739
01:13:59,210 --> 01:14:02,180
"da," "ta," whatever
it is in Hindi--

1740
01:14:02,180 --> 01:14:04,738
those are all going to
count as the same thing.

1741
01:14:04,738 --> 01:14:06,530
And that's going to
help you process speech

1742
01:14:06,530 --> 01:14:09,140
in your native language
but hinder when you try

1743
01:14:09,140 --> 01:14:12,630
to learn a foreign language.

1744
01:14:12,630 --> 01:14:13,130
Yeah?

1745
01:14:13,130 --> 01:14:14,630
AUDIENCE: So something
I'm wondering

1746
01:14:14,630 --> 01:14:18,620
with perceptual narrowing is how
general like the starting point

1747
01:14:18,620 --> 01:14:19,120
is.

1748
01:14:19,120 --> 01:14:21,287
So I'm basically wondering--
because in the studies,

1749
01:14:21,287 --> 01:14:22,920
they compared human
and monkey faces.

1750
01:14:22,920 --> 01:14:23,300
NANCY KANWISHER: Yeah.

1751
01:14:23,300 --> 01:14:24,883
AUDIENCE: And I'm
wondering if there's

1752
01:14:24,883 --> 01:14:28,110
any correlation with
how similar the DNA,

1753
01:14:28,110 --> 01:14:31,310
like how they're able to
discriminate between the faces.

1754
01:14:31,310 --> 01:14:33,800
So whether that's
different types of monkeys,

1755
01:14:33,800 --> 01:14:35,300
or different animals--

1756
01:14:35,300 --> 01:14:37,092
NANCY KANWISHER: I'm
not getting it, right?

1757
01:14:37,092 --> 01:14:39,650
Early on, you can
discriminate both, right?

1758
01:14:39,650 --> 01:14:40,860
So what's the question?

1759
01:14:40,860 --> 01:14:43,220
AUDIENCE: So I'm wondering
what other animals can they

1760
01:14:43,220 --> 01:14:44,090
discriminate, and what--

1761
01:14:44,090 --> 01:14:44,750
NANCY KANWISHER: I see, I see.

1762
01:14:44,750 --> 01:14:45,710
How far does it go?

1763
01:14:45,710 --> 01:14:47,430
Yeah, good question.

1764
01:14:47,430 --> 01:14:50,750
I don't know that anybody
has asked little kids if they

1765
01:14:50,750 --> 01:14:56,030
can discriminate other kinds of
faces other than monkey faces.

1766
01:14:56,030 --> 01:14:58,970
I'm sure there's some limit
to it-- like fish faces?

1767
01:14:58,970 --> 01:15:03,410
Probably, I don't know, yeah.

1768
01:15:03,410 --> 01:15:05,877
But there's also, actually,
in terms of that extended--

1769
01:15:05,877 --> 01:15:07,460
I don't know the
answer to that, yeah.

1770
01:15:07,460 --> 01:15:09,080
There's going to be some limit.

1771
01:15:09,080 --> 01:15:14,240
But in terms of the
question of how long

1772
01:15:14,240 --> 01:15:17,030
can you relearn those
abilities or maintain them,

1773
01:15:17,030 --> 01:15:19,910
it's not like
perceptual narrowing

1774
01:15:19,910 --> 01:15:22,920
is going to happen at
six months automatically.

1775
01:15:22,920 --> 01:15:26,090
So if you manipulate it--
so the studies on humans,

1776
01:15:26,090 --> 01:15:27,135
if you send--

1777
01:15:27,135 --> 01:15:28,760
I feel like I said
this in here before,

1778
01:15:28,760 --> 01:15:30,740
but it must have
been somewhere else--

1779
01:15:30,740 --> 01:15:33,830
if you send parents
home with books

1780
01:15:33,830 --> 01:15:37,610
with monkey pictures in them,
parents of six-month-olds,

1781
01:15:37,610 --> 01:15:39,740
and you say, look,
every night, go

1782
01:15:39,740 --> 01:15:42,410
through the book with your kids
and say, there's Monkey Joe,

1783
01:15:42,410 --> 01:15:44,060
and there's Monkey
Bob, and there's

1784
01:15:44,060 --> 01:15:47,870
Monkey Whoever with
your kids, and you

1785
01:15:47,870 --> 01:15:50,240
have them do that from age
six months to 12 months,

1786
01:15:50,240 --> 01:15:52,520
they don't perceptually
narrow, because they continue

1787
01:15:52,520 --> 01:15:55,460
to get that experience, right?

