WEBVTT

00:00:00.000 --> 00:00:04.780
[LOGO SOUNDS]

00:00:10.000 --> 00:00:12.500
NANCY KANWISHER: So I'm doing
another one of these big mongo

00:00:12.500 --> 00:00:13.875
lectures that
takes a whole week,

00:00:13.875 --> 00:00:16.755
so this is really a
continuation of last time.

00:00:16.755 --> 00:00:18.380
This is the outline
for the whole week.

00:00:18.380 --> 00:00:21.920
We got through most of the
stuff on face perception.

00:00:21.920 --> 00:00:23.360
I'll do some more today.

00:00:23.360 --> 00:00:25.200
We're right there.

00:00:25.200 --> 00:00:27.590
And we're going to go on and
consider this question of,

00:00:27.590 --> 00:00:30.800
what's innate, and how
do you wire up brains?

00:00:30.800 --> 00:00:35.540
So first, a brief recap of
main points from last time.

00:00:35.540 --> 00:00:38.510
What, if anything, is innate
about face perception?

00:00:38.510 --> 00:00:41.360
We considered lots of different
kinds of evidence, behavioral

00:00:41.360 --> 00:00:42.290
and neural.

00:00:42.290 --> 00:00:46.130
And the bottom line is,
maybe not that much.

00:00:46.130 --> 00:00:49.280
So there's a few things
that are sort of suggestive,

00:00:49.280 --> 00:00:52.190
like newborns have this
bias to look at faces more

00:00:52.190 --> 00:00:54.050
than other non-face
stimuli that are

00:00:54.050 --> 00:00:56.900
pretty similar-- schematic
faces versus scrambled schematic

00:00:56.900 --> 00:00:57.740
faces.

00:00:57.740 --> 00:00:59.300
And that's suggestive.

00:00:59.300 --> 00:01:01.520
But then there's the
possibility that that's

00:01:01.520 --> 00:01:04.190
just due to some very,
very simple property

00:01:04.190 --> 00:01:06.680
of those stimuli, namely
just having more junk

00:01:06.680 --> 00:01:09.890
on the top than the bottom, like
eyes on the top than bottom.

00:01:09.890 --> 00:01:11.750
So what would have to
be innate in that case

00:01:11.750 --> 00:01:15.200
would be just the simplest
possible template, not even

00:01:15.200 --> 00:01:17.200
a whole face.

00:01:17.200 --> 00:01:18.950
Similarly, we showed
that there's actually

00:01:18.950 --> 00:01:21.290
very good discrimination
of one face from another,

00:01:21.290 --> 00:01:25.820
even across viewpoint
changes in newborn humans,

00:01:25.820 --> 00:01:29.780
and also in monkeys that
were raised without ever

00:01:29.780 --> 00:01:32.120
being allowed to see faces.

00:01:32.120 --> 00:01:36.260
And both of those things
suggest innate abilities

00:01:36.260 --> 00:01:39.510
to process faces,
but in both cases,

00:01:39.510 --> 00:01:43.100
it's possible to argue that
that ability isn't due to face

00:01:43.100 --> 00:01:44.820
mechanisms in particular.

00:01:44.820 --> 00:01:47.750
It's due to just general
vision and shape perception.

00:01:50.900 --> 00:01:53.870
Third, I showed you
beautiful recent data

00:01:53.870 --> 00:01:56.870
showing that the face
patches in monkeys

00:01:56.870 --> 00:01:58.820
don't develop if
monkeys are reared

00:01:58.820 --> 00:02:00.440
without ever seeing faces.

00:02:03.140 --> 00:02:06.470
Which also suggests that
maybe not that much is innate.

00:02:06.470 --> 00:02:09.229
So all that is fine,
but then there's

00:02:09.229 --> 00:02:12.020
a big, wide open
question that's left

00:02:12.020 --> 00:02:13.820
unanswered by all
of that, which is,

00:02:13.820 --> 00:02:18.890
how do the face areas know to
land right there in everybody,

00:02:18.890 --> 00:02:20.090
robustly?

00:02:20.090 --> 00:02:22.010
That really feels
like something has

00:02:22.010 --> 00:02:25.050
to be innate about
the brain, at least,

00:02:25.050 --> 00:02:27.710
to say where those
things should go.

00:02:27.710 --> 00:02:30.620
OK, so one possibility that
I'm sort of skipping over,

00:02:30.620 --> 00:02:32.240
because it's a whole
little universe,

00:02:32.240 --> 00:02:34.550
and there isn't an answer yet--
people are working on it right

00:02:34.550 --> 00:02:36.925
now, people in this building
are working on it right now,

00:02:36.925 --> 00:02:43.220
but the gist of the idea
is that maybe what's innate

00:02:43.220 --> 00:02:46.070
is some other kind of
simpler selectivity.

00:02:46.070 --> 00:02:48.500
Maybe like selectivity
for curved things.

00:02:48.500 --> 00:02:51.350
Remember how I talked about,
as you go up the visual system,

00:02:51.350 --> 00:02:55.430
you start with selectivity for
spots of light and then edges?

00:02:55.430 --> 00:02:56.930
Well, maybe up
there, you're born

00:02:56.930 --> 00:02:59.120
with selectivity
for curved things,

00:02:59.120 --> 00:03:01.820
or something like that,
that is face-like enough

00:03:01.820 --> 00:03:04.520
that somehow that leads
face selectivity to land

00:03:04.520 --> 00:03:05.262
there later.

00:03:05.262 --> 00:03:07.220
It's kind of vague because
nobody really knows,

00:03:07.220 --> 00:03:09.710
but that's an idea.

00:03:09.710 --> 00:03:12.500
Another possibility that we'll
talk more about in a moment

00:03:12.500 --> 00:03:15.540
is a possibility that the reason
your face patches land right

00:03:15.540 --> 00:03:18.260
there is something about
the long-range structural

00:03:18.260 --> 00:03:20.720
connectivity of that region
to the rest of the brain

00:03:20.720 --> 00:03:22.920
makes that the right place.

00:03:22.920 --> 00:03:25.640
And so all of this is very
actively being investigated,

00:03:25.640 --> 00:03:28.190
and nobody knows the
right answer here.

00:03:28.190 --> 00:03:32.630
Further, I just want to mention
that deep net modeling is just

00:03:32.630 --> 00:03:35.870
very suddenly in the last year
become a very powerful way

00:03:35.870 --> 00:03:38.940
to approach these same questions
from a different angle.

00:03:38.940 --> 00:03:42.890
So with deep nets, you
can ask, what do you

00:03:42.890 --> 00:03:47.840
need to build into a network to
get it to produce face patches?

00:03:47.840 --> 00:03:51.020
So that's a way of
asking, in principle,

00:03:51.020 --> 00:03:53.510
in a network where you can
actually control everything

00:03:53.510 --> 00:03:57.110
about its architecture and
about the stimuli it sees,

00:03:57.110 --> 00:03:59.330
what are the necessary
conditions for it to produce

00:03:59.330 --> 00:04:01.880
something like face patches?

00:04:01.880 --> 00:04:03.950
What do you have to
train it on to get

00:04:03.950 --> 00:04:07.790
it to produce face patches, and
to be able to recognize faces?

00:04:07.790 --> 00:04:11.420
And at the top level, why,
computationally, does it

00:04:11.420 --> 00:04:13.980
make sense to have face
patches in the first place?

00:04:13.980 --> 00:04:15.530
This is kind of the
biggest question

00:04:15.530 --> 00:04:17.447
lurking in the background
of this whole field.

00:04:17.447 --> 00:04:20.060
I'm describing all of these
specialized mechanisms

00:04:20.060 --> 00:04:22.310
in mind and brain, but
really, wouldn't it

00:04:22.310 --> 00:04:25.100
be nice to know why our minds
and brains are organized

00:04:25.100 --> 00:04:27.680
that way, rather than
just that they are?

00:04:27.680 --> 00:04:29.100
And that's a really
hard question,

00:04:29.100 --> 00:04:30.710
and I think there's
a real hope now

00:04:30.710 --> 00:04:34.220
that computational modeling
may get us toward an answer

00:04:34.220 --> 00:04:38.750
sometime in the next decade,
maybe even the next few years.

00:04:38.750 --> 00:04:40.880
OK, so that's the overview.

00:04:40.880 --> 00:04:45.140
I now want to go on quite a
discussion about this notion

00:04:45.140 --> 00:04:48.530
that preexisting connectivity
may be a major constraint

00:04:48.530 --> 00:04:50.820
in wiring up the brain.

00:04:50.820 --> 00:04:52.910
So first, we need
to talk about, how

00:04:52.910 --> 00:04:55.760
would you look at structural
connectivity in human brains?

00:04:55.760 --> 00:04:57.560
And I haven't really
talked about this yet.

00:04:57.560 --> 00:05:00.320
The main method for
being able to look--

00:05:00.320 --> 00:05:03.650
for being able to get some
sense of this in human brains

00:05:03.650 --> 00:05:06.230
is to use another
kind of MRI imaging.

00:05:06.230 --> 00:05:08.660
Uses the same machine
that's an MRI machine,

00:05:08.660 --> 00:05:11.830
but it's going to produce
anatomical images that show us

00:05:11.830 --> 00:05:14.620
not those nice pretty pictures
of brains that you're used to,

00:05:14.620 --> 00:05:19.090
but that show us the
direction of water diffusion.

00:05:19.090 --> 00:05:21.520
And so the principle
is pretty simple.

00:05:21.520 --> 00:05:26.140
Here is a picture
of an optic tract.

00:05:26.140 --> 00:05:28.930
And what it's showing
you is that if you see,

00:05:28.930 --> 00:05:31.240
an optic tract is a
whole bunch of axons

00:05:31.240 --> 00:05:35.950
oriented like this connecting
retinal ganglion cells to what?

00:05:38.560 --> 00:05:40.420
Where do the retinal
ganglion cell axons

00:05:40.420 --> 00:05:42.880
land going through
the optic tract?

00:05:42.880 --> 00:05:43.420
[INAUDIBLE]

00:05:43.420 --> 00:05:44.650
AUDIENCE: LOG?

00:05:44.650 --> 00:05:45.760
NANCY KANWISHER: LGN.

00:05:45.760 --> 00:05:46.330
LGN.

00:05:46.330 --> 00:05:48.670
Lateral geniculate
nucleus of the thalamus.

00:05:48.670 --> 00:05:50.230
So there's that fiber bundle.

00:05:50.230 --> 00:05:52.900
But the main point
for now is that you

00:05:52.900 --> 00:05:55.240
can see that each
of those fibers

00:05:55.240 --> 00:05:58.480
has a layer of fat around it,
and the upshot of all of that

00:05:58.480 --> 00:06:00.580
is that water likes
to diffuse more

00:06:00.580 --> 00:06:03.400
in this direction
than that direction.

00:06:03.400 --> 00:06:05.800
That's the key idea
of diffusion imaging.

00:06:05.800 --> 00:06:09.730
It tells you which direction
water is diffusing most.

00:06:09.730 --> 00:06:13.690
Water is constrained by the
fat layers around those axons,

00:06:13.690 --> 00:06:14.650
that myelin.

00:06:14.650 --> 00:06:16.690
And so you get diffusion
more in this direction

00:06:16.690 --> 00:06:18.800
than orthogonally to it.

00:06:18.800 --> 00:06:22.690
And so the details
of the physics

00:06:22.690 --> 00:06:25.810
of this kind of imaging, which
I'm totally not explaining,

00:06:25.810 --> 00:06:28.930
are such that what
you get out is

00:06:28.930 --> 00:06:31.750
a picture at each point
in the brain of what

00:06:31.750 --> 00:06:35.837
is the direction of maximum
diffusion at that point.

00:06:35.837 --> 00:06:37.420
And so here's a
little picture of lots

00:06:37.420 --> 00:06:39.950
of little vectors
saying, at this point,

00:06:39.950 --> 00:06:41.920
water wants to diffuse
this way, or this way,

00:06:41.920 --> 00:06:43.300
or this way, or this way.

00:06:43.300 --> 00:06:44.600
Everybody with me so far?

00:06:44.600 --> 00:06:46.660
So you get a whole bunch
of little teeny vectors

00:06:46.660 --> 00:06:49.960
all through the brain showing
you the orientation where water

00:06:49.960 --> 00:06:51.880
wants to diffuse at that point.

00:06:51.880 --> 00:06:54.550
And the idea is
that's telling us

00:06:54.550 --> 00:06:57.610
which way fibers are
going at that point.

00:06:57.610 --> 00:06:59.590
And we can therefore infer--

00:06:59.590 --> 00:07:03.850
we can follow these things using
a method called tractography,

00:07:03.850 --> 00:07:05.650
where we just follow
those little vectors

00:07:05.650 --> 00:07:06.490
through the brain.

00:07:06.490 --> 00:07:07.823
And that's what's happened here.

00:07:07.823 --> 00:07:10.150
At each point in the brain,
you start at one point,

00:07:10.150 --> 00:07:14.050
and you just follow these
vectors and see where they go.

00:07:14.050 --> 00:07:16.833
Does that make sense,
sort of intuitively?

00:07:16.833 --> 00:07:18.250
I'm skipping over
lots of details,

00:07:18.250 --> 00:07:20.470
but I want you to get the gist.

00:07:20.470 --> 00:07:24.910
OK, so these beautiful pictures
that you may have seen before

00:07:24.910 --> 00:07:26.800
are diffusion tractography.

00:07:26.800 --> 00:07:32.140
They show you our best guess
of the long-range connections

00:07:32.140 --> 00:07:34.600
between one part of
the brain and another

00:07:34.600 --> 00:07:37.690
based on diffusion tractography.

00:07:37.690 --> 00:07:39.610
And on the theory
that you should

00:07:39.610 --> 00:07:42.800
wear your data
whenever possible,

00:07:42.800 --> 00:07:44.110
here's mine from my lab.

00:07:44.110 --> 00:07:46.610
Whoops, I'm tangling it here.

00:07:46.610 --> 00:07:49.480
So-- I love these things,
they're so beautiful.

00:07:49.480 --> 00:07:52.562
One of my post-docs who's
our tractography whiz

00:07:52.562 --> 00:07:53.770
gave me this beautiful scarf.

00:07:53.770 --> 00:07:54.700
Isn't this nice?

00:07:54.700 --> 00:07:57.040
And so you can see
even more clearly here

00:07:57.040 --> 00:08:00.370
that this is a cross-section
through the brain in this axis

00:08:00.370 --> 00:08:01.460
right here.

00:08:01.460 --> 00:08:05.950
And so these big green
guys are the connections

00:08:05.950 --> 00:08:08.583
that go from the back of the
head down the temporal lobe,

00:08:08.583 --> 00:08:10.000
down the visual
pathway that we've

00:08:10.000 --> 00:08:12.130
been talking about all along.

00:08:12.130 --> 00:08:13.850
OK, that was gratuitous.

00:08:13.850 --> 00:08:16.210
I just thought it was fun.

00:08:16.210 --> 00:08:18.280
OK, so tractography is cool.

00:08:18.280 --> 00:08:20.440
It makes gorgeous pictures
and gorgeous scarves.

00:08:23.800 --> 00:08:26.798
And it works really well to
discover big fiber bundles.

00:08:26.798 --> 00:08:28.840
There are lots of parts
of the brain I showed you

00:08:28.840 --> 00:08:30.550
with that gross
dissection picture

00:08:30.550 --> 00:08:32.770
last time, that there are
big chunks of white matter

00:08:32.770 --> 00:08:35.500
where lots and lots of
parallel fibers go like this.

00:08:35.500 --> 00:08:38.140
And tractography works
well to find those.

00:08:38.140 --> 00:08:41.840
You can really see those very
nicely with diffusion imaging.

00:08:41.840 --> 00:08:47.500
However, it's not so hot for
discovering finer connections.

00:08:47.500 --> 00:08:49.630
It's better than
nothing, but there's

00:08:49.630 --> 00:08:52.370
a lot of ways in which it fails.

00:08:52.370 --> 00:08:54.440
So for example, if
you have water--

00:08:54.440 --> 00:08:56.260
if you have fibers
crossing in some part

00:08:56.260 --> 00:08:58.840
of the brain like this,
you'll get diffusion

00:08:58.840 --> 00:09:00.700
in this direction
and this direction,

00:09:00.700 --> 00:09:02.737
and the tractography
algorithm will be finished.

00:09:02.737 --> 00:09:04.570
It won't know whether
to keep going straight

00:09:04.570 --> 00:09:06.460
or whether to turn.

00:09:06.460 --> 00:09:08.080
So that's just one
of many reasons why

00:09:08.080 --> 00:09:11.170
diffusion tractography
is lovely, and wonderful,

00:09:11.170 --> 00:09:16.340
and the best we have in in-vivo
brains, but it's not so great.

00:09:16.340 --> 00:09:19.780
Anyway, it's all we
have, so we use it.

00:09:19.780 --> 00:09:20.530
OK.

00:09:20.530 --> 00:09:25.510
So we can use tractography
to ask, for example,

00:09:25.510 --> 00:09:28.420
is the long-range connectivity
of the fusiform face

00:09:28.420 --> 00:09:31.690
area distinct from the
long-range connectivity

00:09:31.690 --> 00:09:33.340
of its neighbors?

00:09:33.340 --> 00:09:36.760
In other words, on this idea
that that patch of cortex

00:09:36.760 --> 00:09:40.390
gets wired up to be a
face area, somehow because

00:09:40.390 --> 00:09:44.230
of the connectivity to
and from that region

00:09:44.230 --> 00:09:47.327
to other parts of
the brain, then we

00:09:47.327 --> 00:09:48.910
should predict that
that region should

00:09:48.910 --> 00:09:51.700
have different connectivity
than neighboring cortex.

00:09:51.700 --> 00:09:54.280
Otherwise, connectivity
isn't enough of a signature

00:09:54.280 --> 00:09:56.680
to tell us where
to put a face area.

00:09:56.680 --> 00:09:57.850
I'm seeing blank looks.

00:09:57.850 --> 00:09:59.050
Is this not making sense?

00:09:59.050 --> 00:10:00.190
OK.

