1
00:00:00,000 --> 00:00:02,615
The following content is
provided by MIT open

2
00:00:02,615 --> 00:00:04,756
courseware under a creative
commons license.

3
00:00:04,756 --> 00:00:08,830
Additional information about
our license, and MIT open

4
00:00:08,830 --> 00:00:12,394
courseware, in general, is
available at ocw.mit.edu.

5
00:00:15,450 --> 00:00:17,900
PROFESSOR: Good afternoon.

6
00:00:22,270 --> 00:00:30,250
In response to public
demand here--

7
00:00:30,250 --> 00:00:32,790
let's see how many
people do you--

8
00:00:32,790 --> 00:00:35,940
again, the colors may have faded
by now, but check it

9
00:00:35,940 --> 00:00:37,560
when you tilt your head.

10
00:00:37,560 --> 00:00:38,970
The effect changes.

11
00:00:44,930 --> 00:00:48,530
How many people still have
some sort of an effect?

12
00:00:48,530 --> 00:00:51,780
It's really nice to know that
this course has some lasting

13
00:00:51,780 --> 00:00:53,190
impact on people.

14
00:01:00,100 --> 00:01:08,210
Maybe we should try posting
the Powerpoint on the web,

15
00:01:08,210 --> 00:01:10,430
because then if you want to
build up your own McCullough

16
00:01:10,430 --> 00:01:14,130
effect you can do this and
entertain yourselves.

17
00:01:18,250 --> 00:01:24,680
If you have been following the
story thus far, you should

18
00:01:24,680 --> 00:01:31,790
basically have the idea that
at the front of the visual

19
00:01:31,790 --> 00:01:37,310
system, or of any sensory
system, you've got all sorts

20
00:01:37,310 --> 00:01:41,640
of information coming in.

21
00:01:41,640 --> 00:01:46,830
That the job early processes
in the visual systems do.

22
00:01:46,830 --> 00:01:49,330
The sorts of things that the
early parts of visual cortex

23
00:01:49,330 --> 00:01:53,516
do and say, oh look, there's a
little line at this point in

24
00:01:53,516 --> 00:01:54,660
the visual field.

25
00:01:54,660 --> 00:01:58,830
Oh look, there's a dot there,
and a line here, and it's

26
00:01:58,830 --> 00:02:00,150
moving like that.

27
00:02:00,150 --> 00:02:03,120
Little bits of information
all over the place.

28
00:02:03,120 --> 00:02:06,710
Too much of it for
you to handle.

29
00:02:06,710 --> 00:02:11,140
So, last time I talked about
this bottleneck of attention

30
00:02:11,140 --> 00:02:17,740
that allows only some of it
through to processes that

31
00:02:17,740 --> 00:02:21,980
would do things like,
say, recognition.

32
00:02:21,980 --> 00:02:24,940
And, then, I'm going to run
out of places to draw.

33
00:02:24,940 --> 00:02:26,940
We'll take a little
detour here.

34
00:02:26,940 --> 00:02:32,820
Somewhere up here, you're
going to get perception.

35
00:02:32,820 --> 00:02:38,160
And the job of today's lecture
is to convince you that that

36
00:02:38,160 --> 00:02:42,770
percept, your current
understanding of the world, is

37
00:02:42,770 --> 00:02:46,930
always the result of many,
many inferences.

38
00:02:46,930 --> 00:02:52,530
Many, many guesses of what the
nature of the world might be.

39
00:02:52,530 --> 00:02:59,880
Because, not only is there too
much information coming in,

40
00:02:59,880 --> 00:03:03,490
there's also too little
information coming to specify

41
00:03:03,490 --> 00:03:06,800
exactly what's going on out
there in the world.

42
00:03:06,800 --> 00:03:08,650
Consider just a couple
of the problems --

43
00:03:08,650 --> 00:03:10,280
I think are they on
the hand out?

44
00:03:10,280 --> 00:03:12,790
Couple of them are-- well,
one of the obvious ones.

45
00:03:12,790 --> 00:03:15,520
It's the world is 3D.

46
00:03:15,520 --> 00:03:17,140
Your job is to figure
out what's going

47
00:03:17,140 --> 00:03:18,400
on out in the world.

48
00:03:18,400 --> 00:03:20,680
The input is inherently 2D.

49
00:03:20,680 --> 00:03:25,950
That retina that you've
got is a 2D surface.

50
00:03:25,950 --> 00:03:29,320
So, if you are seeing 3D, you
are recovering that 3D

51
00:03:29,320 --> 00:03:34,640
information from essentially
2D input.

52
00:03:34,640 --> 00:03:40,230
You're collecting information
about light intensities.

53
00:03:40,230 --> 00:03:42,170
You don't care about
light intensity.

54
00:03:42,170 --> 00:03:46,080
You care about surface
properties in the world.

55
00:03:46,080 --> 00:03:49,580
So, everything patch that
you're seeing --

56
00:03:49,580 --> 00:03:57,160
if you're looking at this spot
right here, what you're seeing

57
00:03:57,160 --> 00:04:03,110
is the product of the surface,
the properties of the surface,

58
00:04:03,110 --> 00:04:05,610
and the properties of the
illuminant, the properties of

59
00:04:05,610 --> 00:04:07,000
what's lighting it up.

60
00:04:07,000 --> 00:04:09,870
You don't care about the
properties of the illuminate.

61
00:04:09,870 --> 00:04:15,190
You want to recover just the
properties of the surface.

62
00:04:15,190 --> 00:04:18,350
So how do you successfully
ignore the properties of

63
00:04:18,350 --> 00:04:18,970
illumination?

64
00:04:18,970 --> 00:04:21,360
We'll say a little
bit about that.

65
00:04:21,360 --> 00:04:25,730
The world you're looking is an
essentially stable world.

66
00:04:25,730 --> 00:04:27,700
I mean things move around in
it, but the whole world

67
00:04:27,700 --> 00:04:28,870
doesn't jump around.

68
00:04:28,870 --> 00:04:32,460
But, you're looking at
it from an inherently

69
00:04:32,460 --> 00:04:34,040
unstable vantage point.

70
00:04:34,040 --> 00:04:35,340
You're moving around.

71
00:04:35,340 --> 00:04:38,460
And more to the point, even if
you're salk still, the way

72
00:04:38,460 --> 00:04:39,510
you're looking at the
world is you're

73
00:04:39,510 --> 00:04:42,830
moving your eyes around.

74
00:04:42,830 --> 00:04:44,180
Try this for a moment.

75
00:04:44,180 --> 00:04:46,430
Look at the lower left hand
corner of the screen.