1788
01:15:55,460 --> 01:15:58,580
Interestingly, if the
parents go home and just say,

1789
01:15:58,580 --> 01:16:02,420
look, look, that doesn't do it.

1790
01:16:02,420 --> 01:16:05,090
You have to give them some
social cue that is essentially

1791
01:16:05,090 --> 01:16:07,940
saying, this thing is
different from that thing.

1792
01:16:07,940 --> 01:16:09,560
And if you do that
with an infant,

1793
01:16:09,560 --> 01:16:11,768
even when they don't really
understand language much,

1794
01:16:11,768 --> 01:16:14,065
they get that cue, and they
learn to discriminate--

1795
01:16:14,065 --> 01:16:16,565
or they maintain their ability
to discriminate monkey faces.

1796
01:16:19,520 --> 01:16:20,020
Yeah?

1797
01:16:20,020 --> 01:16:21,562
AUDIENCE: Does that
hold up even when

1798
01:16:21,562 --> 01:16:23,492
they're past the 12 months old?

1799
01:16:23,492 --> 01:16:24,950
NANCY KANWISHER:
Well, I'm guessing

1800
01:16:24,950 --> 01:16:29,090
it will be just like the
case that McKone showed

1801
01:16:29,090 --> 01:16:31,240
with other race effects, right?

1802
01:16:31,240 --> 01:16:32,990
I'm guessing the other
species effect will

1803
01:16:32,990 --> 01:16:35,870
be like the other
race effect in that

1804
01:16:35,870 --> 01:16:38,300
if you, say, start
working in a monkey lab

1805
01:16:38,300 --> 01:16:40,670
when you're eight years old--

1806
01:16:40,670 --> 01:16:43,190
that would be weird,
but you could--

1807
01:16:43,190 --> 01:16:44,960
or you-- I don't know, whatever.

1808
01:16:44,960 --> 01:16:48,410
Anyway, that you would be able
to relearn it on the same time

1809
01:16:48,410 --> 01:16:51,440
scale that you would relearn--

1810
01:16:51,440 --> 01:16:54,800
relearn, or learn for the
first time, previously

1811
01:16:54,800 --> 01:16:56,630
unfamiliar races of faces.

1812
01:16:56,630 --> 01:17:00,270
But maybe those are slightly
different timelines.

1813
01:17:00,270 --> 01:17:01,320
Yeah?

1814
01:17:01,320 --> 01:17:03,120
AUDIENCE: Could you
do something similar

1815
01:17:03,120 --> 01:17:05,370
with the monkey faces,
but with phonemes

1816
01:17:05,370 --> 01:17:06,623
in different languages?

1817
01:17:06,623 --> 01:17:08,040
NANCY KANWISHER:
I'm sure you can,

1818
01:17:08,040 --> 01:17:09,600
and I'm sure that
has been done, but I

1819
01:17:09,600 --> 01:17:10,725
don't know that literature.

1820
01:17:10,725 --> 01:17:11,260
Yeah.

1821
01:17:11,260 --> 01:17:12,630
Yeah, you mean like keep--

1822
01:17:12,630 --> 01:17:13,290
well, OK.

1823
01:17:13,290 --> 01:17:15,480
I mean, it essentially
does get done, right?

1824
01:17:15,480 --> 01:17:20,680
So kids who stay
in environments--

1825
01:17:20,680 --> 01:17:21,750
let me think about this.

1826
01:17:21,750 --> 01:17:24,660
Well, certainly, an
infant who's being

1827
01:17:24,660 --> 01:17:26,460
raised in a
bilingual environment

1828
01:17:26,460 --> 01:17:29,130
will maintain their ability
to discriminate those phonemes

1829
01:17:29,130 --> 01:17:32,348
from any of the languages
they hear, right?

1830
01:17:32,348 --> 01:17:34,890
AUDIENCE: So you're saying, with
the monkeys thing, some kind

1831
01:17:34,890 --> 01:17:37,440
of social cue to know that--

1832
01:17:37,440 --> 01:17:39,240
NANCY KANWISHER: I
suspect that's true.

1833
01:17:39,240 --> 01:17:40,948
I don't know this
literature well enough.

1834
01:17:40,948 --> 01:17:43,317
I do know-- yeah, actually,
it's coming back dimly.

1835
01:17:43,317 --> 01:17:44,400
Heather, do you know this?

1836
01:17:44,400 --> 01:17:45,000
Janet Werker

1837
01:17:45,000 --> 01:17:46,740
AUDIENCE: [INAUDIBLE].