00:10:00.190 --> 00:10:03.490
Just butt in and ask questions
if I'm not making sense.

00:10:03.490 --> 00:10:04.180
OK.

00:10:04.180 --> 00:10:08.260
So question is, do these
connectivity fingerprints

00:10:08.260 --> 00:10:11.680
predict the location of
functional regions, first

00:10:11.680 --> 00:10:12.830
in adults?

00:10:12.830 --> 00:10:15.270
If we don't see it in
adults, then the jig's up.

00:10:15.270 --> 00:10:17.080
So let's start with adults.

00:10:17.080 --> 00:10:17.740
OK.

00:10:17.740 --> 00:10:21.490
So the way that you can
do this is, for each voxel

00:10:21.490 --> 00:10:22.310
in the brain--

00:10:22.310 --> 00:10:23.880
this is a big one,
so you can see it.

00:10:23.880 --> 00:10:25.630
It would actually be
a couple millimeters,

00:10:25.630 --> 00:10:27.100
wouldn't show on this picture.

00:10:27.100 --> 00:10:30.650
What you do is you follow
that tractography and you say,

00:10:30.650 --> 00:10:35.590
oh, look, it went there, and it
goes there, and it goes there.

00:10:35.590 --> 00:10:39.170
And you tally how often,
when you start here,

00:10:39.170 --> 00:10:43.340
you land in each of a bunch of
different big anatomical chunks

00:10:43.340 --> 00:10:44.810
of brain.

00:10:44.810 --> 00:10:48.050
That gives you a description
of the connectivity fingerprint

00:10:48.050 --> 00:10:49.220
of that voxel.

00:10:49.220 --> 00:10:51.230
How strong is its
connection to each

00:10:51.230 --> 00:10:54.590
of these other remote
regions in the brain?

00:10:54.590 --> 00:10:57.410
That's what I mean by a
connectivity fingerprint.

00:10:57.410 --> 00:11:00.440
So now the question
is, can you use

00:11:00.440 --> 00:11:05.750
this connectivity
fingerprint to predict what

00:11:05.750 --> 00:11:07.631
the function of that voxel is?

00:11:07.631 --> 00:11:10.700
That is, is the connectivity
distinctive enough that,

00:11:10.700 --> 00:11:13.370
just based on diffusion
data, we could

00:11:13.370 --> 00:11:15.800
say, what does that voxel do?

00:11:15.800 --> 00:11:19.550
If the fusiform face area has a
whole distinctive connectivity

00:11:19.550 --> 00:11:22.020
fingerprint, then we should
be able to predict it.

00:11:22.020 --> 00:11:23.550
Does this make sense?

00:11:23.550 --> 00:11:24.950
OK, so that's the question.

00:11:24.950 --> 00:11:26.900
And there's a lot of
math, which I'll skip.

00:11:26.900 --> 00:11:28.430
I'll just give you the gist.

00:11:28.430 --> 00:11:30.013
So what we're trying
to figure out is,

00:11:30.013 --> 00:11:32.450
is the fusiform face area
distinct from its neighbors

00:11:32.450 --> 00:11:33.920
in its long-range connectivity?

00:11:33.920 --> 00:11:36.290
That's the question.

00:11:36.290 --> 00:11:37.850
And, in fact, it is.

00:11:37.850 --> 00:11:39.140
And we can show that.

00:11:39.140 --> 00:11:40.790
Again, I'm skipping
over some details,

00:11:40.790 --> 00:11:44.840
but here is a recently-published
paper that shows you

00:11:44.840 --> 00:11:46.550
in ways that should
be familiar now,

00:11:46.550 --> 00:11:49.310
this is functional
MRI activation

00:11:49.310 --> 00:11:50.990
for faces versus objects.

00:11:50.990 --> 00:11:52.940
Fusiform face area,
that's probably

00:11:52.940 --> 00:11:55.910
occipital face area, another
region we'll talk about later.

00:11:55.910 --> 00:11:56.840
The face patches.

00:11:56.840 --> 00:11:58.040
The usual face patches.

00:11:58.040 --> 00:11:59.863
Again, this is an
inflated brain,

00:11:59.863 --> 00:12:01.280
so the dark bits
are the bits that

00:12:01.280 --> 00:12:03.110
used to be folded
up inside the sulcus

00:12:03.110 --> 00:12:06.020
until they were
mathematically inflated.

00:12:06.020 --> 00:12:08.510
So that's the standard
thing we've been looking at.

00:12:08.510 --> 00:12:12.980
This is the prediction based
on diffusion tractography

00:12:12.980 --> 00:12:16.820
alone in the same subject about
where the face patches should

00:12:16.820 --> 00:12:17.750
be.

00:12:17.750 --> 00:12:21.620
So very roughly, what you do is
you take some other subjects,

00:12:21.620 --> 00:12:24.230
and you train them up on
connectivity fingerprints--

00:12:24.230 --> 00:12:28.520
it's kind of like NVPA, but
you train from diffusion data,

00:12:28.520 --> 00:12:31.340
and you try to predict
face selectivity.

00:12:31.340 --> 00:12:34.340
And then you take the diffusion
data from a new subject,

00:12:34.340 --> 00:12:37.580
and you predict where that
face selectivity should be,

00:12:37.580 --> 00:12:40.220
and there's where it's
predicted for the same subject,

00:12:40.220 --> 00:12:42.380
and it's pretty damn good.

00:12:42.380 --> 00:12:44.943
Did everybody get the gist
of what I just went through?

00:12:44.943 --> 00:12:46.610
You don't need to
remember every detail.

00:12:46.610 --> 00:12:50.300
The key idea is, is there
a systematic relationship

00:12:50.300 --> 00:12:52.640
between long-range
connectivity of a voxel

00:12:52.640 --> 00:12:55.010
and its function,
its selectivity?

00:12:55.010 --> 00:12:58.340
And this says yes for faces.

00:12:58.340 --> 00:12:58.940
OK?

00:12:58.940 --> 00:13:00.470
So that's the case for faces.

00:13:00.470 --> 00:13:02.870
That tells us that in
adults, those face regions

00:13:02.870 --> 00:13:05.512
have distinct connectivity.

00:13:05.512 --> 00:13:06.470
This is the same thing.

00:13:06.470 --> 00:13:08.450
I just shrunk it so I
could fit in other stuff.

00:13:08.450 --> 00:13:11.390
Here is doing the
same thing for scenes.

00:13:11.390 --> 00:13:16.220
Functional selectivity PPA
RSC, functional selectivity

00:13:16.220 --> 00:13:18.680
for scenes measured
with functional MRI,

00:13:18.680 --> 00:13:22.250
predicted functional pattern
from the same subject

00:13:22.250 --> 00:13:25.190
with just tractography alone.

00:13:25.190 --> 00:13:27.322
OK?

00:13:27.322 --> 00:13:28.280
Do you have a question?

00:13:28.280 --> 00:13:29.096
AUDIENCE: Oh, no.

00:13:29.096 --> 00:13:29.929
[INTERPOSING VOICES]

00:13:29.929 --> 00:13:31.763
NANCY KANWISHER: It's
pretty good, isn't it?

00:13:31.763 --> 00:13:32.300
Yeah, yeah.

00:13:32.300 --> 00:13:33.500
No, I was dissing diffusion.

00:13:33.500 --> 00:13:36.360
You might be thinking, OK, I was
dissing diffusion tractography.

00:13:36.360 --> 00:13:36.860
It sucks.

00:13:36.860 --> 00:13:37.943
It has all these problems.

00:13:37.943 --> 00:13:39.330
It has all these ambiguities.

00:13:39.330 --> 00:13:41.090
So how could it work so well?

00:13:41.090 --> 00:13:42.060
That's a good question.

00:13:42.060 --> 00:13:43.260
I don't know the answer to that.

00:13:43.260 --> 00:13:44.760
I think in part,
it's because you're

00:13:44.760 --> 00:13:47.610
predicting based on all of
these different connections.

00:13:47.610 --> 00:13:49.760
So even if half
of them are wrong,

00:13:49.760 --> 00:13:52.110
you can still get some
predictive power out of it.

00:13:52.110 --> 00:13:53.860
That's just my guess.

00:13:53.860 --> 00:13:54.670
OK?

00:13:54.670 --> 00:13:56.500
OK, so it works pretty
well for scenes,

00:13:56.500 --> 00:13:58.960
and it works pretty well for
body selectivity as well.

00:13:58.960 --> 00:14:03.790
Functional MRI prediction
from connectivity.

00:14:03.790 --> 00:14:05.470
So that's cool.

00:14:05.470 --> 00:14:08.080
So that says, these all
have distinct connectivity

00:14:08.080 --> 00:14:11.890
fingerprints, but now this
is all done in adults.

00:14:11.890 --> 00:14:17.290
And remember, the way we got
into this long shaggy dog story

00:14:17.290 --> 00:14:20.740
is to ask what these long-range
connections, what role they

00:14:20.740 --> 00:14:22.450
might play in development.

00:14:22.450 --> 00:14:24.880
Remember that I said
last time that most

00:14:24.880 --> 00:14:28.940
of the long-range connections of
the brain are present at birth.

00:14:28.940 --> 00:14:32.950
So that suggests that maybe
these connections are also

00:14:32.950 --> 00:14:33.820
there at birth.

00:14:36.460 --> 00:14:38.980
And it suggests that maybe
indeed those connections could

00:14:38.980 --> 00:14:40.150
play a role in development.

00:14:40.150 --> 00:14:41.485
At least they're probably there.

00:14:41.485 --> 00:14:45.250
They're in a position
to play that role,

00:14:45.250 --> 00:14:48.110
if that's actually what happens.

00:14:48.110 --> 00:14:54.815
So all of this brings us to
the case of rewired ferrets.

00:14:54.815 --> 00:14:55.315
What?

00:14:55.315 --> 00:14:56.350
What am I talking about?

00:14:56.350 --> 00:14:57.970
They're cute, aren't they?

00:14:57.970 --> 00:15:00.220
They're also very good
experimental animals

00:15:00.220 --> 00:15:02.600
to address just this question.

00:15:02.600 --> 00:15:04.260
And Mriganka Sur
in this department

00:15:04.260 --> 00:15:08.290
did this very important
paper a while back

00:15:08.290 --> 00:15:12.792
where he asked whether
connectivity instructs

00:15:12.792 --> 00:15:13.750
functional development.

00:15:13.750 --> 00:15:17.350
That is, whether the
connectivity present at birth

00:15:17.350 --> 00:15:21.130
is sufficient to determine the
function of the region that

00:15:21.130 --> 00:15:22.450
has those connections.

00:15:22.450 --> 00:15:26.350
And he did this by
manipulating connectivity.

00:15:26.350 --> 00:15:28.450
So if you want to ask,
what is the causal role

00:15:28.450 --> 00:15:30.610
of x, you have to manipulate x.

00:15:30.610 --> 00:15:32.740
So we've talked a lot
about this in this class.

00:15:32.740 --> 00:15:34.060
Functional MRI, wonderful.

00:15:34.060 --> 00:15:34.810
You see activity.

00:15:34.810 --> 00:15:38.150
You have no idea what its causal
role is until you mess with it.

00:15:38.150 --> 00:15:40.690
For example, by electrically
stimulating the brain.

00:15:40.690 --> 00:15:44.260
Similarly, connectivity
may be present at birth,

00:15:44.260 --> 00:15:47.118
and may predict where we may
be able to use it to predict

00:15:47.118 --> 00:15:48.160
where the functions land.

00:15:48.160 --> 00:15:50.290
It doesn't tell us that
it's playing a causal role.

00:15:50.290 --> 00:15:52.390
The way to find out if
it's playing a causal role

00:15:52.390 --> 00:15:54.740
is to change it and
see what happens.

00:15:54.740 --> 00:15:57.640
And that's what Mriganka
Sur and his colleagues did.

00:15:57.640 --> 00:16:02.350
So they used ferrets because
they're born very prematurely.

00:16:02.350 --> 00:16:06.010
And so what that means is
that you can operate on them

00:16:06.010 --> 00:16:09.460
surgically right at
birth before they

00:16:09.460 --> 00:16:10.672
have any visual experience.

00:16:10.672 --> 00:16:12.130
They haven't opened
their eyes yet.

00:16:12.130 --> 00:16:14.260
And you can--
turns out-- reroute

00:16:14.260 --> 00:16:15.850
some of the connectivity.

00:16:15.850 --> 00:16:20.050
OK, so this is a diagram of some
bits that should be familiar.

00:16:20.050 --> 00:16:22.870
The retina going to the lateral
geniculate nucleus and then

00:16:22.870 --> 00:16:23.920
up to V1.

00:16:23.920 --> 00:16:26.350
Also true in ferrets.

00:16:26.350 --> 00:16:29.530
In addition, we have primary
auditory cortex that we'll

00:16:29.530 --> 00:16:31.600
talk more about in a few weeks.

00:16:31.600 --> 00:16:33.820
So just like V1,
but for hearing.

00:16:33.820 --> 00:16:34.704
A1.

00:16:34.704 --> 00:16:38.510
A1 also goes to another
nucleus in the thalamus.

00:16:38.510 --> 00:16:40.840
This one called the
medial geniculate nucleus.

00:16:40.840 --> 00:16:43.420
And then it goes from there up
through a complicated chain,

00:16:43.420 --> 00:16:46.300
eventually-- oh, sorry,
it goes this way.

00:16:46.300 --> 00:16:49.090
Thalamus up to A1.

00:16:49.090 --> 00:16:54.010
So that's the basic
wiring of an adult ferret.

00:16:54.010 --> 00:16:56.830
And so what Sur
and his colleagues

00:16:56.830 --> 00:16:58.660
figured out how
to do is redirect

00:16:58.660 --> 00:17:02.420
some of those connections
by surgery at birth.

00:17:02.420 --> 00:17:05.619
So this is a wiring diagram
of the same thing shown here.

00:17:05.619 --> 00:17:08.050
Retina, LGN.

00:17:08.050 --> 00:17:10.359
This is V1, it's also called 17.

00:17:10.359 --> 00:17:13.690
And here is medial geniculate
and auditory cortex.

00:17:13.690 --> 00:17:19.720
And so what they did was
to surgically knock out

00:17:19.720 --> 00:17:27.430
a few of these connections here
in the just-born ferret pups.

00:17:27.430 --> 00:17:30.850
And what happens is if you
knock out this connection

00:17:30.850 --> 00:17:34.900
here, the fibers that start
this way get rerouted,

00:17:34.900 --> 00:17:38.920
and you end up with a ferret
that's wired up like this.

00:17:38.920 --> 00:17:42.790
The important part of this
is this rewired ferret

00:17:42.790 --> 00:17:47.920
has a connection between their
retina and medial geniculate

00:17:47.920 --> 00:17:50.690
nucleus that goes to
primary auditory cortex.

00:17:50.690 --> 00:17:52.960
So we're taking visual
input at the periphery

00:17:52.960 --> 00:17:56.770
and wiring it up into
the auditory system.

00:17:56.770 --> 00:18:00.760
And the point of
all of this is now,

00:18:00.760 --> 00:18:03.700
primary auditory cortex
in this developing ferret

00:18:03.700 --> 00:18:06.320
will be getting visual input.

00:18:06.320 --> 00:18:10.420
And so if the input were
sufficient to determine

00:18:10.420 --> 00:18:13.810
the function of that
region of cortex, then

00:18:13.810 --> 00:18:15.850
what should we find in
these rewired ferrets?

00:18:15.850 --> 00:18:20.300
What should happen in what
would have been primary auditory

00:18:20.300 --> 00:18:20.800
cortex?

00:18:20.800 --> 00:18:22.660
What should it do?

00:18:22.660 --> 00:18:23.661
Christine.

00:18:27.098 --> 00:18:29.062
AUDIENCE: [INAUDIBLE] visual--

00:18:29.062 --> 00:18:30.130
NANCY KANWISHER: Yeah!

00:18:30.130 --> 00:18:32.800
It should behave like
visual cortex, absolutely.

00:18:32.800 --> 00:18:34.690
If everything's
determined by the inputs,

00:18:34.690 --> 00:18:38.440
we change the inputs, it should
behave like visual cortex.

00:18:38.440 --> 00:18:41.020
Well, that would be
freaking crazy, wouldn't it?

00:18:41.020 --> 00:18:42.760
I mean, it's miles
away in the brain.

00:18:42.760 --> 00:18:44.980
It's a totally
different part of brain.

00:18:44.980 --> 00:18:47.870
That will be nuts.

00:18:47.870 --> 00:18:49.500
But that's what happens.

00:18:49.500 --> 00:18:50.410
It's pretty amazing.

00:18:50.410 --> 00:18:52.270
This is a really
important study.

00:18:52.270 --> 00:18:53.820
OK.

00:18:53.820 --> 00:18:56.100
All right.

00:18:56.100 --> 00:18:58.950
So what you find,
first of all, is

00:18:58.950 --> 00:19:02.010
that primary auditory cortex
in the rewired ferrets

00:19:02.010 --> 00:19:04.500
responds to visual input.

00:19:04.500 --> 00:19:05.040
That's cool.

00:19:05.040 --> 00:19:07.340
But you might say, OK, you
wired visual input in there.

00:19:07.340 --> 00:19:09.340
Of course it's going to
respond to visual input.

00:19:09.340 --> 00:19:12.540
So maybe that's not too
cool, but not too surprising.

00:19:12.540 --> 00:19:15.990
But the next part is really
cool and really surprising.

00:19:15.990 --> 00:19:22.020
Remember how I said that
in normal visual cortex--

00:19:22.020 --> 00:19:24.390
in humans and monkeys,
and also ferrets--

00:19:24.390 --> 00:19:26.770
you get these
orientation columns.

00:19:26.770 --> 00:19:28.320
Now, remember, these are--

00:19:28.320 --> 00:19:32.610
what this shows is that as you
move across the cortex in V1--

00:19:32.610 --> 00:19:35.640
we're now talking visual
cortex here-- in visual cortex,

00:19:35.640 --> 00:19:39.600
in normal mammals, you get
this smooth progression

00:19:39.600 --> 00:19:42.202
of orientation selectivity as
you move across the cortex.