76
00:04:49,430 --> 00:04:51,360
Well, actually that might be a
little large, so look at the

77
00:04:51,360 --> 00:04:54,370
lower left hand corner this
McCollough test pattern.

78
00:04:54,370 --> 00:04:56,220
Look at the lower right
hand corner of the

79
00:04:56,220 --> 00:04:58,680
McCollough test pattern.

80
00:04:58,680 --> 00:05:02,930
Did the McCollough effect
test pattern jump

81
00:05:02,930 --> 00:05:03,640
when you did that?

82
00:05:03,640 --> 00:05:04,060
No.

83
00:05:04,060 --> 00:05:05,090
It didn't.

84
00:05:05,090 --> 00:05:06,560
Why didn't it jump?

85
00:05:06,560 --> 00:05:10,800
Because when you're looking
here-- do I have a laser

86
00:05:10,800 --> 00:05:12,450
pointer today?

87
00:05:12,450 --> 00:05:12,920
No.

88
00:05:12,920 --> 00:05:17,785
Oh well -- when you're looking
here, the bulk of that square

89
00:05:17,785 --> 00:05:19,330
is to the right of fixation.

90
00:05:19,330 --> 00:05:21,460
When you're looking here, the
bulk that image is to the left

91
00:05:21,460 --> 00:05:22,300
the fixation.

92
00:05:22,300 --> 00:05:25,270
So it's in two different
spots on the retina.

93
00:05:25,270 --> 00:05:27,420
If I put something here, and
something here, on your

94
00:05:27,420 --> 00:05:28,920
retina, it'll look
like it's moving.

95
00:05:28,920 --> 00:05:31,120
Why didn't that look
like it's moving?

96
00:05:31,120 --> 00:05:36,910
The reason is, that when you
tell your eyes to move, you

97
00:05:36,910 --> 00:05:43,180
send a copy of that command, in
effect, to visual centers

98
00:05:43,180 --> 00:05:47,430
of your brain saying, look, I
just told my eyes to move,

99
00:05:47,430 --> 00:05:52,480
kindly ignore the
resulting smear.

100
00:05:52,480 --> 00:05:54,130
In fact, I want you
to do two things.

101
00:05:54,130 --> 00:05:57,050
I want you-- you, your
visual system--

102
00:05:57,050 --> 00:06:02,060
I want you to shut down during
the course of the eye

103
00:06:02,060 --> 00:06:08,120
movement, and I want you to
compensate for the fact that

104
00:06:08,120 --> 00:06:10,610
everything is been displaced.

105
00:06:10,610 --> 00:06:13,890
You can see what would happen
if that was not the case, by

106
00:06:13,890 --> 00:06:17,620
taking your finger, and
poking your eye.

107
00:06:20,890 --> 00:06:24,220
On your eyelid, you should try
this, because it's more

108
00:06:24,220 --> 00:06:26,430
interesting if you actually
try this, it if you wiggle

109
00:06:26,430 --> 00:06:27,660
your eyeball.

110
00:06:27,660 --> 00:06:30,450
You can do it slowly, or
you can do it quickly.

111
00:06:30,450 --> 00:06:32,640
Look at me, and poke your
eye ball around.

112
00:06:32,640 --> 00:06:37,070
You will notice that all your
friends look funny.

113
00:06:37,070 --> 00:06:40,440
But, you'll notice that things
are jumping around.

114
00:06:40,440 --> 00:06:41,290
Why is that?

115
00:06:41,290 --> 00:06:44,130
Well look.

116
00:06:44,130 --> 00:06:47,770
Millions of years of evolution
did not provide you with a

117
00:06:47,770 --> 00:06:52,290
mechanism that said, I am now
going to poke my eye.

118
00:06:52,290 --> 00:06:56,150
Please send a copy of that
signal to the visual centers

119
00:06:56,150 --> 00:06:57,770
of the brain, saying
to cancel that out.

120
00:06:57,770 --> 00:06:59,660
There's no cancellation
signal here.

121
00:06:59,660 --> 00:07:02,490
And so, you see the image
moving around.

122
00:07:02,490 --> 00:07:06,040
You don't see the image moving
around when you move your eyes

123
00:07:06,040 --> 00:07:10,870
normally, because you're
compensating for it.

124
00:07:10,870 --> 00:07:12,790
All right, so you're
collecting all this

125
00:07:12,790 --> 00:07:13,310
information.

126
00:07:13,310 --> 00:07:17,170
You're doing your best
to register it in a

127
00:07:17,170 --> 00:07:19,080
stable kind of a way.

128
00:07:19,080 --> 00:07:19,650
Oh look at that.

129
00:07:19,650 --> 00:07:21,870
It also says I'm going to
demonstrate the vestibular

130
00:07:21,870 --> 00:07:23,120
ocular reflex.

131
00:07:25,890 --> 00:07:29,110
Let me do that, just because you
might as well get to use

132
00:07:29,110 --> 00:07:32,520
your fingers some more.

133
00:07:32,520 --> 00:07:35,460
You also want the world not to
jump around too much when

134
00:07:35,460 --> 00:07:36,590
you're moving your head.

135
00:07:36,590 --> 00:07:41,170
One of the things you do very
reflexively is, if you rotate

136
00:07:41,170 --> 00:07:43,140
you take your head one
way, your eyes

137
00:07:43,140 --> 00:07:44,920
counter rotate the other.

138
00:07:44,920 --> 00:07:47,670
That's a very quick reflex.

139
00:07:47,670 --> 00:07:49,560
If you want to see how quick
it is, try this.

140
00:07:49,560 --> 00:07:51,650
Hold your finger out
in front of you.

141
00:07:51,650 --> 00:07:55,940
Look at your finger, and move
your head back and forth, and

142
00:07:55,940 --> 00:07:58,550
just keep your eyes
on the finger.

143
00:07:58,550 --> 00:08:01,320
No problem, right?

144
00:08:01,320 --> 00:08:02,100
You know you can do that.

145
00:08:02,100 --> 00:08:05,540
Now, at the same speed, move
your fingers, and try to keep

146
00:08:05,540 --> 00:08:07,110
your eye on the finger.

147
00:08:07,110 --> 00:08:10,160
Can't do it.

148
00:08:10,160 --> 00:08:16,480
It doesn't work, because that
tracking movement -- there are

149
00:08:16,480 --> 00:08:17,380
more neurons involved.

150
00:08:17,380 --> 00:08:20,770
It's a slower process.

151
00:08:20,770 --> 00:08:25,420
The vestibular ocular reflect
is a very short latency kind

152
00:08:25,420 --> 00:08:29,200
of a reflex designed to keep the
input relatively stable.