1838
01:17:46,740 --> 01:17:48,240
NANCY KANWISHER:
OK, so Janet Werker

1839
01:17:48,240 --> 01:17:51,810
is this amazing infant
phoneme perception researcher.

1840
01:17:51,810 --> 01:17:55,650
And I'm pretty sure that
if you present infants

1841
01:17:55,650 --> 01:17:57,630
with just like a TV
in the background

1842
01:17:57,630 --> 01:18:01,200
with a foreign language, even
if the infant doesn't have much

1843
01:18:01,200 --> 01:18:03,270
else to do, that's not enough.

1844
01:18:03,270 --> 01:18:05,580
And you need to look at
them, and engage with them,

1845
01:18:05,580 --> 01:18:09,632
and speak mother-ese-- like,
hey, infant, blah, blah, right?

1846
01:18:09,632 --> 01:18:12,090
I think you need to do all of
that for them to maintain it,

1847
01:18:12,090 --> 01:18:12,668
but I'm--

1848
01:18:12,668 --> 01:18:13,960
AUDIENCE: Yeah, that's correct.

1849
01:18:13,960 --> 01:18:15,668
I think there also
has to be interaction.

1850
01:18:15,668 --> 01:18:18,420
They can't also just be
watching the [INAUDIBLE]..

1851
01:18:18,420 --> 01:18:20,220
It has to be
slightly [INAUDIBLE]

1852
01:18:20,220 --> 01:18:21,340
reciprocal [INAUDIBLE].

1853
01:18:21,340 --> 01:18:23,460
AUDIENCE: And the
fact that [INAUDIBLE]..

1854
01:18:26,108 --> 01:18:27,400
NANCY KANWISHER: Correct, yeah.

1855
01:18:27,400 --> 01:18:29,516
AUDIENCE: So even if it's
not just [INAUDIBLE],,

1856
01:18:29,516 --> 01:18:30,752
it has to be [INAUDIBLE].

1857
01:18:30,752 --> 01:18:32,210
NANCY KANWISHER:
It has to be what?

1858
01:18:32,210 --> 01:18:33,918
AUDIENCE: It has to
be like [INAUDIBLE]..

1859
01:18:33,918 --> 01:18:35,390
It can't be [INAUDIBLE].

1860
01:18:35,390 --> 01:18:38,190
AUDIENCE: Yeah, which makes me
of [INAUDIBLE] or something--

1861
01:18:38,190 --> 01:18:40,490
like if you interact in
different ways, [INAUDIBLE]..

1862
01:18:43,040 --> 01:18:44,430
NANCY KANWISHER: Cool, yeah?

1863
01:18:44,430 --> 01:18:47,270
AUDIENCE: Yeah, I have a
question about how long

1864
01:18:47,270 --> 01:18:49,286
that [INAUDIBLE] lasts.

1865
01:18:49,286 --> 01:18:50,930
If someone spoke
a foreign language

1866
01:18:50,930 --> 01:18:53,220
when they were younger,
then moved somewhere else

1867
01:18:53,220 --> 01:18:56,120
or were adopted and then
stopped speaking the language,

1868
01:18:56,120 --> 01:19:03,420
[INAUDIBLE],, could they
sort of be [INAUDIBLE]??

1869
01:19:03,420 --> 01:19:04,670
NANCY KANWISHER: I don't know.

1870
01:19:04,670 --> 01:19:06,253
I'm sure there's a
literature on that.

1871
01:19:06,253 --> 01:19:07,970
You don't know
that, Dana, do you?

1872
01:19:07,970 --> 01:19:11,090
Sorry, like so you're
raised bilingual,

1873
01:19:11,090 --> 01:19:13,580
and then you stop having
the experience early

1874
01:19:13,580 --> 01:19:16,100
on from your second
language, and then

1875
01:19:16,100 --> 01:19:18,860
you're re-exposed
later at age eight?

1876
01:19:18,860 --> 01:19:19,850
AUDIENCE: [INAUDIBLE].

1877
01:19:19,850 --> 01:19:21,640
AUDIENCE: Yeah, you
still have the--

1878
01:19:21,640 --> 01:19:23,140
yeah, you maintain
the [INAUDIBLE]..

1879
01:19:23,140 --> 01:19:24,100
AUDIENCE: Yeah, like after--

1880
01:19:24,100 --> 01:19:25,475
NANCY KANWISHER:
Well, but wait--

1881
01:19:25,475 --> 01:19:27,935
AUDIENCE: But you're not able
to speak the language, right?