00:19:42.202 --> 00:19:43.410
And that's what's shown here.

00:19:43.410 --> 00:19:44.670
Everybody with the program?

00:19:44.670 --> 00:19:45.420
OK.

00:19:45.420 --> 00:19:50.400
So that's normal primary visual
cortex in an adult animal.

00:19:50.400 --> 00:19:53.520
What do you think primary
auditory cortex looks

00:19:53.520 --> 00:19:55.770
like in the rewired ferrets?

00:19:55.770 --> 00:19:57.570
Damn similar.

00:19:57.570 --> 00:20:01.200
So not only do you
get visual responses

00:20:01.200 --> 00:20:04.380
in what would have been
auditory cortex when you rewire,

00:20:04.380 --> 00:20:06.000
you get orientation columns.

00:20:06.000 --> 00:20:08.550
You get this really
fine-grained structure

00:20:08.550 --> 00:20:10.500
of what everybody
thought this was

00:20:10.500 --> 00:20:12.240
something about visual cortex.

00:20:12.240 --> 00:20:14.190
Well, this says
that visual input

00:20:14.190 --> 00:20:17.220
is sufficient to produce
orientation columns

00:20:17.220 --> 00:20:19.110
in a part of cortex
that otherwise never

00:20:19.110 --> 00:20:21.430
would have had them.

00:20:21.430 --> 00:20:24.090
Does everybody see how
mind-blowing this is?

00:20:24.090 --> 00:20:25.140
OK.

00:20:25.140 --> 00:20:27.030
So that's pretty
cool, but now we

00:20:27.030 --> 00:20:29.370
get to the really cool question.

00:20:29.370 --> 00:20:34.290
When these neurons are
active, does the ferret see,

00:20:34.290 --> 00:20:37.380
or do they hear?

00:20:37.380 --> 00:20:38.220
OK.

00:20:38.220 --> 00:20:39.150
It's rewired.

00:20:39.150 --> 00:20:40.950
It's getting input
from the retina,

00:20:40.950 --> 00:20:42.480
but there's neurons
in what would

00:20:42.480 --> 00:20:44.970
have been primary
auditory cortex now

00:20:44.970 --> 00:20:46.930
responding to visual input.

00:20:46.930 --> 00:20:49.780
What does the ferret
think is going on?

00:20:49.780 --> 00:20:52.560
Does he say, oh, that's
sight, because he's

00:20:52.560 --> 00:20:55.080
learned that visual
input means that's sight?

00:20:55.080 --> 00:20:57.750
Or does he say,
I hear something,

00:20:57.750 --> 00:21:01.470
because that's auditory cortex.

00:21:01.470 --> 00:21:04.530
Everybody in the grip of
what a cool question that is?

00:21:04.530 --> 00:21:05.040
OK.

00:21:05.040 --> 00:21:07.080
And so it could go either way.

00:21:07.080 --> 00:21:08.880
There's really no way
to tell in advance.

00:21:08.880 --> 00:21:12.150
It depends on how you
read out the information

00:21:12.150 --> 00:21:13.410
in that piece of cortex.

00:21:13.410 --> 00:21:17.100
When we do NVPA,
we sit god-like by,

00:21:17.100 --> 00:21:19.982
and we look at a patch of brain,
and we decode what's in there.

00:21:19.982 --> 00:21:21.690
But really, what's
happening in the brain

00:21:21.690 --> 00:21:24.450
is some other part of the brain
is getting input, and decoding,

00:21:24.450 --> 00:21:25.657
and interpreting it.

00:21:25.657 --> 00:21:27.990
And so the question is, what
do later parts of the brain

00:21:27.990 --> 00:21:29.910
make of this?

00:21:29.910 --> 00:21:31.380
And the answer is
the later parts

00:21:31.380 --> 00:21:34.080
of the brain learn that
that's visual information,

00:21:34.080 --> 00:21:38.383
and the ferret reports
seeing stuff, not hearing it.

00:21:38.383 --> 00:21:40.800
Now, you may be thinking, how
the hell do you ask a ferret

00:21:40.800 --> 00:21:43.050
if he's seeing or hearing?

00:21:43.050 --> 00:21:45.780
What you do is you
use non-rewired parts

00:21:45.780 --> 00:21:47.625
of the same ferret's brain.

00:21:47.625 --> 00:21:49.500
Actually, forget if it's
the other hemisphere

00:21:49.500 --> 00:21:51.510
or a different part
of the visual field

00:21:51.510 --> 00:21:53.610
that doesn't get rewired.

00:21:53.610 --> 00:21:57.597
So you have gold standard,
where normal vision is working,

00:21:57.597 --> 00:21:59.430
and normal hearing is
working in the ferret,

00:21:59.430 --> 00:22:01.890
and you train him, press
this button when you see

00:22:01.890 --> 00:22:04.380
and press this button when you
hear, and it's unambiguous.

00:22:04.380 --> 00:22:08.550
And then once he's trained,
you stimulate those A1 neurons

00:22:08.550 --> 00:22:12.690
and you ask him what's going on,
and he says he sees something.

00:22:12.690 --> 00:22:14.320
OK?

00:22:14.320 --> 00:22:18.520
All right, so this is
one of the true classics.

00:22:18.520 --> 00:22:20.170
OK.

00:22:20.170 --> 00:22:25.540
So this means that A1 in this
case, primary auditory cortex,

00:22:25.540 --> 00:22:28.090
is instructed by
its connectivity

00:22:28.090 --> 00:22:31.690
and by the experience that
comes through that connectivity

00:22:31.690 --> 00:22:33.880
to shape its function.

00:22:33.880 --> 00:22:35.140
Everybody got that?

00:22:35.140 --> 00:22:36.850
All right.

00:22:36.850 --> 00:22:39.190
So both experience
and connectivity

00:22:39.190 --> 00:22:44.150
can determine cortical
function, at least in ferrets.

00:22:44.150 --> 00:22:44.650
What?

00:22:44.650 --> 00:22:45.390
Yes, question.

00:22:45.390 --> 00:22:46.682
AUDIENCE: I have two questions.

00:22:46.682 --> 00:22:49.870
So first of all, what
does their V1 look

00:22:49.870 --> 00:22:52.960
like after this rewiring, and
also, can they hear things,

00:22:52.960 --> 00:22:54.033
and if so, where is it?

00:22:54.033 --> 00:22:55.450
NANCY KANWISHER:
Yeah, absolutely.

00:22:55.450 --> 00:22:59.420
OK, so if you look at the
diagram, there is additional--

00:22:59.420 --> 00:23:01.780
well, actually, it's
not in the diagram.

00:23:01.780 --> 00:23:06.680
But there is additional
input that's not shown here.

00:23:06.680 --> 00:23:09.730
So they can hear things through
maybe the other hemisphere,

00:23:09.730 --> 00:23:10.360
I forget.

00:23:10.360 --> 00:23:12.260
They can hear.

00:23:12.260 --> 00:23:15.548
And they can see,
because notice--

00:23:15.548 --> 00:23:16.090
that's right.

00:23:16.090 --> 00:23:18.910
OK, we blocked off
area 17, but these guys

00:23:18.910 --> 00:23:20.810
are higher-level visual areas.

00:23:20.810 --> 00:23:23.950
So they can see both through
their non-rewired hemisphere

00:23:23.950 --> 00:23:26.290
and through some other
bypassing connections

00:23:26.290 --> 00:23:28.810
to other parts of visual cortex.

00:23:28.810 --> 00:23:31.150
Probably, both of those
are going to be affected.

00:23:31.150 --> 00:23:34.300
Your vision is going to be
different if you bypass V1.

00:23:34.300 --> 00:23:37.090
But there will be at least
some visual information.

00:23:39.750 --> 00:23:41.820
OK, so that's ferrets.

00:23:41.820 --> 00:23:45.000
Again, animals, you
can do invasive studies

00:23:45.000 --> 00:23:47.010
and really do the
strong manipulation

00:23:47.010 --> 00:23:49.440
to do a strong test
of a causal role,

00:23:49.440 --> 00:23:51.540
and this is a classic example.

00:23:51.540 --> 00:23:53.430
Of course, we can't
rewire humans--

00:23:53.430 --> 00:23:56.010
or we could, but it
wouldn't be nice.

00:23:56.010 --> 00:23:58.950
But really, we want to know,
how does all that stuff get

00:23:58.950 --> 00:24:00.000
wired up?

00:24:00.000 --> 00:24:01.830
Are these regions also--

00:24:01.830 --> 00:24:04.740
is their function determined
by their connectivity

00:24:04.740 --> 00:24:10.980
present at birth, and due
to the experience of those

00:24:10.980 --> 00:24:12.370
regions have?

00:24:12.370 --> 00:24:12.870
OK.

00:24:12.870 --> 00:24:17.460
Well, we can't do controlled
rearing studies in humans.

00:24:17.460 --> 00:24:19.600
We can't rewire their brains.

00:24:19.600 --> 00:24:23.380
But we can be clever and smart
and think of other cases.

00:24:23.380 --> 00:24:25.800
So here's an
important test case.

00:24:25.800 --> 00:24:30.120
The important test case
is the case of reading.

00:24:30.120 --> 00:24:31.620
Why reading?

00:24:31.620 --> 00:24:34.800
Well, one, we all spend
a lot of time doing it.

00:24:34.800 --> 00:24:38.220
And two, humans have only been
reading for a few thousand

00:24:38.220 --> 00:24:39.690
years.

00:24:39.690 --> 00:24:42.300
And that's not long enough
for natural selection

00:24:42.300 --> 00:24:45.000
to have crafted an
innately-specified circuit just

00:24:45.000 --> 00:24:46.560
for reading.

00:24:46.560 --> 00:24:50.460
So that means that if we did
find a patch of cortex that

00:24:50.460 --> 00:24:53.550
responds selectively to
visually-presented words,

00:24:53.550 --> 00:24:58.020
or letters, that would suggest
that for that case at least,

00:24:58.020 --> 00:25:00.570
experience was
sufficient to wire up,

00:25:00.570 --> 00:25:04.090
to determine the function
of that region of cortex.

00:25:04.090 --> 00:25:05.340
This is all very hypothetical.

00:25:05.340 --> 00:25:07.020
Everybody got the idea?

00:25:07.020 --> 00:25:07.740
OK.

00:25:07.740 --> 00:25:10.980
Now, notice, this does not
apply to hearing words.

00:25:10.980 --> 00:25:13.380
People have been hearing words
for hundreds of thousands

00:25:13.380 --> 00:25:14.880
of years, perhaps millions.

00:25:14.880 --> 00:25:18.450
And so that's plenty of time
for special purpose circuitry,

00:25:18.450 --> 00:25:21.780
and that special
purpose circuitry exists

00:25:21.780 --> 00:25:24.210
and we'll talk about
it in a month or so.

00:25:24.210 --> 00:25:26.550
But now we're talking about
the case of visual word

00:25:26.550 --> 00:25:30.870
recognition-- this recent
cultural invention of humans.

00:25:30.870 --> 00:25:33.120
So that's why it's a special
case, because we know

00:25:33.120 --> 00:25:35.370
that's too recent to be innate.

00:25:35.370 --> 00:25:38.820
And so if we find a
selectivity, it can't be innate.

00:25:38.820 --> 00:25:40.080
All right?

00:25:40.080 --> 00:25:42.900
So that's what I just said.

00:25:42.900 --> 00:25:45.250
So do we have such a thing?

00:25:45.250 --> 00:25:46.890
Well, how would you test for it?

00:25:46.890 --> 00:25:48.940
What would you do?

00:25:48.940 --> 00:25:52.092
Joseph, what would you do?

00:25:52.092 --> 00:25:53.300
You want to know if there's--

00:26:00.050 --> 00:26:01.940
AUDIENCE: I guess I
would show them words,

00:26:01.940 --> 00:26:04.113
and then show them
not words, and see--

00:26:04.113 --> 00:26:05.030
NANCY KANWISHER: Yeah.

00:26:05.030 --> 00:26:07.220
It's not rocket science, guys.

00:26:07.220 --> 00:26:08.900
We just keep doing
the same damn thing.

00:26:08.900 --> 00:26:10.070
Exactly.

00:26:10.070 --> 00:26:11.570
Right.

00:26:11.570 --> 00:26:14.850
So start by--
here's what we did.

00:26:14.850 --> 00:26:17.420
We showed people
visually-presented words

00:26:17.420 --> 00:26:20.390
like that, and we showed them
line drawings of objects.

00:26:23.180 --> 00:26:27.110
And when we did that, we
found that in most subjects,

00:26:27.110 --> 00:26:29.450
there's a tiny little
patch of the bottom

00:26:29.450 --> 00:26:31.402
of their left hemisphere
right near the zones

00:26:31.402 --> 00:26:32.860
we've been talking
about, near face

00:26:32.860 --> 00:26:35.400
selective and other regions
on the bottom of the brain.

00:26:35.400 --> 00:26:39.800
But that tiny little patch
responds significantly more

00:26:39.800 --> 00:26:42.500
to words than pictures.

00:26:45.320 --> 00:26:47.780
Now, we won't do this
now, but you can do it

00:26:47.780 --> 00:26:50.090
as a thought experiment.

00:26:50.090 --> 00:26:52.880
What are the alternative
accounts of that activation?

00:26:52.880 --> 00:26:55.100
Has this shown that that
region is selectively

00:26:55.100 --> 00:26:56.720
involved in reading?

00:26:56.720 --> 00:26:58.290
Of course not.

00:26:58.290 --> 00:27:00.520
There's a million
differences between--

00:27:03.170 --> 00:27:06.320
oh, come on-- these and those.

00:27:06.320 --> 00:27:08.300
How bright they are,
how big they are.

00:27:08.300 --> 00:27:10.370
It's a million differences.

00:27:10.370 --> 00:27:11.940
And so to get
serious about it, we

00:27:11.940 --> 00:27:13.940
have to do the same game
that we've been playing

00:27:13.940 --> 00:27:15.030
all along in this course.

00:27:15.030 --> 00:27:16.430
This is like a
first whack at it.

00:27:16.430 --> 00:27:18.290
You find something, now
we have a candidate.

00:27:18.290 --> 00:27:19.910
But if we want to
get serious, we've

00:27:19.910 --> 00:27:21.577
got to test some other
conditions to see

00:27:21.577 --> 00:27:23.060
if that's really for real.

00:27:23.060 --> 00:27:24.290
OK?

00:27:24.290 --> 00:27:25.040
All right.

00:27:25.040 --> 00:27:27.870
So here's what we did in my lab
when we did this a while back.

00:27:27.870 --> 00:27:31.340
So first of all, this
is left-out data.

00:27:31.340 --> 00:27:33.350
Once you find that
region-- remember,

00:27:33.350 --> 00:27:35.840
if you're trying to characterize
the function of a region,

00:27:35.840 --> 00:27:38.060
I talked briefly
about this, a good way

00:27:38.060 --> 00:27:40.850
to do it is to run
a localizer to find

00:27:40.850 --> 00:27:42.290
that region in each subject.

00:27:42.290 --> 00:27:43.290
Now we found it.

00:27:43.290 --> 00:27:44.540
Now we have those voxels.

00:27:44.540 --> 00:27:46.763
Now we collect some
new data that may

00:27:46.763 --> 00:27:47.930
be a lot like our localizer.

00:27:47.930 --> 00:27:48.680
It doesn't matter.

00:27:48.680 --> 00:27:52.280
We collect some new data
and we look at the response.

00:27:52.280 --> 00:27:55.260
And that just puts us on
stronger statistical footing.

00:27:55.260 --> 00:27:55.760
OK.

00:27:55.760 --> 00:27:57.830
So here is time going this way.

00:27:57.830 --> 00:28:00.230
This is something called an
event-related design, where

00:28:00.230 --> 00:28:03.050
you just present a single
stimulus, and then wait,

00:28:03.050 --> 00:28:05.480
and another stimulus rather
than a whole bunch of them

00:28:05.480 --> 00:28:07.460
mushed together in a block.

00:28:07.460 --> 00:28:09.950
And then you average over
many, many repetitions.

00:28:09.950 --> 00:28:12.710
And so this is the response
over time-- it's seconds,

00:28:12.710 --> 00:28:14.570
it's really slow--

00:28:14.570 --> 00:28:17.832
to words and line
drawings in that region.

00:28:17.832 --> 00:28:20.040
So this is just replicating
what I showed you before.

00:28:20.040 --> 00:28:22.550
It's showing you what
the actual selectivity

00:28:22.550 --> 00:28:27.860
looks like in the real data,
not just in a significance map.

00:28:27.860 --> 00:28:31.590
Why is this thing taking
six seconds to respond?

00:28:31.590 --> 00:28:33.050
This is stimulus
onset out there.

00:28:38.680 --> 00:28:39.370
Yes.

00:28:39.370 --> 00:28:42.730
AUDIENCE: That's the
time between blood flow?

00:28:42.730 --> 00:28:43.660
NANCY KANWISHER: Yeah.

00:28:43.660 --> 00:28:45.250
Remember, the signal
we're looking at

00:28:45.250 --> 00:28:46.420
is based on blood flow.

00:28:46.420 --> 00:28:48.092
The neurons all
fired right here,

00:28:48.092 --> 00:28:50.300
but it takes a while to get
the blood flow to change.

00:28:50.300 --> 00:28:51.300
That's why it's delayed.

00:28:51.300 --> 00:28:52.510
Exactly.

00:28:52.510 --> 00:28:53.240
OK.

00:28:53.240 --> 00:28:53.740
All right.

00:28:53.740 --> 00:28:55.250
So what else are
we going to test?

00:28:55.250 --> 00:28:57.122
Well, you can do lots
of different things.

00:28:57.122 --> 00:28:58.330
We just tried lots of things.

00:28:58.330 --> 00:29:00.970
We said, OK, let's have
other things that are symbols

00:29:00.970 --> 00:29:02.450
but that our
subjects can't read.

00:29:02.450 --> 00:29:06.370
So we tried Chinese
characters, low response.

00:29:06.370 --> 00:29:07.840
We tried digit strings.

00:29:07.840 --> 00:29:09.040
Pretty low response.