153
00:08:29,200 --> 00:08:32,630
So, you take all that lovely
input in, you got all these

154
00:08:32,630 --> 00:08:35,290
little bits information all
over the place, and then

155
00:08:35,290 --> 00:08:39,200
you've gotta make your best
guess about what it is that

156
00:08:39,200 --> 00:08:40,190
you're looking at.

157
00:08:40,190 --> 00:08:42,800
The reason that this is an
interesting picture, it's in

158
00:08:42,800 --> 00:08:45,510
the book, by the way, so when
you can't see here, you can go

159
00:08:45,510 --> 00:08:48,060
and study it in the book,
until you can see the

160
00:08:48,060 --> 00:08:50,170
dalmatian dog that
really is there.

161
00:08:50,170 --> 00:08:52,170
How many people can
see it now?

162
00:08:52,170 --> 00:08:55,470
Oh, we got most of them.

163
00:08:55,470 --> 00:08:59,410
His head is right above my
finger, front paws, back paws.

164
00:08:59,410 --> 00:09:05,730
He's on a road sloping from
lower left to upper right.

165
00:09:05,730 --> 00:09:11,020
The point of a picture like
this, is that it's slows down

166
00:09:11,020 --> 00:09:14,560
the process of inference enough,
that you can sort of

167
00:09:14,560 --> 00:09:15,810
feel it happened.

168
00:09:15,810 --> 00:09:20,410
Normally, when I look out at
you, for instance, my visual

169
00:09:20,410 --> 00:09:24,020
system kicks up one, and only
one, interpretation of what

170
00:09:24,020 --> 00:09:27,900
I'm looking at, so rapidly, that
I never notice all the

171
00:09:27,900 --> 00:09:29,950
work that's involved.

172
00:09:29,950 --> 00:09:32,220
The purpose of this picture, and
really the purpose of this

173
00:09:32,220 --> 00:09:36,230
lecture, is to show some of
the work that's involved.

174
00:09:36,230 --> 00:09:42,220
So what you've got here is a
bunch of isolated little black

175
00:09:42,220 --> 00:09:44,560
and white regions.

176
00:09:44,560 --> 00:09:47,770
In order to figure out what's
going on here, you've got to

177
00:09:47,770 --> 00:09:50,010
decide who goes together.

178
00:09:50,010 --> 00:09:51,920
How are we going to decide
who goes together?

179
00:09:51,920 --> 00:09:54,200
Well lets think of this in the
context of a bunch of little

180
00:09:54,200 --> 00:09:55,620
line segments.

181
00:09:58,870 --> 00:10:01,290
You can, sort of, decide that
some of these guys might go

182
00:10:01,290 --> 00:10:03,950
with some of the other ones but
what I'm going to do, is

183
00:10:03,950 --> 00:10:08,810
rotate all of them by 90
degrees, if I recall.

184
00:10:08,810 --> 00:10:09,000
No.

185
00:10:09,000 --> 00:10:10,490
Maybe some other orientation.

186
00:10:10,490 --> 00:10:15,580
But anyway, now they are the
same lines on the screen, but

187
00:10:15,580 --> 00:10:18,110
now some of them hang
together, right?

188
00:10:18,110 --> 00:10:22,170
Got that sort of potato
shape thing there.

189
00:10:22,170 --> 00:10:23,190
Why?

190
00:10:23,190 --> 00:10:26,970
What makes these guys go along
with each other now in a way

191
00:10:26,970 --> 00:10:27,910
that they didn't before?

192
00:10:27,910 --> 00:10:35,170
Well, if you're a little chunk
of brain, whose job it is to

193
00:10:35,170 --> 00:10:38,290
figure out where contours are
out in the world, what you're

194
00:10:38,290 --> 00:10:41,570
getting from earlier in the
visual system, is word that

195
00:10:41,570 --> 00:10:43,450
there's a little bit of contour
here, a little bit of

196
00:10:43,450 --> 00:10:45,370
a contour here, a little
bit of a contour here.

197
00:10:45,370 --> 00:10:47,300
I wonder if those should
go together?

198
00:10:47,300 --> 00:10:50,260
Now how do we decide that the
little bits of contour might

199
00:10:50,260 --> 00:10:51,070
go together?

200
00:10:51,070 --> 00:10:53,750
Well, you might do something
like this.

201
00:11:00,650 --> 00:11:06,070
If you've got a line here, and
you're asking what's the best

202
00:11:06,070 --> 00:11:15,080
bet about if this is really a
piece of a continuing contour?

203
00:11:15,080 --> 00:11:17,120
Where's this likely
to go next?

204
00:11:17,120 --> 00:11:20,530
Well, it might make a hair pin
turn and go off that way.

205
00:11:20,530 --> 00:11:21,900
Doesn't seem really likely.

206
00:11:21,900 --> 00:11:27,050
More likely, it's going to go
off in something like the

207
00:11:27,050 --> 00:11:28,530
direction that it's pointing.

208
00:11:28,530 --> 00:11:33,880
And that turns out to be
the potato shaped ones.

209
00:11:33,880 --> 00:11:37,530
This is a very rule
governed behavior.

210
00:11:37,530 --> 00:11:42,760
If you've got a bunch a little
line segments, as long as the

211
00:11:42,760 --> 00:11:47,540
deviation here isn't more than
about, as I recall, 30 degrees

212
00:11:47,540 --> 00:11:51,920
or so, from co-linear, you're
willing to string those

213
00:11:51,920 --> 00:11:54,100
together pretty happily.

214
00:11:54,100 --> 00:11:58,680
If it starts to be more than,
that your unlikely to string

215
00:11:58,680 --> 00:11:59,270
them together.

216
00:11:59,270 --> 00:12:02,360
The beginning of an effort
to tie little pieces of

217
00:12:02,360 --> 00:12:08,490
information together into
larger structures.

218
00:12:08,490 --> 00:12:09,770
All right, what do
you see here?

219
00:12:12,490 --> 00:12:13,730
Two lines crossing each other.

220
00:12:13,730 --> 00:12:17,220
A reasonable interpretation.

221
00:12:17,220 --> 00:12:19,080
Though, it's not the
only possible

222
00:12:19,080 --> 00:12:20,140
interpretation of this.

223
00:12:20,140 --> 00:12:22,800
I mean, it could be something
like this.

224
00:12:22,800 --> 00:12:27,850
Two birds kissing each other,
or something like that.

225
00:12:27,850 --> 00:12:40,600
Why do you see and, in fact,
this is something like that.

226
00:12:40,600 --> 00:12:43,720
And you see that as an x with
two lines crossing each other,

227
00:12:43,720 --> 00:12:58,070
but if I provide enough other
details here, it's a sea lion?