1882
01:19:27,935 --> 01:19:28,560
AUDIENCE: Yeah.

1883
01:19:28,560 --> 01:19:29,840
AUDIENCE: But you
still [INAUDIBLE]..

1884
01:19:29,840 --> 01:19:30,290
AUDIENCE: But I guess you--

1885
01:19:30,290 --> 01:19:31,665
NANCY KANWISHER:
But then, that's

1886
01:19:31,665 --> 01:19:33,565
not consistent with
perceptual narrowing.

1887
01:19:33,565 --> 01:19:35,690
AUDIENCE: If you're exposed
to it before two years?

1888
01:19:35,690 --> 01:19:36,170
AUDIENCE: Yeah.

1889
01:19:36,170 --> 01:19:36,260
NANCY KANWISHER: Yeah.

1890
01:19:36,260 --> 01:19:37,635
AUDIENCE: And then
you move away?

1891
01:19:37,635 --> 01:19:39,010
NANCY KANWISHER:
Well, if it goes

1892
01:19:39,010 --> 01:19:40,880
beyond that six-month
thing, yeah, OK.

1893
01:19:40,880 --> 01:19:42,620
AUDIENCE: I think
that's the case, yeah.

1894
01:19:42,620 --> 01:19:44,287
You might not have
the higher structure,

1895
01:19:44,287 --> 01:19:48,170
but if you like you the
syntax and some vocabulary,

1896
01:19:48,170 --> 01:19:51,170
you'll have a better accent
than someone who did not

1897
01:19:51,170 --> 01:19:53,090
have that early
experience, might not

1898
01:19:53,090 --> 01:19:55,990
be able to differentiate
[INAUDIBLE]..

1899
01:19:55,990 --> 01:19:56,650
But--

1900
01:19:56,650 --> 01:19:57,942
AUDIENCE: You just [INAUDIBLE].

1901
01:20:00,460 --> 01:20:02,937
AUDIENCE: [INAUDIBLE],,
I think that's correct.

1902
01:20:02,937 --> 01:20:04,020
NANCY KANWISHER: OK, good.

1903
01:20:04,020 --> 01:20:04,770
One more question.

1904
01:20:04,770 --> 01:20:05,440
Josh?

1905
01:20:05,440 --> 01:20:07,023
AUDIENCE: So do we
know of cases where

1906
01:20:07,023 --> 01:20:12,000
there's [INAUDIBLE] a mismatch
between [INAUDIBLE] sort

1907
01:20:12,000 --> 01:20:13,300
of information?

1908
01:20:13,300 --> 01:20:13,800
Like--

1909
01:20:13,800 --> 01:20:14,460
NANCY KANWISHER: Like this?

1910
01:20:14,460 --> 01:20:16,350
AUDIENCE: Yeah, like--
with this property

1911
01:20:16,350 --> 01:20:18,720
in some of the domain of
some of the [INAUDIBLE]..

1912
01:20:18,720 --> 01:20:20,495
Basically be [INAUDIBLE].

1913
01:20:20,495 --> 01:20:21,870
NANCY KANWISHER:
Oh, god, I don't

1914
01:20:21,870 --> 01:20:25,290
have my dictionary of
knowledge filed that way

1915
01:20:25,290 --> 01:20:27,090
so I can pull up an
instance of that,

1916
01:20:27,090 --> 01:20:29,115
but I'm sure there
are loads of those.

1917
01:20:29,115 --> 01:20:30,153
AUDIENCE: [INAUDIBLE].

1918
01:20:30,153 --> 01:20:32,070
NANCY KANWISHER: Yeah,
well, because when we--

1919
01:20:32,070 --> 01:20:34,320
because we're making all
these assumptions about which

1920
01:20:34,320 --> 01:20:37,108
behavioral ability is subserved
by some particular activation

1921
01:20:37,108 --> 01:20:37,650
in the brain.

1922
01:20:37,650 --> 01:20:40,238
And mostly, we
don't know, right?

1923
01:20:40,238 --> 01:20:42,030
We know when we have
the rare opportunities

1924
01:20:42,030 --> 01:20:42,990
to do causal tests.

1925
01:20:42,990 --> 01:20:44,850
We have a better
idea that that system

1926
01:20:44,850 --> 01:20:48,750
is at least causally involved
in that behavioral ability.

1927
01:20:48,750 --> 01:20:53,890
But yeah, often, those links
are much looser than we'd like.

1928
01:20:53,890 --> 01:20:56,550
All right, see you
guys Wednesday.