00:29:09.040 --> 00:29:11.800
That's pretty remarkable,
because words and digit strings

00:29:11.800 --> 00:29:14.480
are pretty similar in how we use
them and what they look like.

00:29:14.480 --> 00:29:15.397
So that's pretty good.

00:29:17.800 --> 00:29:22.840
We tried consonant strings, like
this, that you can't pronounce.

00:29:22.840 --> 00:29:25.150
And we got the same response.

00:29:25.150 --> 00:29:26.330
And this is important.

00:29:26.330 --> 00:29:31.360
It tells us this region
is not a word region.

00:29:31.360 --> 00:29:34.550
Instead, it's something
about recognizing letters.

00:29:34.550 --> 00:29:37.240
But for the purposes of the
current argument, that's OK.

00:29:37.240 --> 00:29:41.470
It's still something that has
no basis in human evolution,

00:29:41.470 --> 00:29:44.530
and so if we find selectivity
for letters that are presumably

00:29:44.530 --> 00:29:46.660
used in the process
of reading, that

00:29:46.660 --> 00:29:49.380
must have come from experience.

00:29:49.380 --> 00:29:51.600
OK?

00:29:51.600 --> 00:29:52.500
What else did we do?

00:29:52.500 --> 00:29:54.810
OK, that's what I just said.

00:29:54.810 --> 00:29:58.530
Now, I submit that this
is a pretty good argument

00:29:58.530 --> 00:30:02.130
that that region must have
been wired up by experience.

00:30:02.130 --> 00:30:03.510
But you could niggle.

00:30:03.510 --> 00:30:08.250
You could say, well, there
are more straight edges

00:30:08.250 --> 00:30:09.810
with the words and consonants.

00:30:09.810 --> 00:30:11.760
The digits are
curvier, or whatever.

00:30:11.760 --> 00:30:13.950
You could make up
some story about how

00:30:13.950 --> 00:30:18.690
that isn't necessarily
selective for letters and words,

00:30:18.690 --> 00:30:21.540
and therefore, maybe
it's not necessarily

00:30:21.540 --> 00:30:23.550
wired up by experience.

00:30:23.550 --> 00:30:26.040
Further, who knows?

00:30:26.040 --> 00:30:28.890
Maybe everybody just has that
weird selectivity in there

00:30:28.890 --> 00:30:32.820
even if they never
learned to read.

00:30:32.820 --> 00:30:36.270
So it would really be nice
to make a stronger case.

00:30:36.270 --> 00:30:39.600
And what we did was we couldn't
find people in Cambridge

00:30:39.600 --> 00:30:43.200
who couldn't read, who didn't
have other things going on,

00:30:43.200 --> 00:30:48.360
but we could find people
who did read Hebrew.

00:30:48.360 --> 00:30:50.310
And we had-- where's
my Hebrew data?

00:30:50.310 --> 00:30:51.300
All right, hang on.

00:30:51.300 --> 00:30:52.440
OK, right.

00:30:52.440 --> 00:30:55.110
So here are our
non-Hebrew readers.

00:30:57.617 --> 00:30:58.200
This is funny.

00:30:58.200 --> 00:30:59.490
This is an old graph.

00:30:59.490 --> 00:31:02.430
It's not so impressive-looking.

00:31:02.430 --> 00:31:05.640
This is-- I forgot to
switch out our newer data.

00:31:05.640 --> 00:31:09.000
OK, so what we found is in
people who don't read Hebrew,

00:31:09.000 --> 00:31:11.642
the response was lower
to Hebrew than to words.

00:31:11.642 --> 00:31:13.350
Looks like it's almost
as high, actually.

00:31:13.350 --> 00:31:16.530
When we ran more subjects, it's
actually quite a bit lower.

00:31:16.530 --> 00:31:18.390
Nonetheless, when
we ran people who

00:31:18.390 --> 00:31:22.360
read both English and Hebrew,
the Hebrew response is higher.

00:31:22.360 --> 00:31:26.520
And that nails the case that
it's actually that individual's

00:31:26.520 --> 00:31:29.620
experience that determines the
selectivity of this region.

00:31:29.620 --> 00:31:31.500
It depends on what
orthographies you know.

00:31:31.500 --> 00:31:33.900
If you know how to read Hebrew,
you get a high response.

00:31:33.900 --> 00:31:35.940
If you don't, you
get a lower response.

00:31:35.940 --> 00:31:39.210
Everybody get that this
pretty much nails the case?

00:31:39.210 --> 00:31:41.160
OK, so where are we?

00:31:41.160 --> 00:31:44.040
All of this was
to say, do we ever

00:31:44.040 --> 00:31:49.243
see selectivity in the
brain that can't be innate?

00:31:49.243 --> 00:31:51.660
And I submit to you, this is
selectivity in the brain that

00:31:51.660 --> 00:31:54.300
can't be innate, that
has to be learned.

00:31:54.300 --> 00:31:56.280
And in fact, our data
show that it depends

00:31:56.280 --> 00:31:59.220
on the subject's experience.

00:31:59.220 --> 00:32:00.450
OK.

00:32:00.450 --> 00:32:03.210
So-- good.

00:32:03.210 --> 00:32:05.950
So yes, we have such a thing.

00:32:05.950 --> 00:32:09.450
It's called the
visual word form area.

00:32:09.450 --> 00:32:14.520
Now, what about this idea that
connectivity of that region

00:32:14.520 --> 00:32:17.730
is playing a role-- it's in
a very systematic location.

00:32:17.730 --> 00:32:20.550
It's that little orange
thing right there.

00:32:20.550 --> 00:32:21.358
Yes, question.

00:32:21.358 --> 00:32:22.150
AUDIENCE: Question.

00:32:22.150 --> 00:32:24.740
I'm just trying to think
through the alternative.

00:32:24.740 --> 00:32:26.520
The brain has to be
shaped by experience,

00:32:26.520 --> 00:32:28.425
otherwise you would never
learn anything, right?

00:32:28.425 --> 00:32:29.130
NANCY KANWISHER: Absolutely.

00:32:29.130 --> 00:32:31.680
AUDIENCE: Even if this
didn't show that difference,

00:32:31.680 --> 00:32:32.700
it would just mean
the difference is

00:32:32.700 --> 00:32:33.750
something you're not measuring.

00:32:33.750 --> 00:32:35.640
NANCY KANWISHER:
Absolutely, absolutely.

00:32:35.640 --> 00:32:38.325
You wouldn't be able to
understand the sentence I'm

00:32:38.325 --> 00:32:40.200
saying right now without
changing your brain,

00:32:40.200 --> 00:32:41.910
because by the time you get
to the end of the sentence,

00:32:41.910 --> 00:32:43.500
you need to remember what
I said at the beginning

00:32:43.500 --> 00:32:45.980
of the sentence, so there's
little things structurally

00:32:45.980 --> 00:32:47.730
wiggling around in
your brain and changing

00:32:47.730 --> 00:32:50.052
synaptic connectivity
online all the time

00:32:50.052 --> 00:32:52.260
or you wouldn't be able to
think, let alone remember.

00:32:52.260 --> 00:32:53.280
Absolutely.

00:32:53.280 --> 00:32:55.150
So the question here
is more specific.

00:32:55.150 --> 00:32:58.230
It's not whether the brain
changes with experience.

00:32:58.230 --> 00:32:59.700
Absolutely, it does.

00:32:59.700 --> 00:33:04.620
It's whether experience can
explain these particular cell

00:33:04.620 --> 00:33:06.840
activities and where
they came from.

00:33:06.840 --> 00:33:09.780
I'm glad you asked
that question.

00:33:09.780 --> 00:33:11.040
OK.

00:33:11.040 --> 00:33:13.860
OK, so now, we've just
argued that the selectivity

00:33:13.860 --> 00:33:16.685
of that little dot, at least,
must be due to experience.

00:33:16.685 --> 00:33:18.060
Doesn't tell us
about the others,

00:33:18.060 --> 00:33:19.560
but tells us that one must be.

00:33:19.560 --> 00:33:23.400
And now we're asking,
can its selectivity--

00:33:23.400 --> 00:33:27.780
can that location be
determined by the connectivity

00:33:27.780 --> 00:33:30.890
of that region?

00:33:30.890 --> 00:33:35.330
So to get to that, we use
diffusion tractography.

00:33:35.330 --> 00:33:37.670
And the hypothesis
here is that it's

00:33:37.670 --> 00:33:40.040
these long-range connections
that determine where

00:33:40.040 --> 00:33:42.410
those functional regions land.

00:33:42.410 --> 00:33:45.107
This is me with a bunch of
functional regions in my head.

00:33:45.107 --> 00:33:46.190
Doesn't matter which ones.

00:33:46.190 --> 00:33:48.830
We're just asking
the general question.

00:33:48.830 --> 00:33:50.960
And so I'm going to skip
over all the details,

00:33:50.960 --> 00:33:52.993
but just give you the
gist of a recent paper

00:33:52.993 --> 00:33:54.410
that we published
looking at this.

00:33:54.410 --> 00:33:57.560
We asked-- we found the
visual word form area.

00:33:57.560 --> 00:34:01.850
That's right down in there,
about there, left hemisphere.

00:34:01.850 --> 00:34:08.150
And we scanned kids at age
eight and age five, same kids.

00:34:08.150 --> 00:34:09.679
Age five, then age eight.

00:34:09.679 --> 00:34:11.810
Here's the age eight data.

00:34:11.810 --> 00:34:15.170
These kids have learned to
read in between the two scans.

00:34:15.170 --> 00:34:17.900
And here is the response
of their visual word form

00:34:17.900 --> 00:34:21.590
area to words, faces, objects,
and scrambled objects.

00:34:21.590 --> 00:34:24.920
Nice and selective, just
like a good visual word form

00:34:24.920 --> 00:34:27.230
area should respond.

00:34:27.230 --> 00:34:29.480
So it's there by age eight.

00:34:29.480 --> 00:34:32.659
What we then do is we take
the data in the same kid

00:34:32.659 --> 00:34:36.320
across those three years,
align the data, and say,

00:34:36.320 --> 00:34:39.920
what were those voxels doing
in that kid at age five

00:34:39.920 --> 00:34:42.139
before they learned to read?

00:34:42.139 --> 00:34:43.969
This is another way
of showing that it's

00:34:43.969 --> 00:34:45.650
experience that was necessary.

00:34:45.650 --> 00:34:48.409
And boom, they were
not word selective.

00:34:48.409 --> 00:34:49.250
They shouldn't be.

00:34:49.250 --> 00:34:51.350
These kids hadn't
learned to read yet.

00:34:51.350 --> 00:34:54.100
But it's still kind of nice
to be able to show that.

00:34:54.100 --> 00:34:54.960
All right?

00:34:54.960 --> 00:34:59.040
But now, the
hypothesis is that it's

00:34:59.040 --> 00:35:02.550
the connectivity at age
five that predicts where

00:35:02.550 --> 00:35:04.630
this region is going to land.

00:35:04.630 --> 00:35:06.990
So we use that same
rigmarole that I showed you

00:35:06.990 --> 00:35:11.190
earlier for adults, where
we used just diffusion data

00:35:11.190 --> 00:35:14.580
to predict where the
functional region will arise.

00:35:14.580 --> 00:35:18.060
But we use the diffusion
data from five-year-olds

00:35:18.060 --> 00:35:19.890
to predict where
that region would

00:35:19.890 --> 00:35:23.010
arise when the kids were eight.

00:35:23.010 --> 00:35:24.690
And it turns out
you can do that.

00:35:24.690 --> 00:35:27.390
You can predict actually
fine-grained individual

00:35:27.390 --> 00:35:30.300
differences in exactly
where the visual word form

00:35:30.300 --> 00:35:33.810
area will arise at age
eight from that same kid's

00:35:33.810 --> 00:35:37.740
connectivity at age five.

00:35:37.740 --> 00:35:39.330
So does everybody
see how that fits

00:35:39.330 --> 00:35:41.820
one of the necessary
conditions for this idea

00:35:41.820 --> 00:35:45.420
that the locations where
these things land later

00:35:45.420 --> 00:35:48.150
in development is
determined by connectivity

00:35:48.150 --> 00:35:50.140
that exists before?

00:35:50.140 --> 00:35:52.007
Now, our study was
done in humans,

00:35:52.007 --> 00:35:53.340
so we didn't have a causal test.

00:35:53.340 --> 00:35:56.320
All we can say is it was there
before, and it's sufficient.

00:35:56.320 --> 00:35:58.680
But we don't know if that's
actually how it worked.

00:35:58.680 --> 00:36:00.820
That's how it is
working on humans.

00:36:00.820 --> 00:36:02.820
But if you put it together
with the ferret data,

00:36:02.820 --> 00:36:05.110
it's pretty suggestive.

00:36:05.110 --> 00:36:05.610
All right?

00:36:05.610 --> 00:36:06.420
Yeah.

00:36:06.420 --> 00:36:08.055
AUDIENCE: Where is
it connected to?

00:36:08.055 --> 00:36:08.888
NANCY KANWISHER: Ah.

00:36:08.888 --> 00:36:09.690
Very good question.

00:36:09.690 --> 00:36:12.480
I'm being very
vague, connectivity.

00:36:12.480 --> 00:36:15.180
This is a long,
complicated issue.

00:36:15.180 --> 00:36:19.080
Most likely, it's connected to
language-y areas, which we'll

00:36:19.080 --> 00:36:21.630
talk about in a
month or so, that

00:36:21.630 --> 00:36:25.290
are out on the lateral surface
and up in the frontal lobe.

00:36:25.290 --> 00:36:26.970
There are papers
claiming that it's

00:36:26.970 --> 00:36:29.850
connected to language-y areas.

00:36:29.850 --> 00:36:32.010
But I'm kind of a
methodological hard ass,

00:36:32.010 --> 00:36:33.900
and I don't quite
believe those data.

00:36:33.900 --> 00:36:36.225
I mean, I think they
have a medium case,

00:36:36.225 --> 00:36:37.350
but they haven't nailed it.

00:36:37.350 --> 00:36:38.550
I've tried to nail it.

00:36:38.550 --> 00:36:41.080
It's hard for all of the
reasons that this method

00:36:41.080 --> 00:36:42.330
that I was complaining about--

00:36:42.330 --> 00:36:45.150
I'm complaining about it
because I'm bitter about it.

00:36:45.150 --> 00:36:46.580
I want this method to be better.

00:36:46.580 --> 00:36:49.080
I want to know what those actual
structural connections are.

00:36:49.080 --> 00:36:54.270
I wish we could put a seed
in the visual word form area

00:36:54.270 --> 00:36:56.305
and follow those
tracks and say not just

00:36:56.305 --> 00:36:57.930
there's enough of a
fingerprint that we

00:36:57.930 --> 00:37:02.190
can predict its function, but
here are the exact connections.

00:37:02.190 --> 00:37:06.930
And it's, mm, not quite up
to that task, in my view.

00:37:06.930 --> 00:37:08.990
It's a big bummer.

00:37:08.990 --> 00:37:11.490
I've wasted a lot of the last
year trying to get that method

00:37:11.490 --> 00:37:15.540
to work, and I haven't quite
given up yet, but I'm close.

00:37:15.540 --> 00:37:16.153
It's OK.

00:37:16.153 --> 00:37:18.570
It's just not good enough to
answer those questions, which

00:37:18.570 --> 00:37:21.530
is very frustrating because
they're pressing questions.

00:37:21.530 --> 00:37:22.030
Yeah.

00:37:22.030 --> 00:37:22.510
AUDIENCE: Can I ask
one more question?

00:37:22.510 --> 00:37:23.427
NANCY KANWISHER: Yeah.

00:37:23.427 --> 00:37:26.880
AUDIENCE: So people
who are blind shouldn't

00:37:26.880 --> 00:37:28.290
have this region active.

00:37:28.290 --> 00:37:30.290
NANCY KANWISHER: Ooh,
very interesting question.

00:37:30.290 --> 00:37:32.520
What do you think?

00:37:32.520 --> 00:37:34.110
People who are blind read.

00:37:37.130 --> 00:37:37.880
What do you think?

00:37:37.880 --> 00:37:40.370
AUDIENCE: So the connection
between here and the visual

00:37:40.370 --> 00:37:46.060
system for the blind people goes
from that region and touching,

00:37:46.060 --> 00:37:46.820
since they're--

00:37:46.820 --> 00:37:47.580
I don't know.

00:37:47.580 --> 00:37:48.890
NANCY KANWISHER: Yeah.

00:37:48.890 --> 00:37:50.180
Yeah, it's not obvious.

00:37:50.180 --> 00:37:51.170
It's not obvious.

00:37:51.170 --> 00:37:52.680
There are several papers--

00:37:52.680 --> 00:37:53.840
which I was going to
put in this lecture

00:37:53.840 --> 00:37:54.860
and I just couldn't fit.

00:37:54.860 --> 00:37:56.270
But there are
several papers that

00:37:56.270 --> 00:38:00.500
argue that tactile Braille
reading in congenitally

00:38:00.500 --> 00:38:04.430
blind people activates
that same region.

00:38:04.430 --> 00:38:05.980
They're pretty good papers.

00:38:05.980 --> 00:38:07.220
I sort of believe it.

00:38:07.220 --> 00:38:09.830
I have-- as I say, I'm a
little bit of a hard ass,

00:38:09.830 --> 00:38:12.860
so I'm not 100% convinced,
but they're pretty compelling,

00:38:12.860 --> 00:38:16.410
and it's a very
interesting question.

00:38:16.410 --> 00:38:17.787
And it's a whole saga.

00:38:17.787 --> 00:38:18.620
It's so interesting.

00:38:18.620 --> 00:38:19.700
I'm going to try
to incorporate more

00:38:19.700 --> 00:38:21.825
of this in a later lecture,
because I didn't fit it

00:38:21.825 --> 00:38:22.610
in here.

00:38:22.610 --> 00:38:24.080
Yeah.

00:38:24.080 --> 00:38:26.690
And the idea would
be, if you had

00:38:26.690 --> 00:38:29.840
to guess, what will those
connections be that drive that?

00:38:29.840 --> 00:38:31.220
Certainly not visual input.