228
00:12:58,070 --> 00:12:59,320
I don't know what it is.

229
00:13:02,710 --> 00:13:05,130
The point is, he's still
a lousy artist.

230
00:13:05,130 --> 00:13:09,110
It hasn't gotten any better.

231
00:13:09,110 --> 00:13:10,910
The point is more or
less the same.

232
00:13:10,910 --> 00:13:18,190
You've got this little process
that's worrying about that

233
00:13:18,190 --> 00:13:25,360
little piece of line segment,
gets to this junction, and is

234
00:13:25,360 --> 00:13:28,630
busy doing the three roads
diversion of yellow wood kind

235
00:13:28,630 --> 00:13:31,170
of thing, trying to decide
which way to go.

236
00:13:31,170 --> 00:13:36,340
And, it's guessing that all
else being equal, I should

237
00:13:36,340 --> 00:13:37,590
probably go with that.

238
00:13:47,380 --> 00:13:50,330
It sometimes goes by the name
of good continuation.

239
00:13:50,330 --> 00:13:53,880
It's one of a variety of so-
called grouping rules that

240
00:13:53,880 --> 00:13:58,070
were developed first by the
Gestalt psychologists starting

241
00:13:58,070 --> 00:14:02,950
in the early part of the
twentieth century, and they're

242
00:14:02,950 --> 00:14:10,110
really rules for figuring out
how bits of the scene might

243
00:14:10,110 --> 00:14:12,260
hang together.

244
00:14:12,260 --> 00:14:17,230
You can see a similar sort
of process going on here.

245
00:14:17,230 --> 00:14:19,800
You can see that as three
isolated line segments, but

246
00:14:19,800 --> 00:14:21,120
you probably don't.

247
00:14:21,120 --> 00:14:26,060
You see that as a curvy lines
that occluded, right?

248
00:14:26,060 --> 00:14:28,590
If I suddenly reveal this,
you don't go, ooh.

249
00:14:28,590 --> 00:14:31,170
Amazing!

250
00:14:31,170 --> 00:14:33,210
Kind of what I thought
was there.

251
00:14:33,210 --> 00:14:37,530
That's also highly
rule governed.

252
00:14:40,040 --> 00:14:50,950
If you've got a line segment
and you've got another line

253
00:14:50,950 --> 00:14:55,090
segment, you're perfectly happy
to see this one and this

254
00:14:55,090 --> 00:14:56,340
one is connected.

255
00:14:59,870 --> 00:15:00,550
Well how about this.

256
00:15:00,550 --> 00:15:01,870
This will do.

257
00:15:01,870 --> 00:15:04,450
If I put one up here.

258
00:15:04,450 --> 00:15:06,810
All the more if I erase
the stuff in between.

259
00:15:11,470 --> 00:15:13,590
That, you're less likely
to see as connected.

260
00:15:13,590 --> 00:15:16,880
Now why are you less likely
to see that as connected?

261
00:15:16,880 --> 00:15:22,840
The rule turns out to be, that
if I can connect two lines

262
00:15:22,840 --> 00:15:29,500
with a smooth curve,
I'm in business.

263
00:15:29,500 --> 00:15:30,900
I'll be willing to see
those as connected.

264
00:15:30,900 --> 00:15:33,320
But, if I have to put an
inflection in it to make it

265
00:15:33,320 --> 00:15:36,440
work, then it doesn't
look as convincing.

266
00:15:36,440 --> 00:15:39,190
I mean it's not that I deny
the possibility this could

267
00:15:39,190 --> 00:15:46,560
ever connect with that, but if
I give people a bunch of

268
00:15:46,560 --> 00:15:52,880
stimuli like this, and ask
good connection or a bad

269
00:15:52,880 --> 00:15:54,040
connection?

270
00:15:54,040 --> 00:15:57,390
The good connections are the
ones that can be done can with

271
00:15:57,390 --> 00:15:58,650
a single smooth curve.

272
00:15:58,650 --> 00:15:59,570
That can't quite.

273
00:15:59,570 --> 00:16:00,790
I think you probably
have to get an

274
00:16:00,790 --> 00:16:02,550
inflection in there somewhere.

275
00:16:02,550 --> 00:16:07,130
And if you have to inflict
the curve, it fails.

276
00:16:07,130 --> 00:16:13,170
People report it as looking less
convincingly continuous.

277
00:16:13,170 --> 00:16:16,260
All right, it's voting
times here.

278
00:16:16,260 --> 00:16:19,050
Here we've got a whole bunch
isolated guys, but they do

279
00:16:19,050 --> 00:16:21,780
seem to have something to
do with each other.

280
00:16:21,780 --> 00:16:26,400
If you had to pick, this being
organized by columns or rows,

281
00:16:26,400 --> 00:16:28,980
how many vote for columns?

282
00:16:28,980 --> 00:16:31,420
How many vote for rows?

283
00:16:31,420 --> 00:16:33,410
Ain't much to chose, right?

284
00:16:33,410 --> 00:16:38,180
But, if I do this, OK, how
many vote for columns?

285
00:16:38,180 --> 00:16:40,300
How many vote for rows?

286
00:16:40,300 --> 00:16:42,230
So, we've now skewed it
very heavily in the

287
00:16:42,230 --> 00:16:44,190
direction of columns.

288
00:16:44,190 --> 00:16:46,100
And, all that I've done
is change the

289
00:16:46,100 --> 00:16:49,170
proximity of elements.

290
00:16:49,170 --> 00:16:55,270
Once the distance from one item
to the next item in a

291
00:16:55,270 --> 00:16:59,030
vertical direction is closer
than to the items in the

292
00:16:59,030 --> 00:17:04,870
horizontal direction, it's
another one just these gestalt

293
00:17:04,870 --> 00:17:08,590
grouping rules of proximity
takes over, and says, well,

294
00:17:08,590 --> 00:17:11,940
all else being equal, if I had
to guess who goes with who,

295
00:17:11,940 --> 00:17:13,640
the guys that are close
to each other.

296
00:17:13,640 --> 00:17:17,670
They probably go with
each other.

297
00:17:17,670 --> 00:17:21,430
Multiple rules operate at the
same time, so I'll keep the

298
00:17:21,430 --> 00:17:24,590
proximity rule working here.

299
00:17:24,590 --> 00:17:28,940
Now, if you have to vote, how
many vote for columns?

300
00:17:28,940 --> 00:17:30,570
How many votes for rows?