00:38:31.220 --> 00:38:33.030
They're not getting
visual input.

00:38:33.030 --> 00:38:35.300
So it would have to be input
from language-y regions

00:38:35.300 --> 00:38:37.370
or something like
that, that would also

00:38:37.370 --> 00:38:39.110
be present in blind people.

00:38:39.110 --> 00:38:41.340
See what I mean?

00:38:41.340 --> 00:38:42.740
OK.

00:38:42.740 --> 00:38:43.280
All right.

00:38:43.280 --> 00:38:46.730
Anyway, all of this just
to say that it looks

00:38:46.730 --> 00:38:48.913
like the visual
word form area is

00:38:48.913 --> 00:38:50.330
kind of special
in the human brain

00:38:50.330 --> 00:38:53.120
because, one, it shows us
that at least one region gets

00:38:53.120 --> 00:38:56.490
its selectivity from
experience, and two,

00:38:56.490 --> 00:38:59.120
because it develops later,
it gave us this opportunity

00:38:59.120 --> 00:39:02.480
to ask if the connectivity was
present before the function

00:39:02.480 --> 00:39:08.120
as a sort of weak test of this
hypothesis that connectivity

00:39:08.120 --> 00:39:08.990
determines function.

00:39:11.540 --> 00:39:12.380
All right.

00:39:12.380 --> 00:39:13.070
Boom.

00:39:13.070 --> 00:39:14.390
All right, so where are we?

00:39:14.390 --> 00:39:17.360
This really is a shaggy
dog story lecture.

00:39:17.360 --> 00:39:17.870
OK.

00:39:17.870 --> 00:39:21.110
So we started off by saying
a lot of the basic structure

00:39:21.110 --> 00:39:22.730
of the brain is innate.

00:39:22.730 --> 00:39:25.730
Most of the neurons in your
brain, you had at birth.

00:39:25.730 --> 00:39:27.273
Most of the
long-range connections

00:39:27.273 --> 00:39:28.190
were present at birth.

00:39:28.190 --> 00:39:31.550
They weren't yet myelinated,
but they were there.

00:39:31.550 --> 00:39:34.760
We've argued that some of these
selective cortical regions

00:39:34.760 --> 00:39:37.400
appear to depend on experience.

00:39:37.400 --> 00:39:39.080
For example, the
face-deprived monkeys

00:39:39.080 --> 00:39:41.180
don't have face patches.

00:39:41.180 --> 00:39:45.380
And the ferrets see the
response of an auditory cortex

00:39:45.380 --> 00:39:46.970
when their auditory
cortex has been

00:39:46.970 --> 00:39:48.470
rewired to get visual input.

00:39:51.500 --> 00:39:54.050
And further, I've argued
that the visual word form

00:39:54.050 --> 00:39:59.000
area, the selectivity of
that region can't be innate,

00:39:59.000 --> 00:40:02.120
and yet it arises at
a consistent location,

00:40:02.120 --> 00:40:04.310
possibly because of these
long-range connections

00:40:04.310 --> 00:40:07.120
of that region.

00:40:07.120 --> 00:40:09.790
So all of this looks
very experiential,

00:40:09.790 --> 00:40:12.860
aside from the structural
stuff that's present at birth.

00:40:12.860 --> 00:40:15.790
So is Kant toast?

00:40:15.790 --> 00:40:17.680
I started last
lecture as saying he

00:40:17.680 --> 00:40:20.830
was reacting against
the empiricist,

00:40:20.830 --> 00:40:23.410
saying not everything is
derived from experience.

00:40:23.410 --> 00:40:26.860
We need to have a priori
conditions of cognition.

00:40:26.860 --> 00:40:29.440
Remember, he said,
"space can be given

00:40:29.440 --> 00:40:34.390
prior to all actual perceptions,
and so exist in the mind

00:40:34.390 --> 00:40:36.070
a priori.

00:40:36.070 --> 00:40:38.980
And it can contain, prior to
all experience, principles

00:40:38.980 --> 00:40:42.100
which determine the
relations of these objects."

00:40:42.100 --> 00:40:45.310
So he's basically saying we
have an innate representation

00:40:45.310 --> 00:40:46.352
of space.

00:40:46.352 --> 00:40:48.310
And I've just been giving
you all this evidence

00:40:48.310 --> 00:40:50.950
for all the other cases
that experience seems

00:40:50.950 --> 00:40:53.390
to be playing the major role.

00:40:53.390 --> 00:40:57.100
So is it all over for Kant?

00:40:57.100 --> 00:41:00.802
Well, actually, Kant was talking
about space and time primarily,

00:41:00.802 --> 00:41:02.260
and we haven't
considered that yet.

00:41:02.260 --> 00:41:04.840
So let's get back to space.

00:41:04.840 --> 00:41:08.230
Remember these spatial
representations

00:41:08.230 --> 00:41:10.510
that I talked about
in the rodent brain.

00:41:10.510 --> 00:41:12.340
Four different kinds
of neurons that

00:41:12.340 --> 00:41:15.550
are present in adult rodents
that play wonderfully

00:41:15.550 --> 00:41:18.100
different roles in navigation.

00:41:18.100 --> 00:41:20.500
Remember, there are place
cells that fire only

00:41:20.500 --> 00:41:23.560
when the rodent is in a given
known place in his environment.

00:41:23.560 --> 00:41:25.810
There are direction
cells that fire only

00:41:25.810 --> 00:41:27.670
when the rodent is
oriented in a given

00:41:27.670 --> 00:41:29.290
direction in his environment.

00:41:29.290 --> 00:41:31.750
There are border
cells that fire only

00:41:31.750 --> 00:41:34.695
when the rodent is near a
border of the space he's in,

00:41:34.695 --> 00:41:36.070
like right now,
I have cells that

00:41:36.070 --> 00:41:38.403
are firing because I'm next
to this border of this space

00:41:38.403 --> 00:41:41.830
that I'm in, and Anna does not
have any of those cells firing

00:41:41.830 --> 00:41:44.380
because she's in the
middle of this space.

00:41:44.380 --> 00:41:46.930
And there are grid cells that
have this amazing property

00:41:46.930 --> 00:41:50.800
of firing in a hexagonal
array of little micro place

00:41:50.800 --> 00:41:54.820
cells spaced evenly
in a hexagonal array.

00:41:54.820 --> 00:41:58.180
OK, so all of this apparatus
that I talked about last time

00:41:58.180 --> 00:42:00.910
that seems to be playing a role
in your concept of where you

00:42:00.910 --> 00:42:05.810
are, where you're oriented,
and the space around you,

00:42:05.810 --> 00:42:08.033
if we had to take some
representation of space

00:42:08.033 --> 00:42:09.700
that Kant might have
been talking about,

00:42:09.700 --> 00:42:11.440
this would be it.

00:42:11.440 --> 00:42:13.900
So is this stuff innate?

00:42:13.900 --> 00:42:17.650
Well, happily, all this work
was done originally in rodents.

00:42:17.650 --> 00:42:20.260
All the most detailed
work was done in rodents,

00:42:20.260 --> 00:42:23.570
so we can ask that question,
because it's an animal.

00:42:23.570 --> 00:42:24.320
OK?

00:42:24.320 --> 00:42:25.370
All right.

00:42:25.370 --> 00:42:29.270
So what the Mosers
and their colleagues--

00:42:29.270 --> 00:42:33.050
the husband-wife team who
got the Nobel Prize in 2014

00:42:33.050 --> 00:42:34.820
for their work on
the grid cells--

00:42:34.820 --> 00:42:37.610
and O'Keefe and their
colleagues in London,

00:42:37.610 --> 00:42:39.680
who discovered place
cells in the first place--

00:42:39.680 --> 00:42:42.470
two different groups
simultaneously realized what

00:42:42.470 --> 00:42:44.467
a huge, big, fabulous
question this was,

00:42:44.467 --> 00:42:46.550
and they both did the
experiment at the same time,

00:42:46.550 --> 00:42:50.210
and they published it together
at the same time about four

00:42:50.210 --> 00:42:51.200
years ago in--

00:42:51.200 --> 00:42:52.970
I forget-- Science or Nature.

00:42:52.970 --> 00:42:54.500
Big event in the field.

00:42:54.500 --> 00:42:57.020
So they both realize
the same thing.

00:42:57.020 --> 00:43:01.130
The way rodents grow up,
they hang out in a dark nest.

00:43:01.130 --> 00:43:05.720
They're very premature at birth,
and they can't really do much.

00:43:05.720 --> 00:43:06.890
They can't move around.

00:43:06.890 --> 00:43:08.690
All they can do
is turn their head

00:43:08.690 --> 00:43:10.190
toward a nipple and suck milk.

00:43:10.190 --> 00:43:12.470
That's kind of it.

00:43:12.470 --> 00:43:14.660
And so there they are,
in the nest, in the dark.

00:43:14.660 --> 00:43:18.440
Their eyes don't even
open until the end

00:43:18.440 --> 00:43:19.910
of the second week of life.

00:43:19.910 --> 00:43:22.055
And at the same time,
it's the first time

00:43:22.055 --> 00:43:23.930
they emerge from the
nest, and the first time

00:43:23.930 --> 00:43:27.440
they have any experience
navigating, any real experience

00:43:27.440 --> 00:43:29.750
of space.

00:43:29.750 --> 00:43:32.000
And so we can ask
which of those cells

00:43:32.000 --> 00:43:35.570
are present, the very
first experience.

00:43:35.570 --> 00:43:37.400
And it turns out that--

00:43:37.400 --> 00:43:38.900
sorry, this is a
little hard to see.

00:43:38.900 --> 00:43:40.640
There's a light yellow overlay.

00:43:40.640 --> 00:43:42.950
This is the window when
they first open their eyes

00:43:42.950 --> 00:43:45.560
and leave the nest,
between postnatal day

00:43:45.560 --> 00:43:49.160
12 and 14, the end of
the second week of life.

00:43:49.160 --> 00:43:51.290
And what you see is the
head direction cells are

00:43:51.290 --> 00:43:54.440
present immediately, as
soon as the-- can first

00:43:54.440 --> 00:43:59.120
collect neurophysiology data
from these newborn rat pups.

00:43:59.120 --> 00:44:01.560
They're there right away.

00:44:01.560 --> 00:44:05.420
Place cells, you can
get them pretty early,

00:44:05.420 --> 00:44:10.700
and grid cells soon after that.

00:44:10.700 --> 00:44:14.300
So this suggests that in
the rodents, at least,

00:44:14.300 --> 00:44:18.290
their representation
of space as entailed

00:44:18.290 --> 00:44:21.560
in the properties of these
neurons is largely innate.

00:44:25.860 --> 00:44:29.160
So just like Kant said
way back in the 1700s.

00:44:29.160 --> 00:44:30.030
Everybody get this?

00:44:30.030 --> 00:44:31.760
It's pretty cool.

00:44:31.760 --> 00:44:33.510
It's a rare opportunity
where you can just

00:44:33.510 --> 00:44:36.480
take a huge, big
philosophical question

00:44:36.480 --> 00:44:39.510
and, boom, answer it with data.

00:44:39.510 --> 00:44:40.110
Yeah.

00:44:40.110 --> 00:44:40.950
Awesome.

00:44:40.950 --> 00:44:41.760
OK.

00:44:41.760 --> 00:44:42.390
Yes.

00:44:42.390 --> 00:44:43.040
AUDIENCE: Wait, sorry--

00:44:43.040 --> 00:44:43.980
NANCY KANWISHER: I'm
sorry, is it Martin?

00:44:43.980 --> 00:44:44.690
Yeah.

00:44:44.690 --> 00:44:46.690
AUDIENCE: Sorry, are you
saying that it's innate

00:44:46.690 --> 00:44:47.760
or that it's learned?

00:44:47.760 --> 00:44:48.760
NANCY KANWISHER: Innate.

00:44:48.760 --> 00:44:49.260
Innate.

00:44:49.260 --> 00:44:49.875
AUDIENCE: --takes time--

00:44:49.875 --> 00:44:51.630
NANCY KANWISHER:
Because-- oh, yeah.

00:44:51.630 --> 00:44:53.280
OK, important point.

00:44:53.280 --> 00:44:55.980
OK, we don't know before
then whether they existed.

00:44:55.980 --> 00:44:56.970
They were in the nest.

00:44:56.970 --> 00:44:59.460
You can't really
do neurophysiology

00:44:59.460 --> 00:45:01.020
on the rodents in the nest.

00:45:01.020 --> 00:45:03.000
The point is, none of
the relevant experience

00:45:03.000 --> 00:45:04.320
has happened before then.

00:45:04.320 --> 00:45:06.810
They haven't opened their
eyes, they haven't navigated.

00:45:06.810 --> 00:45:08.550
So none of the
experience that could

00:45:08.550 --> 00:45:12.690
be relevant for navigation has
happened before right here,

00:45:12.690 --> 00:45:15.840
on the very first time
that you can test it,

00:45:15.840 --> 00:45:19.140
and the very first time
that they could possibly

00:45:19.140 --> 00:45:21.150
be in the world, seeing
the world, navigating,

00:45:21.150 --> 00:45:22.600
they have them.

00:45:22.600 --> 00:45:24.630
But what you point to
is an important point.

00:45:24.630 --> 00:45:26.088
I mentioned this
briefly last time,

00:45:26.088 --> 00:45:28.447
but it's really worth repeating.

00:45:28.447 --> 00:45:30.030
Innate-- I guess the
word "innate" can

00:45:30.030 --> 00:45:32.790
be used different ways, but
what I mean by innate here,

00:45:32.790 --> 00:45:36.330
the relevant part of innate, the
content to the big questions,

00:45:36.330 --> 00:45:39.540
is whether it's
specified at birth, not

00:45:39.540 --> 00:45:41.640
whether it exists at birth.

00:45:41.640 --> 00:45:44.370
Remember, I gave
the case of puberty.

00:45:44.370 --> 00:45:47.100
Puberty happens way
after birth, but it's not

00:45:47.100 --> 00:45:48.270
the result of experience.

00:45:48.270 --> 00:45:49.770
It's part of a genetic program.

00:45:49.770 --> 00:45:52.170
It's just going to happen.

00:45:52.170 --> 00:45:54.420
I mean, I guess if you don't
eat anything, you'll die

00:45:54.420 --> 00:45:57.090
and then it won't happen,
but within broad latitude,

00:45:57.090 --> 00:45:59.070
it's not the result
of experience.

00:45:59.070 --> 00:46:03.690
And so you can have maturation
on a biological autopilot that

00:46:03.690 --> 00:46:05.700
continues independent
of experience,

00:46:05.700 --> 00:46:07.590
and that's the relevant
kind of innate.

00:46:07.590 --> 00:46:10.190
I realize I was
probably confusing.

00:46:10.190 --> 00:46:13.060
Innate for this purpose
doesn't mean present at birth.

00:46:13.060 --> 00:46:16.050
It means determined
at birth, essentially,

00:46:16.050 --> 00:46:18.980
independent of experience.

00:46:18.980 --> 00:46:19.843
Good.

00:46:19.843 --> 00:46:21.260
You guys are asking
good questions

00:46:21.260 --> 00:46:23.300
and it's helping me be clearer.

00:46:23.300 --> 00:46:24.560
OK.

00:46:24.560 --> 00:46:26.540
OK, so that's cool.

00:46:26.540 --> 00:46:29.060
That says that
those cells are all

00:46:29.060 --> 00:46:32.000
present very early on,
and presumably independent

00:46:32.000 --> 00:46:33.620
of experience.

00:46:33.620 --> 00:46:35.600
What about re-orientation?

00:46:35.600 --> 00:46:38.180
Remember, re-orientation
is this cool thing

00:46:38.180 --> 00:46:39.863
that I carried on
for a long time about

00:46:39.863 --> 00:46:41.030
because it's so interesting.

00:46:41.030 --> 00:46:43.820
Reorientation is this
particular aspect

00:46:43.820 --> 00:46:45.140
of the navigation system.

00:46:45.140 --> 00:46:48.680
It's been studied behaviorally
in rodents, in young humans,

00:46:48.680 --> 00:46:50.480
and human adults.

00:46:50.480 --> 00:46:52.190
And lots of other
animals, actually.

00:46:52.190 --> 00:46:54.620
And the key thing
about reorientation

00:46:54.620 --> 00:46:57.560
is this is how an animal
gets their bearing when

00:46:57.560 --> 00:46:58.760
they're disoriented.

00:46:58.760 --> 00:47:02.630
And the key finding is they use
the shape of space around them.

00:47:02.630 --> 00:47:06.690
They don't use landmarks
to reorient themselves.

00:47:06.690 --> 00:47:07.670
That's the key finding.

00:47:07.670 --> 00:47:09.630
This is all stuff I
talked about before.

00:47:09.630 --> 00:47:13.310
And the evidence that animals
use the shape of space

00:47:13.310 --> 00:47:18.170
to reorient is, when you have
shown a rodent that there's

00:47:18.170 --> 00:47:22.670
goodies in that corner, the
left side of the short wall,

00:47:22.670 --> 00:47:25.520
essentially, and then you
disorient him and put him back

00:47:25.520 --> 00:47:28.430
in the box, he goes 50/50
to those two corners,

00:47:28.430 --> 00:47:29.900
showing that he's
learned something

00:47:29.900 --> 00:47:33.080
like the food is on the
left side of the short wall.

00:47:33.080 --> 00:47:36.020
Not in words, presumably,
but some mental language

00:47:36.020 --> 00:47:38.870
that holds that information.

00:47:38.870 --> 00:47:44.210
OK, so that's using the shape
of space for reorientation.

00:47:44.210 --> 00:47:46.963
Is that ability to use
the shape of space--

00:47:46.963 --> 00:47:48.380
this is a different
sense of space

00:47:48.380 --> 00:47:50.930
than head direction cells, the
shape of space around you--

00:47:50.930 --> 00:47:55.580
is that present
independent of experience?

00:47:55.580 --> 00:47:58.940
Well, again, we can't
test that in humans

00:47:58.940 --> 00:48:01.488
because we can't deprive
humans of experiencing

00:48:01.488 --> 00:48:02.780
the shape of space around them.

00:48:02.780 --> 00:48:03.890
Was there a question?

00:48:03.890 --> 00:48:04.390
No?

00:48:04.390 --> 00:48:06.650
OK, all right.

00:48:06.650 --> 00:48:11.230
But we can test it in
animals with something

00:48:11.230 --> 00:48:14.540
called controlled rearing
that I've talked about before.

00:48:14.540 --> 00:48:18.490
So again, we can't test
this-- even in animals,

00:48:18.490 --> 00:48:19.960
it's hard to test at birth.

00:48:19.960 --> 00:48:23.380
Lots of animals can't navigate
very well at birth, right?

00:48:23.380 --> 00:48:25.708
So we want to test
them after birth,

00:48:25.708 --> 00:48:28.000
but we don't want them to
have the relevant experience,

00:48:28.000 --> 00:48:29.050
because that's
what we're asking,

00:48:29.050 --> 00:48:30.640
is would this
ability be there even

00:48:30.640 --> 00:48:33.340
without the relevant experience.

00:48:33.340 --> 00:48:34.000
OK.

00:48:34.000 --> 00:48:38.470
So the answer to all of
this, the way around this

00:48:38.470 --> 00:48:40.480
is to use controlled rearing.

00:48:40.480 --> 00:48:44.050
Just like Sugita did with
the face-deprived monkeys,

00:48:44.050 --> 00:48:47.170
and just like our Carl also did
with face-deprived monkeys--

00:48:47.170 --> 00:48:50.890
the behavioral study and
the functional MRI study.

00:48:50.890 --> 00:48:54.070
But this will be a
controlled rearing

00:48:54.070 --> 00:48:56.830
study in a different organism,
and it's pretty cute.

00:48:56.830 --> 00:48:57.805
It goes like this.

00:48:57.805 --> 00:48:59.530
This is a group
in Italy that has

00:48:59.530 --> 00:49:01.300
a whole lab that
uses this paradigm,

00:49:01.300 --> 00:49:03.530
and it's very, very powerful.

00:49:03.530 --> 00:49:07.630
So what they do is they--

00:49:07.630 --> 00:49:08.650
again, I just said this.

00:49:08.650 --> 00:49:09.820
The whole idea is
raise an animal

00:49:09.820 --> 00:49:11.650
without the relevant
experience, figure out

00:49:11.650 --> 00:49:13.870
if the ability arises anyway.

00:49:13.870 --> 00:49:17.950
So in this case,
what they do is they

00:49:17.950 --> 00:49:21.670
get fertilized eggs, chicken
eggs from a local hatchery

00:49:21.670 --> 00:49:24.730
that's conveniently
near their lab.

00:49:24.730 --> 00:49:27.460
They bring those fertilized
eggs into the lab

00:49:27.460 --> 00:49:30.790
and put them in an incubator,
and they hatch them

00:49:30.790 --> 00:49:31.375
in darkness.

00:49:34.020 --> 00:49:36.643
Then for the first few days,
you get a nice little chicken.

00:49:36.643 --> 00:49:39.060
It's in the light here, but
that's just so you can see it.

00:49:39.060 --> 00:49:40.990
It actually hatches
in the darkness,

00:49:40.990 --> 00:49:43.860
so there's no visual experience.

00:49:43.860 --> 00:49:48.810
Then you put them in
cages of different shapes.

00:49:48.810 --> 00:49:51.570
Either a nice rectangular
shape like this

00:49:51.570 --> 00:49:53.880
that would be relevant
for reorienting,

00:49:53.880 --> 00:49:57.900
or a circular space like that
that has no geometric cues

00:49:57.900 --> 00:50:00.660
because it's symmetrical.

00:50:00.660 --> 00:50:03.420
So they spend their first
three days of life in one

00:50:03.420 --> 00:50:05.820
or the other of
those containers.

00:50:08.640 --> 00:50:11.920
You then, in order to get a
behavioral result out of them,

00:50:11.920 --> 00:50:13.920
you have to use their
natural behavior, which is

00:50:13.920 --> 00:50:16.710
that they imprint on mama bird.

00:50:16.710 --> 00:50:20.280
And you may know that imprinting
is pretty non-specific.

00:50:20.280 --> 00:50:23.970
Baby birds will imprint on
nearly anything that moves.

00:50:23.970 --> 00:50:26.190
So they take a big,
red plastic object,

00:50:26.190 --> 00:50:28.200
and they dangle it in
the middle of the cage,

00:50:28.200 --> 00:50:30.000
and little chicks
follow the red object.

00:50:30.000 --> 00:50:31.050
That's mom.

00:50:31.050 --> 00:50:34.270
That's what they do.

00:50:34.270 --> 00:50:40.290
So then you can use that
behavior to test their ability.

00:50:40.290 --> 00:50:43.020
And so you get
them in the groove.

00:50:43.020 --> 00:50:48.780
You show them mom, and mom
disappears behind an occluder.

00:50:48.780 --> 00:50:52.410
And then you let the
chick go follow mom,

00:50:52.410 --> 00:50:53.760
which the chick wants to do.

00:50:53.760 --> 00:50:55.320
So they do a few
trials like that.

00:50:55.320 --> 00:50:56.070
They've imprinted.

00:50:56.070 --> 00:50:57.237
They're going to follow mom.

00:50:57.237 --> 00:50:59.190
This gives us a way
to ask the chick,

00:50:59.190 --> 00:51:00.750
where do you think mom is?

00:51:00.750 --> 00:51:02.970
And that gives us
a way to ask, what

00:51:02.970 --> 00:51:06.090
cues are you using to reorient,
even though you've been raised

00:51:06.090 --> 00:51:09.390
without geometric information.

00:51:09.390 --> 00:51:10.200
All right.

00:51:10.200 --> 00:51:12.372
And the thing I really
love about this--

00:51:12.372 --> 00:51:13.830
oh, I guess it's
on a later slide--

00:51:13.830 --> 00:51:15.820
is that after you do
the whole experiment,

00:51:15.820 --> 00:51:17.730
you take one or two
trials on that chick,

00:51:17.730 --> 00:51:18.870
you're done with
that chick, they

00:51:18.870 --> 00:51:20.550
have the relevant
experience, you give them

00:51:20.550 --> 00:51:22.842
back to the hatchery and the
hatchery does their thing.

00:51:22.842 --> 00:51:25.890
So it's just like a really
nice little symbiotic

00:51:25.890 --> 00:51:29.700
science-farming enterprise.

00:51:29.700 --> 00:51:32.070
OK, so here's
actually what they do.

00:51:32.070 --> 00:51:35.610
So here's how the
re-orientation test goes.

00:51:35.610 --> 00:51:39.270
After this chick is raised in
one of those two environments--

00:51:39.270 --> 00:51:42.090
the circular one with no
geometric information,

00:51:42.090 --> 00:51:45.450
or the rectangular one with
geometric information--

00:51:45.450 --> 00:51:49.980
and they've learned to
follow big red plastic mom,

00:51:49.980 --> 00:51:53.640
you then put the chick
in this box here.

00:51:53.640 --> 00:51:55.740
The chick is in there
in this wire mesh

00:51:55.740 --> 00:51:58.150
that holds them in there
so he can't run around.

00:51:58.150 --> 00:51:59.940
He's in this rectangular
space, and there

00:51:59.940 --> 00:52:05.520
are four symmetrical
occluders in the corner.

00:52:05.520 --> 00:52:07.860
You then take the red object--

00:52:07.860 --> 00:52:10.860
mom-- and hide it behind
one of the blue panels

00:52:10.860 --> 00:52:13.770
in full view of the chick.

00:52:13.770 --> 00:52:16.290
So now the chick
knows where mom is.

00:52:16.290 --> 00:52:20.460
Now you bring down
an opaque cylinder

00:52:20.460 --> 00:52:21.960
around where the chick is.

00:52:25.460 --> 00:52:27.620
And while the opaque
cylinder is down,

00:52:27.620 --> 00:52:30.840
you rotate the box 90 degrees.

00:52:30.840 --> 00:52:35.000
So now, the chick has no way
to tell things are rotated,

00:52:35.000 --> 00:52:39.650
I'm disoriented, what's what,
how do I know where to go.

00:52:39.650 --> 00:52:43.310
So this is reorientation in
a newly-hatched chick that's

00:52:43.310 --> 00:52:47.070
been reared under
controlled conditions.

00:52:47.070 --> 00:52:47.570
All right.

00:52:47.570 --> 00:52:51.110
So now, once you rotate
the box, then you

00:52:51.110 --> 00:52:54.980
lift up the opaque
occluder, and the cage,

00:52:54.980 --> 00:52:57.038
and you see where
the chick goes.

00:52:57.038 --> 00:52:57.830
Everybody get this?

00:52:57.830 --> 00:52:59.038
It's a little bit convoluted.

00:52:59.038 --> 00:53:00.590
But it's just a version--

00:53:00.590 --> 00:53:02.982
it's a chick version of the
same reorientation task we've

00:53:02.982 --> 00:53:04.190
been talking about all along.

00:53:08.360 --> 00:53:11.600
You do 16 trials, and then
you give the chick back

00:53:11.600 --> 00:53:13.100
to the hatchery.

00:53:13.100 --> 00:53:13.610
OK.

00:53:13.610 --> 00:53:17.300
So here's what happens
for chicks that are raised

00:53:17.300 --> 00:53:18.890
in that rectangular cage.

00:53:18.890 --> 00:53:21.560
They have geometric experience
during those first three

00:53:21.560 --> 00:53:23.090
days of life.

00:53:23.090 --> 00:53:24.920
So this is kind
of a control case.

00:53:24.920 --> 00:53:27.980
And what you find
is that when you've

00:53:27.980 --> 00:53:31.190
hid mom in a corner that
is on the right side

00:53:31.190 --> 00:53:33.800
of the short wall,
they go preferentially

00:53:33.800 --> 00:53:36.260
to the two corners
consistent with that more

00:53:36.260 --> 00:53:39.620
than the other two corners,
consistent with the idea

00:53:39.620 --> 00:53:41.960
that they can use geometric
information to reorient

00:53:41.960 --> 00:53:42.890
themselves.

00:53:42.890 --> 00:53:46.305
They're not perfect, but
they're way better than chance.

00:53:46.305 --> 00:53:47.180
Does that make sense?

00:53:47.180 --> 00:53:49.650
They go to the two corners
that are consistent,

00:53:49.650 --> 00:53:51.930
showing that they can use
the geometric information.

00:53:51.930 --> 00:53:53.638
But these are the
chicks that were raised

00:53:53.638 --> 00:53:55.520
with the geometric experience.

00:53:55.520 --> 00:53:57.530
What about the chicks
raised in the cylinder,

00:53:57.530 --> 00:54:01.550
without geometric experience?

00:54:01.550 --> 00:54:04.620
They do the same thing.

00:54:04.620 --> 00:54:06.620
And this is the first
time they've experienced--

00:54:06.620 --> 00:54:10.010
this testing condition is the
first time they've experienced

00:54:10.010 --> 00:54:12.950
any space that
isn't symmetrical,

00:54:12.950 --> 00:54:14.930
any place where
they could possibly

00:54:14.930 --> 00:54:17.090
use geometric
information to orient,

00:54:17.090 --> 00:54:18.785
and they do it on
the first trials.

00:54:21.780 --> 00:54:23.700
Everybody got that?

00:54:23.700 --> 00:54:26.070
So that tells us
that this ability

00:54:26.070 --> 00:54:29.550
to reorient based on
the shape of space

00:54:29.550 --> 00:54:34.530
when you're disoriented
doesn't require experience

00:54:34.530 --> 00:54:38.650
with the geometry of space.

00:54:38.650 --> 00:54:42.970
Now, you might be thinking,
well, that cylindrical cage,

00:54:42.970 --> 00:54:46.277
it doesn't have something
to break the symmetry,

00:54:46.277 --> 00:54:47.860
but there's still
something geometric.

00:54:47.860 --> 00:54:49.450
There's a floor, there's a wall.

00:54:49.450 --> 00:54:51.250
I agree, that bugged me too.

00:54:51.250 --> 00:54:54.100
They did another experiment in
which they raised the chicks

00:54:54.100 --> 00:54:55.420
in total darkness.

00:54:55.420 --> 00:54:59.110
First three days, no
visual experience at all,

00:54:59.110 --> 00:55:01.300
and the chicks still do that.

00:55:01.300 --> 00:55:03.460
So no visual experience.

00:55:03.460 --> 00:55:04.962
That's an even stronger case.

00:55:04.962 --> 00:55:06.670
Was there a question
percolating in here?

00:55:06.670 --> 00:55:09.610
I felt like-- no, OK.

00:55:09.610 --> 00:55:10.450
All right.

00:55:10.450 --> 00:55:13.300
So yes, the reorientation
system-- actually,

00:55:13.300 --> 00:55:14.650
that's not well expressed.

00:55:14.650 --> 00:55:18.070
The ability to use
geometry to reorient

00:55:18.070 --> 00:55:20.620
is not based on any
experience with geometry.

00:55:20.620 --> 00:55:24.130
It must be innate in the sense
of not requiring experience.

00:55:29.870 --> 00:55:30.585
So go Kant.

00:55:33.870 --> 00:55:35.070
All right.

00:55:35.070 --> 00:55:36.450
So where have we gotten to?

00:55:36.450 --> 00:55:38.670
Let's recap.

00:55:38.670 --> 00:55:39.810
What's innate?

00:55:39.810 --> 00:55:41.040
OK, in the face system--

00:55:41.040 --> 00:55:43.500
I went through this before,
maybe not that much.

00:55:43.500 --> 00:55:46.800
We could quibble some of
the cases are ambiguous,

00:55:46.800 --> 00:55:50.814
but the main evidence
suggests that--

00:55:50.814 --> 00:55:52.770
before you posit that
something's innate,

00:55:52.770 --> 00:55:54.210
it's like the
evidence-- you have

00:55:54.210 --> 00:55:56.418
to have strong evidence for
innateness to argue with.

00:55:56.418 --> 00:55:58.440
The default case is
not innate, right?

00:55:58.440 --> 00:56:01.530
It's kind of an extreme
claim, and so the default

00:56:01.530 --> 00:56:04.890
is not innate, and so right now,
we don't have a strong argument

00:56:04.890 --> 00:56:07.680
that any of the face system
is innate other than this bias

00:56:07.680 --> 00:56:10.080
to look more at
faces, which as I said

00:56:10.080 --> 00:56:12.420
might be a very
rudimentary template.

00:56:12.420 --> 00:56:14.003
OK.

00:56:14.003 --> 00:56:16.170
I talked about the role of
connectivity and cortical

00:56:16.170 --> 00:56:16.980
development.

00:56:16.980 --> 00:56:19.290
Most of those
long-range connections

00:56:19.290 --> 00:56:21.180
are present at birth.

00:56:21.180 --> 00:56:24.990
I showed that connectivity can
causally affect development

00:56:24.990 --> 00:56:28.170
in the case of the
rewired ferrets.

00:56:28.170 --> 00:56:31.140
I showed that category selective
regions in human adults

00:56:31.140 --> 00:56:33.420
have distinctive connectivity.

00:56:33.420 --> 00:56:35.910
And I showed that in the
visual word form area,

00:56:35.910 --> 00:56:40.830
the distinctive connectivity
is present before the function.

00:56:40.830 --> 00:56:41.370
OK.

00:56:41.370 --> 00:56:44.790
So that tells us that there's
one region in the brain

00:56:44.790 --> 00:56:49.140
that we know the selectivity
of that region can't be innate.

00:56:49.140 --> 00:56:51.060
It doesn't tell us
about all the others.

00:56:51.060 --> 00:56:52.097
Who knows?

00:56:52.097 --> 00:56:53.430
It's kind of an existence proof.

00:56:53.430 --> 00:56:55.360
They might all be
learned by experience.

00:56:55.360 --> 00:56:56.420
We look at faces a lot.

00:56:56.420 --> 00:56:57.420
We look at scenes a lot.

00:56:57.420 --> 00:56:58.620
We look at bodies a lot.

00:56:58.620 --> 00:57:01.770
Maybe they all have the
same experiential basis.

00:57:01.770 --> 00:57:02.910
Doesn't prove it.

00:57:02.910 --> 00:57:05.610
It just says maybe.

00:57:05.610 --> 00:57:06.960
All right.

00:57:06.960 --> 00:57:10.110
But then I showed that for
the space system, actually,

00:57:10.110 --> 00:57:14.310
we do have pretty strong
evidence that a lot of it

00:57:14.310 --> 00:57:16.590
is innate, both in
that the head direction

00:57:16.590 --> 00:57:20.710
cells are present before
any visual experience

00:57:20.710 --> 00:57:22.800
or any navigation.

00:57:22.800 --> 00:57:25.080
And I showed that the
chicks can reorient

00:57:25.080 --> 00:57:27.450
based on the geometry of
space, even if they've never

00:57:27.450 --> 00:57:31.800
seen space or geometry before.

00:57:31.800 --> 00:57:35.130
So bottom line, face
system, who knows,

00:57:35.130 --> 00:57:38.010
but no strong evidence
for innateness.

00:57:38.010 --> 00:57:40.290
Visual word form
area, strong evidence

00:57:40.290 --> 00:57:42.420
that it's experientially
based, and space

00:57:42.420 --> 00:57:45.240
system, strong evidence
that a lot of it is innate.

00:57:48.210 --> 00:57:50.460
OK.

00:57:50.460 --> 00:57:51.240
All right.

00:57:51.240 --> 00:57:52.230
I got us to here.

00:57:52.230 --> 00:57:52.770
All right.

00:57:52.770 --> 00:57:55.620
Now, all of this time,
I've been talking about,

00:57:55.620 --> 00:58:01.740
how do we wire up this system
and its cognitive correlates

00:58:01.740 --> 00:58:02.520
in development?

00:58:02.520 --> 00:58:05.070
What do you have to build
in to get a system like this

00:58:05.070 --> 00:58:06.300
in development?

00:58:06.300 --> 00:58:08.190
What can you get
through learning?

00:58:08.190 --> 00:58:12.100
What do you have to
build in, and so forth.