301
00:17:30,570 --> 00:17:34,040
So now, I've skewed it very
heavily in the direction of

302
00:17:34,040 --> 00:17:40,000
rows, even though the proximity
rule, is still going

303
00:17:40,000 --> 00:17:40,460
for columns.

304
00:17:40,460 --> 00:17:43,000
Those things are closer to
each other in a vertical

305
00:17:43,000 --> 00:17:44,390
direction that horizontal.

306
00:17:44,390 --> 00:17:48,880
In this case, the similarity
is trumping that.

307
00:17:48,880 --> 00:17:51,300
You can balance these off
against each other.

308
00:17:51,300 --> 00:17:54,220
How similar to does it need to
be to compensate for a two to

309
00:17:54,220 --> 00:17:59,200
one difference in distance for
instance or something like

310
00:17:59,200 --> 00:18:01,970
that, but the important point
here is, that what you're

311
00:18:01,970 --> 00:18:04,170
trying to do.

312
00:18:04,170 --> 00:18:07,880
These are, you know,
demonstration versions of

313
00:18:07,880 --> 00:18:10,220
presumably what you're
doing all the time.

314
00:18:10,220 --> 00:18:20,340
I'm looking out there, and I'm
seeing regions of redness.

315
00:18:20,340 --> 00:18:24,430
I see sort of disconnected
regions of redness, but I'm

316
00:18:24,430 --> 00:18:26,890
guessing, they're all
part of her top.

317
00:18:26,890 --> 00:18:32,080
It's your top there.

318
00:18:32,080 --> 00:18:33,460
It's you.

319
00:18:33,460 --> 00:18:34,160
No, not her.

320
00:18:34,160 --> 00:18:36,890
That's pink, you're wearing.

321
00:18:36,890 --> 00:18:37,900
The woman behind you.

322
00:18:37,900 --> 00:18:40,190
Well, the woman next
to you is also red.

323
00:18:40,190 --> 00:18:49,030
But, in this case, you can see
the role of proximity here.

324
00:18:49,030 --> 00:18:51,450
So, the similarity thing is
telling me, all those red

325
00:18:51,450 --> 00:18:54,040
things are tied together, and
I'm making it into sort of one

326
00:18:54,040 --> 00:18:55,160
piece of clothing.

327
00:18:55,160 --> 00:19:00,340
The proximity thing is saying,
well I don't think her red

328
00:19:00,340 --> 00:19:03,490
top, and her red top are
the same red top.

329
00:19:03,490 --> 00:19:05,160
That would be a very
weird assumption.

330
00:19:05,160 --> 00:19:08,070
On the other hand, the guy she's
sitting next to is also

331
00:19:08,070 --> 00:19:09,290
wearing red.

332
00:19:09,290 --> 00:19:12,340
So maybe they're just
wearing one garment.

333
00:19:12,340 --> 00:19:13,670
No.

334
00:19:13,670 --> 00:19:14,620
I'm probably not going
to come up with

335
00:19:14,620 --> 00:19:15,530
that assumption either.

336
00:19:15,530 --> 00:19:21,960
But what my visual system is
doing, is continuously trying

337
00:19:21,960 --> 00:19:29,840
to cut the world up into
meaningful chunks, that are

338
00:19:29,840 --> 00:19:33,170
going to be worth subsequent
analysis.

339
00:19:33,170 --> 00:19:36,690
I don't want to go off and
analyze every little pixel in

340
00:19:36,690 --> 00:19:37,170
this scene.

341
00:19:37,170 --> 00:19:40,950
I don't have the brain
power to do that.

342
00:19:40,950 --> 00:19:43,420
I want to have meaningful
chunks that are worth

343
00:19:43,420 --> 00:19:46,000
analyzing, so here I might
decide to the meaningful

344
00:19:46,000 --> 00:19:48,170
chunks were the rows.

345
00:19:48,170 --> 00:19:49,910
They are a something or other.

346
00:19:52,450 --> 00:19:53,700
OK.

347
00:19:57,630 --> 00:19:59,530
There's a couple of reasons
why you need to do this

348
00:19:59,530 --> 00:20:06,870
grouping business over little
elements in the world.

349
00:20:06,870 --> 00:20:07,450
One of them--

350
00:20:07,450 --> 00:20:10,260
I was sort of cartooning over
here-- which is that early in

351
00:20:10,260 --> 00:20:13,120
the visual system, the chunks of
the brain that are looking

352
00:20:13,120 --> 00:20:15,540
at it bits of the world are
only looking at very teeny

353
00:20:15,540 --> 00:20:17,940
bits, and you're going to have
to tie those together.

354
00:20:17,940 --> 00:20:22,770
The other reason, is that out
in the world, contours don't

355
00:20:22,770 --> 00:20:24,260
behave well.

356
00:20:24,260 --> 00:20:27,750
They tend to do awkward things
like disappear on you.

357
00:20:27,750 --> 00:20:31,010
And you don't want to get the
idea that these are--

358
00:20:39,330 --> 00:20:43,050
if I've got a contour like that,
if for some reason bits

359
00:20:43,050 --> 00:20:46,430
of it are deleted, maybe because
there's an occluder,

360
00:20:46,430 --> 00:20:48,900
or maybe because something
just bad happened in the

361
00:20:48,900 --> 00:20:52,950
image, you don't want to lose
this whole structure, because

362
00:20:52,950 --> 00:20:55,880
bits of it have been deleted.

363
00:20:55,880 --> 00:21:00,880
So you have a lot clever
mechanisms designed to help

364
00:21:00,880 --> 00:21:05,150
you find where the edges
are of things out

365
00:21:05,150 --> 00:21:07,870
there in the scene.

366
00:21:07,870 --> 00:21:10,300
And, putting the bits of
edges together into a

367
00:21:10,300 --> 00:21:11,440
long coherent one.

368
00:21:11,440 --> 00:21:13,290
If you play with Photoshop,
you can go and

369
00:21:13,290 --> 00:21:14,300
find the edges I think.

370
00:21:14,300 --> 00:21:17,520
Isn't there like a find edges
filter, or something?

371
00:21:17,520 --> 00:21:19,080
So, try that sometime.

372
00:21:19,080 --> 00:21:28,110
Do find edges on an image of,
say, a person, and you'll see

373
00:21:28,110 --> 00:21:30,940
that it comes up with a lot of
edges that you recognize as

374
00:21:30,940 --> 00:21:33,050
being related to this person,
but its fragmented

375
00:21:33,050 --> 00:21:34,150
all over the place.

376
00:21:34,150 --> 00:21:38,490
And, in fact, getting your
computer to figure out which

377
00:21:38,490 --> 00:21:43,420
bits go with each other is
a tricky piece of work.

378
00:21:43,420 --> 00:21:44,790
It's tricky for you too.

379
00:21:44,790 --> 00:21:46,470
You just don't know that it's
tricky, because, it

380
00:21:46,470 --> 00:21:47,250
works all the time.

381
00:21:47,250 --> 00:21:47,680
All right.

382
00:21:47,680 --> 00:21:53,110
So, this beautifully boring
stimulus is there, because you

383
00:21:53,110 --> 00:21:59,390
see this beautiful vertical
contour, right?

384
00:21:59,390 --> 00:22:02,110
Now, all I'm going to do, I'm
going to leave it there, but

385
00:22:02,110 --> 00:22:05,340
I'm going to put a new
background on it.

386
00:22:05,340 --> 00:22:07,280
Isn't that lovely?

387
00:22:07,280 --> 00:22:08,830
Why is that interesting?

388
00:22:08,830 --> 00:22:12,960
Well, among the reasons it is
interesting, is because it's

389
00:22:12,960 --> 00:22:15,430
the same gray stuff that
was there before.

390
00:22:15,430 --> 00:22:18,190
So, the top of that bar, and
the bottom of that bar are

391
00:22:18,190 --> 00:22:20,410
still identical.

392
00:22:20,410 --> 00:22:24,710
All four of those little
rectangles are all identical.

393
00:22:24,710 --> 00:22:27,490
Even though they no longer
look identical.

394
00:22:27,490 --> 00:22:30,000
Because what the system
is doing, is--

395
00:22:30,000 --> 00:22:31,870
actually can I do this?

396
00:22:31,870 --> 00:22:33,820
I forget what I programmed.

397
00:22:33,820 --> 00:22:35,320
Oh, there we go.

398
00:22:35,320 --> 00:22:38,180
little key's going to
turn into that one.

399
00:22:38,180 --> 00:22:40,500
Isn't that fun?

400
00:22:40,500 --> 00:22:41,690
Get back there.

401
00:22:41,690 --> 00:22:43,010
There we go.

402
00:22:43,010 --> 00:22:46,600
-- but my real point --
so what that is, is a

403
00:22:46,600 --> 00:22:49,350
simultaneous contrast
affect, by the way.

404
00:22:49,350 --> 00:22:53,890
This square looks bright,
because it's surrounded by

405
00:22:53,890 --> 00:22:55,230
darker stuff.

406
00:22:55,230 --> 00:22:57,840
That square looks dark, because
it's surrounded by

407
00:22:57,840 --> 00:22:58,960
brighter stuff.

408
00:22:58,960 --> 00:23:03,410
And even though this bar is
continuous with it's gray

409
00:23:03,410 --> 00:23:08,140
level from bottom to top, it
picks up it's apparent

410
00:23:08,140 --> 00:23:11,170
brightness from the

411
00:23:11,170 --> 00:23:13,190
immediately surrounding contours.

412
00:23:13,190 --> 00:23:17,710
The interesting aspect of this
from the point of view of

413
00:23:17,710 --> 00:23:25,600
understanding were edges are, is
that if the bar is brighter

414
00:23:25,600 --> 00:23:29,440
than the background here, and
darker than the background up

415
00:23:29,440 --> 00:23:33,820
there, there must be a place in
the middle where it's gone.

416
00:23:33,820 --> 00:23:35,220
Where there is no contour.

417
00:23:35,220 --> 00:23:36,470
But you don't see that.

418
00:23:39,980 --> 00:23:45,740
I was too busy making this
little thing move, and I

419
00:23:45,740 --> 00:23:46,540
forgot to do that.

420
00:23:46,540 --> 00:23:47,630
Oh well.

421
00:23:47,630 --> 00:23:50,340
It's not a small region.

422
00:23:50,340 --> 00:23:53,170
There's a fairly sizable region
of that middle there

423
00:23:53,170 --> 00:23:56,870
where there is no physical
contour.

424
00:23:56,870 --> 00:23:58,890
But you fill it in.

425
00:23:58,890 --> 00:24:05,240
You know, in some fashion,
that contour is there.

426
00:24:05,240 --> 00:24:07,280
That's called a subjective
contour, where you're

427
00:24:07,280 --> 00:24:12,260
completing a contour that
doesn't have any real support

428
00:24:12,260 --> 00:24:12,820
in the image.

429
00:24:12,820 --> 00:24:15,360
That's the piece that your
Photoshop filter will have a

430
00:24:15,360 --> 00:24:17,360
hard time doing, by the way.

431
00:24:17,360 --> 00:24:24,370
Oh, we can also do a second
order affect here,

432
00:24:24,370 --> 00:24:25,270
that's kind of --

433
00:24:25,270 --> 00:24:26,980
well, no, we'll do that later.

434
00:24:26,980 --> 00:24:27,050
Oh.

435
00:24:27,050 --> 00:24:27,480
Come on.

436
00:24:27,480 --> 00:24:28,100
Go away.

437
00:24:28,100 --> 00:24:31,630
You've moved often enough.

438
00:24:31,630 --> 00:24:32,850
OK.

439
00:24:32,850 --> 00:24:34,440
Let's continue the
same point here.

440
00:24:34,440 --> 00:24:37,390
All right, everybody sees
this rectangle, right?

441
00:24:37,390 --> 00:24:43,980
And what are those
black things?

442
00:24:46,830 --> 00:24:47,750
Three quarter circles.

443
00:24:47,750 --> 00:24:47,950
Yes.

444
00:24:47,950 --> 00:24:50,370
The literalists figured out that
these are Pacmen or three

445
00:24:50,370 --> 00:24:52,880
quarter circles, or something
like that, but you didn't

446
00:24:52,880 --> 00:24:54,090
really see that when
it came up.

447
00:24:54,090 --> 00:24:56,770
You said, oh that's a rectangle
sitting on top of

448
00:24:56,770 --> 00:25:01,190
four circles of some variety,
and, in fact, you still see it

449
00:25:01,190 --> 00:25:04,800
as a rectangle sitting on
top of four circles.

450
00:25:04,800 --> 00:25:08,270
In fact, you're probably
reasonably convinced that you

451
00:25:08,270 --> 00:25:09,540
can see the contour.

452
00:25:12,750 --> 00:25:13,590
That's weird.

453
00:25:13,590 --> 00:25:15,490
I'm reasonably convinced that
there's an interesting

454
00:25:15,490 --> 00:25:19,070
artifact on the screen that's
creating contours.

455
00:25:19,070 --> 00:25:20,170
I don't know what
that's about.

456
00:25:20,170 --> 00:25:21,170
It looks better over there.

457
00:25:21,170 --> 00:25:24,740
It looks less bogus
over there.