00:58:12.100 --> 00:58:16.470
But it's a related but
different question to ask,

00:58:16.470 --> 00:58:19.890
is that the only possible
way it could work,

00:58:19.890 --> 00:58:21.540
or are there situations
where we might

00:58:21.540 --> 00:58:25.050
have a very different kind
of organization of the brain?

00:58:25.050 --> 00:58:27.210
Are there other
possible organizations

00:58:27.210 --> 00:58:30.570
that might develop under
different circumstances that

00:58:30.570 --> 00:58:32.500
would still work?

00:58:32.500 --> 00:58:35.190
And the two relevant cases
that people have looked at

00:58:35.190 --> 00:58:37.440
are cases of brain damage.

00:58:37.440 --> 00:58:40.410
So if you have brain
damage in adulthood,

00:58:40.410 --> 00:58:42.780
and you lose a little
piece, can that piece

00:58:42.780 --> 00:58:44.250
move over and reorganize?

00:58:44.250 --> 00:58:47.190
Is there another possible
organization that would work?

00:58:50.610 --> 00:58:52.590
Or what about if
you have very, very

00:58:52.590 --> 00:58:57.540
different visual experience,
like you're born blind.

00:58:57.540 --> 00:58:59.250
Then do you get the
same organization,

00:58:59.250 --> 00:59:01.500
or does everything go
haywire and you have

00:59:01.500 --> 00:59:05.610
a totally different kind
of brain organization?

00:59:05.610 --> 00:59:08.190
All right, so I'll give
you a little bit of data

00:59:08.190 --> 00:59:10.420
on each of those questions.

00:59:10.420 --> 00:59:11.340
All right.

00:59:11.340 --> 00:59:16.350
So first of all, can the brain
reorganize after brain damage?

00:59:16.350 --> 00:59:18.670
The main domain where
people have studied this--

00:59:18.670 --> 00:59:22.050
which we haven't talked about
yet, but we will in a month--

00:59:22.050 --> 00:59:23.640
is the case of language.

00:59:23.640 --> 00:59:26.130
So it's just something there
are lots of studies of this.

00:59:26.130 --> 00:59:28.297
People have been onto this
question for a long time.

00:59:28.297 --> 00:59:32.550
In fact, Broca wrote about
this question 200 years ago.

00:59:32.550 --> 00:59:36.870
So the basic findings are
that if you have damage

00:59:36.870 --> 00:59:40.110
to your language parts of
your brain in adulthood,

00:59:40.110 --> 00:59:42.570
that is not good.

00:59:42.570 --> 00:59:45.120
Often, you'll recover a
little bit of function,

00:59:45.120 --> 00:59:48.730
but you really
won't get it back.

00:59:48.730 --> 00:59:50.913
It's just a big massive drag.

00:59:50.913 --> 00:59:52.830
There are people we will
talk about in a month

00:59:52.830 --> 00:59:56.670
when we get to the
language section who

00:59:56.670 --> 00:59:59.940
have had massive left hemisphere
strokes that basically take out

00:59:59.940 --> 01:00:03.060
their entire language system.

01:00:03.060 --> 01:00:06.840
And it doesn't come back
years after that stroke.

01:00:06.840 --> 01:00:09.600
We'll see, actually, that
they're cognitively pretty

01:00:09.600 --> 01:00:10.860
normal in every other respect.

01:00:10.860 --> 01:00:13.650
It's quite amazing how much
they can do without language,

01:00:13.650 --> 01:00:14.940
which is fascinating.

01:00:14.940 --> 01:00:16.950
But for present purposes,
the main finding

01:00:16.950 --> 01:00:21.450
is brain damage in
adulthood that takes out

01:00:21.450 --> 01:00:23.880
language functions, not good.

01:00:23.880 --> 01:00:26.370
Not much recovery, not
much reorganization.

01:00:26.370 --> 01:00:27.990
By the way, there's
a whole-- it's

01:00:27.990 --> 01:00:30.720
very trendy in popular media
to talk about, oh, the brain is

01:00:30.720 --> 01:00:32.730
plastic, you can
rewire your brain,

01:00:32.730 --> 01:00:36.090
take this-- use this smartphone
app and rewire your brain.

01:00:36.090 --> 01:00:39.267
Mostly, that stuff
is just bullshit.

01:00:39.267 --> 01:00:41.100
You can learn a task,
and you can get better

01:00:41.100 --> 01:00:43.050
at that task, no question.

01:00:43.050 --> 01:00:45.420
But you can't make
yourself smarter.

01:00:45.420 --> 01:00:47.670
You can't rewire
your whole brain.

01:00:47.670 --> 01:00:50.400
That's garbage.

01:00:50.400 --> 01:00:51.270
All right.

01:00:51.270 --> 01:00:53.850
Back to aphasia.

01:00:53.850 --> 01:00:55.050
OK.

01:00:55.050 --> 01:00:59.040
The story is very different
for brain damage in kids.

01:00:59.040 --> 01:01:04.410
If you have brain damage in
the first few months of life

01:01:04.410 --> 01:01:07.780
to language parts of
the brain, as an adult,

01:01:07.780 --> 01:01:10.020
your language function
is pretty good.

01:01:10.020 --> 01:01:11.490
It's not quite perfect.

01:01:11.490 --> 01:01:14.190
Took people a while to discover
that it isn't quite perfect,

01:01:14.190 --> 01:01:15.870
but it's surprisingly good.

01:01:15.870 --> 01:01:18.620
For everyday uses, you
might not even notice.

01:01:18.620 --> 01:01:21.570
You have to test people on
esoteric syntactic things

01:01:21.570 --> 01:01:23.850
to discover that, actually,
it's not quite right.

01:01:23.850 --> 01:01:25.440
But it's very good.

01:01:25.440 --> 01:01:28.770
And typically, what you
see, if you scan these kids,

01:01:28.770 --> 01:01:30.300
is that a lot of
language function

01:01:30.300 --> 01:01:33.450
has reorganized and shifted
over to homologous regions

01:01:33.450 --> 01:01:34.665
in the right hemisphere.

01:01:37.440 --> 01:01:40.080
OK, so that's better news.

01:01:40.080 --> 01:01:44.520
After age five, if you have
brain damage, not so good.

01:01:44.520 --> 01:01:47.340
So it's like there's
some critical period

01:01:47.340 --> 01:01:48.850
for when the brain is plastic.

01:01:48.850 --> 01:01:51.300
You can move language over
to the right hemisphere up

01:01:51.300 --> 01:01:53.895
until around age five, and
after that, you can't really.

01:01:56.860 --> 01:01:59.740
All right, so these consider--
right, that's what I just said.

01:01:59.740 --> 01:02:03.670
So these considerations
have been pulled together

01:02:03.670 --> 01:02:06.820
under something called
the Kennard Principle.

01:02:06.820 --> 01:02:08.620
And the Kennard
Principle basically

01:02:08.620 --> 01:02:11.215
says, if you're going to have
brain damage, have it early.

01:02:14.040 --> 01:02:15.610
Better not to have
the brain damage,

01:02:15.610 --> 01:02:18.510
but if you have to
have it, have it early.

01:02:18.510 --> 01:02:20.550
And that's based on
findings like this--

01:02:20.550 --> 01:02:24.300
the fact that the kids who
have left hemisphere damage

01:02:24.300 --> 01:02:25.920
have much better
language function

01:02:25.920 --> 01:02:29.280
as adults than adults
who have the same kind

01:02:29.280 --> 01:02:31.410
of left hemisphere damage.

01:02:31.410 --> 01:02:34.080
OK, so that's a
reasonable summary

01:02:34.080 --> 01:02:35.970
of the language literature.

01:02:35.970 --> 01:02:41.070
However, this finding
doesn't always hold.

01:02:41.070 --> 01:02:45.990
And it has led others to put
forth the Hebb Principle, which

01:02:45.990 --> 01:02:48.150
is sort of the opposite.

01:02:48.150 --> 01:02:51.190
The idea of the Hebb Principle
is that, first of all,

01:02:51.190 --> 01:02:52.150
it depends.

01:02:52.150 --> 01:02:54.390
It depends on where
the damage is.

01:02:54.390 --> 01:02:58.770
It depends on when you
test after brain damage.

01:02:58.770 --> 01:03:02.277
But the key insight that will
make this seem more sensible--

01:03:02.277 --> 01:03:04.110
at first, you feel like
it's very intuitive.

01:03:04.110 --> 01:03:07.230
Kids are more plastic in
all kinds of ways, right?

01:03:07.230 --> 01:03:10.680
Watch me using a computer,
it drives my students insane,

01:03:10.680 --> 01:03:11.490
I'm so slow.

01:03:11.490 --> 01:03:13.290
One of my students once--

01:03:13.290 --> 01:03:15.120
back when I used to
actually scan subjects,

01:03:15.120 --> 01:03:17.103
one of my students
was watching me scan,

01:03:17.103 --> 01:03:19.020
and he's just getting
more and more impatient,

01:03:19.020 --> 01:03:23.070
and he finally is like, it's
like watching my mother.

01:03:23.070 --> 01:03:26.400
It's just like, you cannot
become as fluent at things when

01:03:26.400 --> 01:03:27.895
you start doing
it when you're 50.

01:03:27.895 --> 01:03:28.770
It's just what it is.

01:03:28.770 --> 01:03:31.290
We've all seen that
manifest in various ways.

01:03:31.290 --> 01:03:33.930
OK, so that's generally
true, and that's

01:03:33.930 --> 01:03:35.730
consistent with this
Kennard principles

01:03:35.730 --> 01:03:39.383
that you have more flexibility
when you're younger than older,

01:03:39.383 --> 01:03:41.550
which is also why you guys
should learn lots of math

01:03:41.550 --> 01:03:43.770
and computer science now
while your brains are still

01:03:43.770 --> 01:03:44.550
good at it.

01:03:44.550 --> 01:03:46.500
Don't wait until you're
40 when it's harder.

01:03:46.500 --> 01:03:47.380
You will need it.

01:03:47.380 --> 01:03:48.880
No matter what field
you are in, you

01:03:48.880 --> 01:03:50.850
will need it, so
do all of that now.

01:03:50.850 --> 01:03:52.020
OK.

01:03:52.020 --> 01:03:54.000
But to get back to
the topic at hand,

01:03:54.000 --> 01:03:57.600
what is the idea behind
the Hebb principle?

01:03:57.600 --> 01:04:02.640
The idea is, think
about building a house.

01:04:02.640 --> 01:04:04.350
You can't build the
first floor if you

01:04:04.350 --> 01:04:06.750
haven't built the foundation.

01:04:06.750 --> 01:04:08.640
Similarly, you might
imagine that there

01:04:08.640 --> 01:04:10.950
are lots of aspects
of cognition that

01:04:10.950 --> 01:04:14.710
are necessary precursors for
other aspects of cognition.

01:04:14.710 --> 01:04:17.353
And if you're wiring
up a whole brain,

01:04:17.353 --> 01:04:19.770
you're not going to develop
those second order ones if you

01:04:19.770 --> 01:04:21.390
don't get the first order ones.

01:04:21.390 --> 01:04:25.560
And so if you have
damage early in life,

01:04:25.560 --> 01:04:29.130
you may have bigger
long-term consequences.

01:04:29.130 --> 01:04:31.680
Really concrete kind
of silly example.

01:04:31.680 --> 01:04:34.560
Suppose you have damage
to primary auditory cortex

01:04:34.560 --> 01:04:36.343
at birth, and you're deaf.

01:04:36.343 --> 01:04:38.760
Well, you're going to have a
harder time learning language

01:04:38.760 --> 01:04:41.243
because you need to
hear to get language.

01:04:41.243 --> 01:04:42.660
I mean, if you
have smart parents,

01:04:42.660 --> 01:04:44.850
they'll teach you sign
language, you'll be OK.

01:04:44.850 --> 01:04:47.550
But this is a necessary
prior condition.

01:04:47.550 --> 01:04:51.270
And so more generally, it turns
out that in a lot of domains,

01:04:51.270 --> 01:04:53.820
some aspects of
brain and cognition

01:04:53.820 --> 01:04:56.940
are necessary precursors for
others, and in those cases,

01:04:56.940 --> 01:04:59.140
the Kennard Principle
doesn't hold.

01:04:59.140 --> 01:04:59.640
OK?

01:05:03.180 --> 01:05:04.230
Blah, blah, blah.

01:05:04.230 --> 01:05:06.090
OK, now let's get--
this is all sort

01:05:06.090 --> 01:05:07.440
of in-principle vague stuff.

01:05:07.440 --> 01:05:10.230
OK, what about visual cortex?

01:05:10.230 --> 01:05:13.080
What about all this stuff
we've been talking about here?

01:05:13.080 --> 01:05:15.570
All of these specialized
regions for different features

01:05:15.570 --> 01:05:17.040
and different
categories, and you

01:05:17.040 --> 01:05:19.470
may notice I've now added
visually-presented words

01:05:19.470 --> 01:05:20.190
on there.

01:05:20.190 --> 01:05:23.200
Remember, visually-presented,
not auditorily.

01:05:23.200 --> 01:05:24.700
Auditory is a whole
different thing.

01:05:24.700 --> 01:05:26.610
This is seeing
words and letters.

01:05:26.610 --> 01:05:31.500
OK, so all of this organization,
can this stuff move around?

01:05:31.500 --> 01:05:35.940
If you lose this thing, can
you regrow it over there?

01:05:35.940 --> 01:05:39.030
Well, not really.

01:05:39.030 --> 01:05:40.860
As I've been talking
about, if you

01:05:40.860 --> 01:05:43.530
have brain damage in
adulthood, you basically

01:05:43.530 --> 01:05:45.910
lose the corresponding
mental function.

01:05:45.910 --> 01:05:48.630
That's why we have all these
neuropsychological syndromes.

01:05:48.630 --> 01:05:51.052
If people could relearn and
just move the function over,

01:05:51.052 --> 01:05:52.260
you wouldn't have a syndrome.

01:05:52.260 --> 01:05:55.050
You might have a transient
problem as you relearned.

01:05:55.050 --> 01:05:57.960
But in fact, if people
get achromatopsia--

01:05:57.960 --> 01:05:59.580
can't see color vision--

01:05:59.580 --> 01:06:01.890
they're not going to
get better, or not much.

01:06:01.890 --> 01:06:03.347
Agnosia, if they
can't see shape,

01:06:03.347 --> 01:06:04.680
they're not going to get better.

01:06:04.680 --> 01:06:06.690
Akinetopsia, they
can't see motion

01:06:06.690 --> 01:06:08.477
after a stroke in adulthood.

01:06:08.477 --> 01:06:09.810
They're not going to get better.

01:06:09.810 --> 01:06:14.430
Prosopagnosia, topographic
disorientation, and alexia--

01:06:14.430 --> 01:06:16.770
inability to read due
to a stroke-- basically,

01:06:16.770 --> 01:06:19.740
people don't really
recover from these things.

01:06:19.740 --> 01:06:21.570
There's a beautiful
recent article

01:06:21.570 --> 01:06:28.470
by a German neuroscientist
who had a stroke

01:06:28.470 --> 01:06:31.350
and couldn't read at--

01:06:31.350 --> 01:06:34.300
I don't know-- age 50,
60, something like that.

01:06:34.300 --> 01:06:36.510
And so made himself an
experimental subject,

01:06:36.510 --> 01:06:39.240
and was just determined
to relearn to read.

01:06:39.240 --> 01:06:41.287
And he did every
possible thing, and he's

01:06:41.287 --> 01:06:42.870
written about this
very interestingly,

01:06:42.870 --> 01:06:45.330
and there's an article
I can put on the website

01:06:45.330 --> 01:06:47.280
if anybody wants to read it.

01:06:47.280 --> 01:06:50.040
He basically retaught
himself to read,

01:06:50.040 --> 01:06:52.710
but he's doing it in
completely different ways

01:06:52.710 --> 01:06:54.390
from what all of you are doing.

01:06:54.390 --> 01:06:55.770
He doesn't have that bit.

01:06:55.770 --> 01:06:57.390
He didn't develop
a new one of those.

01:06:57.390 --> 01:07:00.900
He developed a very different
compensatory strategy that's

01:07:00.900 --> 01:07:03.030
very slow and doesn't
work anywhere near

01:07:03.030 --> 01:07:06.690
as well as reading
does for any of us.

01:07:06.690 --> 01:07:10.440
So basically, in adulthood,
these things can't move around.

01:07:10.440 --> 01:07:12.690
So now, are we talking Kennard
or are we talking Hebb?

01:07:16.170 --> 01:07:19.592
What happens if you get
the damage in childhood?

01:07:19.592 --> 01:07:21.300
Well, I'm raising this
question because I

01:07:21.300 --> 01:07:23.050
think it's big, and
deep, and interesting,

01:07:23.050 --> 01:07:25.770
but there basically isn't
much of an answer to it.

01:07:25.770 --> 01:07:26.692
It's hard to answer.

01:07:26.692 --> 01:07:28.150
I'll give you just
a shred of data,

01:07:28.150 --> 01:07:30.150
but basically, I think
we don't know the answer,

01:07:30.150 --> 01:07:33.480
and I'm dying to
know the answer.

01:07:33.480 --> 01:07:35.730
I'll give you just the
one paper that I know

01:07:35.730 --> 01:07:37.226
of that's relevant to this.

01:07:37.226 --> 01:07:40.140
This is a study from
quite a while ago.

01:07:40.140 --> 01:07:41.760
It's the case of a
patient who's known

01:07:41.760 --> 01:07:43.560
in the literature as Adam.

01:07:43.560 --> 01:07:46.920
And Adam sustained
bilateral damage

01:07:46.920 --> 01:07:50.280
to his ventral visual
pathway, both sides,

01:07:50.280 --> 01:07:53.910
at day one of age
due to a stroke.

01:07:53.910 --> 01:07:56.910
Actually, strokes around
birth are surprisingly common,

01:07:56.910 --> 01:07:59.100
like this happens.

01:07:59.100 --> 01:08:03.038
So this guy basically lost
cortex in a lot of the regions

01:08:03.038 --> 01:08:05.580
that we've been talking about
on the bottom of the brain that

01:08:05.580 --> 01:08:07.380
do high-level vision.

01:08:07.380 --> 01:08:11.880
OK, so he was tested for
this study at age 16.

01:08:11.880 --> 01:08:14.190
Now, his visual
acuity, his ability

01:08:14.190 --> 01:08:16.560
to see fine-grained
stuff is not great,

01:08:16.560 --> 01:08:19.590
and his object recognition
is not perfect,

01:08:19.590 --> 01:08:21.390
but it's not terrible either.

01:08:21.390 --> 01:08:25.770
He can recognize common objects
from photographs and line

01:08:25.770 --> 01:08:27.689
drawings reasonably well.

01:08:27.689 --> 01:08:30.600
So he has some residual vision.

01:08:30.600 --> 01:08:34.750
But he can't recognize
faces at all.

01:08:34.750 --> 01:08:39.397
So he is a fan of this TV
series called Baywatch,

01:08:39.397 --> 01:08:40.439
which I don't know about.

01:08:40.439 --> 01:08:42.180
I don't know if that's
like-- anyway, this study was

01:08:42.180 --> 01:08:43.055
done a long time ago.

01:08:43.055 --> 01:08:46.229
Anyway, some beach
TV series that

01:08:46.229 --> 01:08:49.590
has the same set of characters,
and he was obsessed with this,

01:08:49.590 --> 01:08:51.720
and he watched it
for an hour every day

01:08:51.720 --> 01:08:53.981
for a year and a half.

01:08:53.981 --> 01:08:55.439
And that's just
relevant because we

01:08:55.439 --> 01:08:58.560
know that he has
lots of experience

01:08:58.560 --> 01:09:01.229
looking at these individuals.

01:09:01.229 --> 01:09:05.250
But when tested in the lab
on pictures from Baywatch,

01:09:05.250 --> 01:09:08.279
he couldn't recognize any
of the major protagonists.

01:09:08.279 --> 01:09:12.810
That's just a measure of how
severely prosopagnosic he was.

01:09:12.810 --> 01:09:16.158
So that suggests that when the
relevant parts of the brain,

01:09:16.158 --> 01:09:18.450
that the relevant parts are
already specified at birth,

01:09:18.450 --> 01:09:20.992
and if you lose those parts,
you can't just put that function

01:09:20.992 --> 01:09:23.189
somewhere else.

01:09:23.189 --> 01:09:25.566
So that suggests-- I'm not
leaning too hard on this

01:09:25.566 --> 01:09:27.149
because there's just
very little data.

01:09:27.149 --> 01:09:30.479
This is the best there is.

01:09:30.479 --> 01:09:32.100
So it suggests that those--

01:09:32.100 --> 01:09:36.270
at least the general region
is already specified.

01:09:36.270 --> 01:09:38.100
Can anybody think about
why that might be?

01:09:38.100 --> 01:09:41.160
Why can't you just train up
some other part of cortex?

01:09:41.160 --> 01:09:43.080
Say, his object
recognition is pretty good.

01:09:43.080 --> 01:09:46.149
Why can't you train that part
of the object recognition system

01:09:46.149 --> 01:09:48.074
and just say, OK,
learn to do faces?

01:09:50.767 --> 01:09:52.100
Nobody knows the answer to this.

01:09:52.100 --> 01:09:52.600
Yes.

01:09:52.600 --> 01:09:54.710
AUDIENCE: I don't know
about the [INAUDIBLE]

01:09:54.710 --> 01:09:58.300
it's gone completely, just
maybe because throughout time

01:09:58.300 --> 01:10:03.363
very far back in evolution,
it's a face region.

01:10:03.363 --> 01:10:04.280
NANCY KANWISHER: Yeah.

01:10:04.280 --> 01:10:08.300
Yes, but still-- yeah, I mean,
it's clear that we have it,

01:10:08.300 --> 01:10:11.720
and we probably have it for
some reason and all of that.

01:10:11.720 --> 01:10:14.870
But why couldn't you
just grow a new one over

01:10:14.870 --> 01:10:16.220
in a different part of cortex?

01:10:16.220 --> 01:10:17.810
What's wrong with that
other bit of cortex?

01:10:17.810 --> 01:10:19.602
What might it not have
that you might need.

01:10:19.602 --> 01:10:20.118
[INAUDIBLE]?

01:10:20.118 --> 01:10:21.410
AUDIENCE: The right connection?

01:10:21.410 --> 01:10:23.258
NANCY KANWISHER: Yes!

01:10:23.258 --> 01:10:24.800
I just showed you
guys that there are

01:10:24.800 --> 01:10:26.020
very distinctive connections.

01:10:26.020 --> 01:10:27.020
This is all speculation.

01:10:27.020 --> 01:10:27.860
Nobody knows why.

01:10:27.860 --> 01:10:30.312
I'm just saying
that one guess is

01:10:30.312 --> 01:10:32.270
that the reason these
things can't just take up

01:10:32.270 --> 01:10:33.812
residence someplace
else is they need

01:10:33.812 --> 01:10:39.260
those particular connections to
get the right input to process.

01:10:39.260 --> 01:10:42.990
OK, anyway, this is going
way beyond the data.

01:10:42.990 --> 01:10:45.740
But in principle, people could
get more data of this kind

01:10:45.740 --> 01:10:47.210
and answer this question.

01:10:47.210 --> 01:10:49.490
If I can find the
relevant subjects,

01:10:49.490 --> 01:10:51.860
I'm aiming to do this.

01:10:51.860 --> 01:10:53.690
OK, so let's take
one other case.

01:10:53.690 --> 01:10:58.383
Very different kind of
change to ask, what happens--

01:10:58.383 --> 01:11:00.050
so basically, bottom
line of all of this

01:11:00.050 --> 01:11:03.170
is, stuff doesn't
move around that much.

01:11:03.170 --> 01:11:05.450
Early brain damage
to language regions,

01:11:05.450 --> 01:11:07.280
they can shift to the
homologous regions

01:11:07.280 --> 01:11:08.940
in the right hemisphere.

01:11:08.940 --> 01:11:10.488
But all the other
data that I know of

01:11:10.488 --> 01:11:12.530
suggests you can't just
take anything and move it

01:11:12.530 --> 01:11:15.350
around a few centimeters over.

01:11:15.350 --> 01:11:17.630
At least if you have
the damage in adulthood,

01:11:17.630 --> 01:11:20.090
and maybe even if you
have it pretty early.

01:11:20.090 --> 01:11:21.500
OK, all right.

01:11:21.500 --> 01:11:25.310
So now we're going to say,
OK, might this organization

01:11:25.310 --> 01:11:27.350
nonetheless be very
different if you

01:11:27.350 --> 01:11:30.080
had very different experience?

01:11:30.080 --> 01:11:34.550
So let's take the case
of congenital blindness.

01:11:34.550 --> 01:11:36.080
OK, so how is the
brain organized

01:11:36.080 --> 01:11:38.960
in congenital blindness?

01:11:38.960 --> 01:11:40.520
Well, let's take V1.

01:11:40.520 --> 01:11:43.340
Here's this big chunk of
cortex back here, nice

01:11:43.340 --> 01:11:47.795
big chunk of cortex that, in
all of you guys, does vision.

01:11:47.795 --> 01:11:49.670
What does it do in
congenitally blind people?

01:11:49.670 --> 01:11:50.628
Does it just sit there?

01:11:50.628 --> 01:11:51.560
Do the cells die out?

01:11:51.560 --> 01:11:53.090
Do they just go
dum-dee-dum-dee-dum and they

01:11:53.090 --> 01:11:53.840
don't do anything?

01:11:53.840 --> 01:11:57.900
It's a lot of cortex to
waste on all of that.

01:11:57.900 --> 01:12:00.290
Well, it turns
out, astonishingly,

01:12:00.290 --> 01:12:02.570
that what visual cortex
does in blind people

01:12:02.570 --> 01:12:04.140
is a whole bunch
of other things,

01:12:04.140 --> 01:12:08.300
including,
astonishingly, language.

01:12:08.300 --> 01:12:11.120
So you present a sentence
to subjects through Braille

01:12:11.120 --> 01:12:13.730
or auditorily to blind
subjects in the scanner,

01:12:13.730 --> 01:12:17.460
and you see activation of V1.

01:12:17.460 --> 01:12:22.550
Further, you might think,
well, OK, whatever.

01:12:22.550 --> 01:12:25.760
Just turns on, it has
nothing to do with anything.

01:12:25.760 --> 01:12:29.990
But TMS studies-- V1 is right
near the surface of the brain.

01:12:29.990 --> 01:12:32.870
You can zap that region and ask
if you're disrupting function,

01:12:32.870 --> 01:12:34.760
and you can interfere
with language task

01:12:34.760 --> 01:12:37.830
by zapping V1 in
congenitally blind people.

01:12:37.830 --> 01:12:39.410
So it's not just activated.

01:12:39.410 --> 01:12:42.320
It's doing causal
work in blind people.

01:12:42.320 --> 01:12:43.790
This is mind-blowing.

01:12:43.790 --> 01:12:49.040
This is like a totally
different patch of cortex.

01:12:49.040 --> 01:12:51.560
So yeah, it's hard to think
of more different functions

01:12:51.560 --> 01:12:55.160
than low-level vision and
high-level abstract language

01:12:55.160 --> 01:12:57.020
processing.

01:12:57.020 --> 01:13:01.100
So that suggests radical
possible reorganization,

01:13:01.100 --> 01:13:03.050
in this case, with
different experience.

01:13:07.080 --> 01:13:09.958
OK, what about those regions
on the bottom surface

01:13:09.958 --> 01:13:10.500
of the brain?

01:13:10.500 --> 01:13:13.740
The face, place,
word, and body regions

01:13:13.740 --> 01:13:15.810
that we've been talking
about for so long.

01:13:15.810 --> 01:13:17.352
What do they do in blind people?

01:13:17.352 --> 01:13:19.560
Somebody already asked me
before, maybe [INAUDIBLE]..

01:13:19.560 --> 01:13:20.393
Somebody over there.

01:13:20.393 --> 01:13:23.310
It's my spatial code.

01:13:23.310 --> 01:13:26.100
And there's a lot
of claims that they

01:13:26.100 --> 01:13:29.850
have similar selectivity,
which I'm not totally sure of,

01:13:29.850 --> 01:13:35.460
but let me show you
one piece of data.

01:13:35.460 --> 01:13:36.960
I promised you that
there were going

01:13:36.960 --> 01:13:41.340
to be further contradictions
in the whole saga of the role

01:13:41.340 --> 01:13:43.210
of experience in wiring
up these regions,

01:13:43.210 --> 01:13:46.327
so here's one more
contradictory piece of data.

01:13:46.327 --> 01:13:48.660
OK, this is a paper that was
published just a few months

01:13:48.660 --> 01:13:52.410
ago, and the title of the
paper is that the development

01:13:52.410 --> 01:13:55.230
of visual category selectivity--
that means face place body

01:13:55.230 --> 01:13:56.550
regions, all that stuff--

01:13:56.550 --> 01:13:59.490
in the ventral visual
cortex does not

01:13:59.490 --> 01:14:02.100
require visual experience.

01:14:02.100 --> 01:14:03.370
OK.

01:14:03.370 --> 01:14:04.144
What?

01:14:04.144 --> 01:14:07.230
What, what, what?

01:14:07.230 --> 01:14:08.690
OK, here's what they did.

01:14:08.690 --> 01:14:10.500
They scanned-- pretty
crazy experiment--

01:14:10.500 --> 01:14:14.400
they scanned congenitally
blind subjects while they heard

01:14:14.400 --> 01:14:17.460
sounds that were associated
with faces, bodies, objects,

01:14:17.460 --> 01:14:18.340
and scenes.

01:14:18.340 --> 01:14:21.960
So for example, they might
hear laughing, chewing, blowing

01:14:21.960 --> 01:14:23.280
a kiss, whistling sounds.

01:14:23.280 --> 01:14:24.840
Those are face-related sounds.

01:14:24.840 --> 01:14:28.470
Or they might hear scratching,
hand-clapping, finger-snapping,

01:14:28.470 --> 01:14:30.360
bare footsteps,
knuckle cracking.

01:14:30.360 --> 01:14:32.710
Those are body-related
sounds, et cetera.

01:14:32.710 --> 01:14:36.720
So they're lying in the
scanner hearing these sounds.

01:14:36.720 --> 01:14:38.370
Probably cracking up.

01:14:38.370 --> 01:14:43.110
Now the question is, do
we see face, place, body,

01:14:43.110 --> 01:14:46.290
and object regions
activated from sounds

01:14:46.290 --> 01:14:48.090
in congenitally blind
people listening

01:14:48.090 --> 01:14:50.730
to those categories of sounds?

01:14:50.730 --> 01:14:54.400
And the crazy answer is,
kind of sort of a little bit.

01:14:54.400 --> 01:14:55.590
It's not super strong.

01:14:55.590 --> 01:14:57.940
The data are not
mind-blowing, but let

01:14:57.940 --> 01:14:59.190
me just show you what we have.

01:14:59.190 --> 01:15:01.482
OK, this is the bottom of
the brain, back of the brain.

01:15:01.482 --> 01:15:03.150
Everybody oriented here?

01:15:03.150 --> 01:15:03.870
OK.

01:15:03.870 --> 01:15:04.542
Occipital lobe.

01:15:04.542 --> 01:15:06.000
This is where all
the good stuff is

01:15:06.000 --> 01:15:07.250
that we've been talking about.

01:15:07.250 --> 01:15:07.890
OK.

01:15:07.890 --> 01:15:10.800
So this is now the
sighted control subjects

01:15:10.800 --> 01:15:12.900
looking at visual stimuli.

01:15:12.900 --> 01:15:15.790
So this is a significant
map, P levels.

01:15:15.790 --> 01:15:19.260
And so what you see is
facial activity in red,

01:15:19.260 --> 01:15:22.350
object selectivity in
green, scene selectivity

01:15:22.350 --> 01:15:25.810
in blue, purple,
whatever that is-- blue.

01:15:25.810 --> 01:15:28.230
So that should look
sort of familiar.

01:15:28.230 --> 01:15:31.230
Faces, lateral, scenes, medial.

01:15:31.230 --> 01:15:32.848
Objects, people debate about.

01:15:32.848 --> 01:15:35.140
I haven't talked about it
much, because-- anyway, faces

01:15:35.140 --> 01:15:37.500
and scenes, so stuff
to pay attention to.

01:15:37.500 --> 01:15:39.360
OK.

01:15:39.360 --> 01:15:42.150
And over here-- this
map is the same.

01:15:42.150 --> 01:15:44.550
It just says, never mind
if that voxel reaches

01:15:44.550 --> 01:15:45.960
statistical significance.

01:15:45.960 --> 01:15:50.760
Just plot what category
that voxel responds most to.

01:15:50.760 --> 01:15:53.220
So you just see a big swath.

01:15:53.220 --> 01:15:53.970
All right.

01:15:53.970 --> 01:15:57.660
Now, what do we see for
sighted controls listening

01:15:57.660 --> 01:16:00.000
to the auditory stimuli?

01:16:00.000 --> 01:16:02.970
Not much reaches significance.

01:16:02.970 --> 01:16:05.270
If you drop the threshold
way down and look at this,

01:16:05.270 --> 01:16:06.910
maybe a little bit.

01:16:06.910 --> 01:16:10.050
These are somewhat
correlated, but it's lousy.

01:16:10.050 --> 01:16:14.370
So sighted subjects listening
to those sounds, not much.

01:16:14.370 --> 01:16:16.840
What do you think happens
with blind subjects listening

01:16:16.840 --> 01:16:19.830
to those sounds?

01:16:19.830 --> 01:16:23.400
Well, you get face
selectivity here

01:16:23.400 --> 01:16:25.480
that's statistically
significant.

01:16:25.480 --> 01:16:28.170
And if you drop the threshold
and look at the overall map,

01:16:28.170 --> 01:16:30.690
you see a resemblance
of this map

01:16:30.690 --> 01:16:33.990
to the sighted map, the visual
map in the sighted subjects,

01:16:33.990 --> 01:16:37.540
and this correlation
is highly significant.

01:16:37.540 --> 01:16:40.950
So this is totally weird.

01:16:40.950 --> 01:16:43.740
It says, yes, there's a
similar spatial layout

01:16:43.740 --> 01:16:47.580
on the brain of these same
selectivities in congenitally

01:16:47.580 --> 01:16:49.635
blind subjects who
never saw those stimuli.

01:16:52.170 --> 01:16:54.030
And that's the basis
of their argument,

01:16:54.030 --> 01:16:57.960
that the development of
visually category selectivity

01:16:57.960 --> 01:17:00.450
doesn't require experience.

01:17:00.450 --> 01:17:03.570
But now you may be thinking,
what about that paper

01:17:03.570 --> 01:17:05.880
on face-deprived monkeys?

01:17:05.880 --> 01:17:08.310
The title of which
is, "Seeing faces

01:17:08.310 --> 01:17:11.820
is necessary for
face-domain formation,"

01:17:11.820 --> 01:17:14.020
namely for face patches.

01:17:14.020 --> 01:17:16.950
So these two findings, these
two claims in the titles

01:17:16.950 --> 01:17:20.130
are completely contradictory.

01:17:20.130 --> 01:17:21.330
So we're out of time.

01:17:21.330 --> 01:17:22.740
Nobody knows the answer to this.

01:17:22.740 --> 01:17:24.100
It's an ongoing puzzle.

01:17:24.100 --> 01:17:25.830
There are all kinds
of possibilities.

01:17:25.830 --> 01:17:27.270
They're different
species, they're

01:17:27.270 --> 01:17:28.350
different kinds of tests.

01:17:28.350 --> 01:17:29.970
There are many
things you could say,

01:17:29.970 --> 01:17:32.670
but we're really right on
the horn of a big conundrum

01:17:32.670 --> 01:17:33.750
in the field.

01:17:33.750 --> 01:17:36.210
And all I have to say is
welcome to the cutting edge.

01:17:36.210 --> 01:17:37.500
It's a mess there.

01:17:37.500 --> 01:17:39.590
OK, thank you.