458
00:25:24,740 --> 00:25:26,740
You can probably see the
whole rectangle.

459
00:25:26,740 --> 00:25:31,350
The white on white borders
are not there.

460
00:25:31,350 --> 00:25:34,480
There's simply is no physical
contour there.

461
00:25:34,480 --> 00:25:36,090
There may be here, because
the project's

462
00:25:36,090 --> 00:25:38,140
doing something mutant.

463
00:25:38,140 --> 00:25:41,380
But, there's certainly no
contour there, even though the

464
00:25:41,380 --> 00:25:44,400
rectangle at the center looks
somewhat brighter.

465
00:25:44,400 --> 00:25:48,580
Again, what you're doing is,
making a guess about what is

466
00:25:48,580 --> 00:25:52,720
it that actually created that
image that's landing on my

467
00:25:52,720 --> 00:25:53,700
retina now?

468
00:25:53,700 --> 00:25:57,840
It could be a little conference
of pac-people.

469
00:25:57,840 --> 00:26:01,010
Four little three quarter
circles that got together to

470
00:26:01,010 --> 00:26:02,210
talk to each other.

471
00:26:02,210 --> 00:26:05,810
But that doesn't seem the
most likely possibility.

472
00:26:05,810 --> 00:26:10,320
What seems more likely here, is
that it's a white rectangle

473
00:26:10,320 --> 00:26:16,250
sitting on top of four
black circles.

474
00:26:16,250 --> 00:26:21,040
And you end up completing
that contour.

475
00:26:21,040 --> 00:26:23,620
Or this circle for
that matter.

476
00:26:23,620 --> 00:26:26,710
What are you doing here?

477
00:26:26,710 --> 00:26:33,320
You don't need to have fancy
computer graphics to do this.

478
00:26:33,320 --> 00:26:37,500
One of the advantages of the
material in this particular

479
00:26:37,500 --> 00:26:44,460
lecture is that it provides
great material for doodling in

480
00:26:44,460 --> 00:26:47,460
other lectures.

481
00:26:47,460 --> 00:26:49,420
So, if you like subjective
contours, you

482
00:26:49,420 --> 00:26:50,690
can make your own.

483
00:26:53,320 --> 00:27:00,510
And, so how's that look?

484
00:27:00,510 --> 00:27:04,320
Got a subjective
contour there?

485
00:27:04,320 --> 00:27:05,760
It's not perfectly circular.

486
00:27:05,760 --> 00:27:07,300
That's OK.

487
00:27:07,300 --> 00:27:11,370
But what seems to happen is that
you generate a hypothesis

488
00:27:11,370 --> 00:27:15,710
that says, the whole problem I
got here, is contours don't

489
00:27:15,710 --> 00:27:17,610
just end in the world, they
tend to continue.

490
00:27:17,610 --> 00:27:21,640
Well, if this guy's continuing,
well, what

491
00:27:21,640 --> 00:27:22,350
happened here?

492
00:27:22,350 --> 00:27:22,530
Well.

493
00:27:22,530 --> 00:27:26,540
Maybe it ran into another
edge, it's being hidden.

494
00:27:26,540 --> 00:27:29,160
All else being equal, it
probably ran into an edge

495
00:27:29,160 --> 00:27:31,510
that's orthogonal to the
direction it's going, let's

496
00:27:31,510 --> 00:27:33,970
guess that.

497
00:27:33,970 --> 00:27:36,310
And, if we guess a whole bunch
of little orthogonal edges,

498
00:27:36,310 --> 00:27:40,030
we're back to that earlier
demonstration with a bunch of

499
00:27:40,030 --> 00:27:41,160
little line segments.

500
00:27:41,160 --> 00:27:43,930
I can tie those little
bits together.

501
00:27:43,930 --> 00:27:46,320
They make a kind of
a circle thing.

502
00:27:46,320 --> 00:27:50,970
So I end up seeing that
imaginary circle.

503
00:27:50,970 --> 00:27:53,350
And, in fact, I'm going to start
filling in the contour

504
00:27:53,350 --> 00:27:55,280
all the way around.

505
00:27:55,280 --> 00:27:58,360
That suggests, by the way,
that if I was to tilt all

506
00:27:58,360 --> 00:28:04,690
these lines a little bit off
of straight radial, so that

507
00:28:04,690 --> 00:28:08,880
the virtual line segment was
not forming a nice, neat

508
00:28:08,880 --> 00:28:12,110
circle, that the impression
of a subjective

509
00:28:12,110 --> 00:28:13,410
circle would get weaker.

510
00:28:13,410 --> 00:28:15,920
And you can decide whether
or not that's true here.

511
00:28:15,920 --> 00:28:18,600
So, see this circle?

512
00:28:18,600 --> 00:28:25,920
Now, the question is, we'll give
you three choices here.

513
00:28:25,920 --> 00:28:27,630
Stronger, weaker or
about the same?

514
00:28:27,630 --> 00:28:29,440
How vote that this one
is stronger than

515
00:28:29,440 --> 00:28:31,490
the previous one?

516
00:28:31,490 --> 00:28:34,380
How many vote for just
about the same?

517
00:28:34,380 --> 00:28:36,830
How many vote for weaker?

518
00:28:36,830 --> 00:28:38,695
Just about the same.

519
00:28:38,695 --> 00:28:39,680
That's a boring demo.

520
00:28:39,680 --> 00:28:43,260
I'll have to change
that next year.

521
00:28:43,260 --> 00:28:44,850
Because, it didn't
work well enough.

522
00:28:47,510 --> 00:28:50,000
OK.

523
00:28:50,000 --> 00:28:53,060
Another example.

524
00:28:53,060 --> 00:28:57,030
If your following along on the
notes, we have now gotten to

525
00:28:57,030 --> 00:28:59,690
the so-called
Craik-O'Brien-Cornsweet

526
00:28:59,690 --> 00:29:05,980
illusion, named after Craik,
O'Brien and Cornsweet. suites.

527
00:29:05,980 --> 00:29:14,470
We're continuing the edge
business, but, now, what I

528
00:29:14,470 --> 00:29:17,880
want to do is tie that into that
topic that I mentioned

529
00:29:17,880 --> 00:29:23,970
early in the lecture, about how
it is that, what you're

530
00:29:23,970 --> 00:29:26,550
interested in is things the
surface properties in the

531
00:29:26,550 --> 00:29:28,960
world, and you're not interested
in lighting.

532
00:29:28,960 --> 00:29:31,170
Lighting is very boring.

533
00:29:31,170 --> 00:29:33,610
So, what do you see here?

534
00:29:36,890 --> 00:29:46,000
Some bold soul described this
complicated image, boy, slow

535
00:29:46,000 --> 00:29:50,750
group, gray thing.

536
00:29:50,750 --> 00:29:52,000
Two gray things.

537
00:29:56,600 --> 00:29:57,540
One is darker.

538
00:29:57,540 --> 00:30:01,800
Oh boy, I'm going to bring
my pliers next time.

539
00:30:01,800 --> 00:30:02,740
This is like pulling teeth.

540
00:30:02,740 --> 00:30:03,630
Yes.

541
00:30:03,630 --> 00:30:09,023
AUDIENCE: It's like the top of
the pyramid, and there's

542
00:30:09,023 --> 00:30:11,880
shading on the other side.

543
00:30:11,880 --> 00:30:12,580
PROFESSOR: Oh.

544
00:30:12,580 --> 00:30:13,070
Yeah.

545
00:30:13,070 --> 00:30:15,580
Ok, so the light's coming from
the left, or something.

546
00:30:15,580 --> 00:30:16,020
OK.

547
00:30:16,020 --> 00:30:16,220
Yeah.

548
00:30:16,220 --> 00:30:20,680
That's a complicated inference
about these.

549
00:30:20,680 --> 00:30:20,970
OK.

550
00:30:20,970 --> 00:30:25,440
But what you don't particularly
see, is this.

551
00:30:25,440 --> 00:30:30,110
If I take the edge out, if I
take that edge away from the

552
00:30:30,110 --> 00:30:32,300
middle here, what you discover
is the whole

553
00:30:32,300 --> 00:30:35,970
thing is the same gray.

554
00:30:35,970 --> 00:30:40,180
That's not obvious.

555
00:30:40,180 --> 00:30:43,370
If you were to draw the
luminance profile, drag a

556
00:30:43,370 --> 00:30:47,370
photo detector across this
thing, what you would get is

557
00:30:47,370 --> 00:30:49,050
something like.

558
00:30:49,050 --> 00:30:49,930
Which side is bright?

559
00:30:49,930 --> 00:30:50,190
OK.

560
00:30:50,190 --> 00:30:50,980
That side bright.

561
00:30:50,980 --> 00:30:55,790
So it rises, then drops across
the edge, and then

562
00:30:55,790 --> 00:30:58,540
rises back like that.

563
00:30:58,540 --> 00:31:02,140
So, that if you take out
the actual edge, it's

564
00:31:02,140 --> 00:31:05,040
equal on the two sides.

565
00:31:05,040 --> 00:31:07,110
So, that's kind of weird.

566
00:31:07,110 --> 00:31:14,750
Why does it look like, now,
they didn't believe me.

567
00:31:14,750 --> 00:31:17,820
They thought I was doing
something evil here.

568
00:31:23,130 --> 00:31:23,600
Here we go.

569
00:31:23,600 --> 00:31:23,830
Look.

570
00:31:23,830 --> 00:31:27,210
That's why I had this thing
slowly sneak up, so you could,

571
00:31:27,210 --> 00:31:32,490
well, semi slowly,
fast-ly sneak up.

572
00:31:32,490 --> 00:31:37,190
What's going on here is that
the visual system knows

573
00:31:37,190 --> 00:31:40,490
something about edges
and about lighting.

574
00:31:40,490 --> 00:31:46,050
Edges in the world tend
to be fairly abrupt.

575
00:31:46,050 --> 00:31:50,150
Lightning changes, so it's
bright here, and dimmer over

576
00:31:50,150 --> 00:31:55,570
here, lighting changes tend
to be fairly gradual.

577
00:31:55,570 --> 00:31:59,770
So, if what I'm interested in
is seeing what's on the

578
00:31:59,770 --> 00:32:08,220
surface, as opposed to seeing
the product of boring light

579
00:32:08,220 --> 00:32:09,980
and shade variations.

580
00:32:09,980 --> 00:32:18,510
What I might want to do is to
look for abrupt changes, and

581
00:32:18,510 --> 00:32:21,790
to, in effect, suppress
gradual changes.

582
00:32:21,790 --> 00:32:24,160
And that's what's
going on here.

583
00:32:27,380 --> 00:32:31,200
We might as well do an
entertaining second order

584
00:32:31,200 --> 00:32:32,650
effect here.

585
00:32:35,730 --> 00:32:38,050
Remember that negative
after image thing?

586
00:32:38,050 --> 00:32:42,100
You know, look at red, you see
green, and stuff like that?

587
00:32:42,100 --> 00:32:44,900
What I should be able to do
here, is produce a negative

588
00:32:44,900 --> 00:32:50,926
version of this affect, where
you'll end up seeing this side

589
00:32:50,926 --> 00:32:54,030
is dark and this side is light,
even though I'm not

590
00:32:54,030 --> 00:32:57,990
going to change anything
over here, over here.

591
00:32:57,990 --> 00:33:02,110
Stare at that center
line, right.

592
00:33:02,110 --> 00:33:05,870
Stare rigorously at the center
line, and keep staring.

593
00:33:05,870 --> 00:33:13,760
And then when I do that, so,
this is two illusions

594
00:33:13,760 --> 00:33:15,380
concentrates on top of that.

595
00:33:15,380 --> 00:33:18,250
Do it again, was that?

596
00:33:18,250 --> 00:33:18,680
All right.

597
00:33:18,680 --> 00:33:25,740
For the people incapable of
following instructions the

598
00:33:25,740 --> 00:33:27,510
first time, try it again.

599
00:33:27,510 --> 00:33:35,050
So stare at the center and
hold your fixation there.

600
00:33:35,050 --> 00:33:37,550
Actually, people who got it the
first time, keep staring

601
00:33:37,550 --> 00:33:39,440
at the center, but the people
who got it the first time, can

602
00:33:39,440 --> 00:33:41,440
try something tricky-er which
is move your fixation, a

603
00:33:41,440 --> 00:33:44,460
little bit, and you'll change
the proportion that looks

604
00:33:44,460 --> 00:33:45,710
light or dark.

605
00:33:49,240 --> 00:33:51,770
Isn't that fun?

606
00:33:51,770 --> 00:33:54,070
You should understand
why that works.

607
00:33:54,070 --> 00:33:57,400
If you don't, write yourself
a little note on your paper

608
00:33:57,400 --> 00:34:02,020
saying I don't understand why
that works and figure it out.

609
00:34:02,020 --> 00:34:02,620
All right.

610
00:34:02,620 --> 00:34:05,540
So the important thing is that
what you're trying to do here,

611
00:34:05,540 --> 00:34:09,280
is you're trying to get rid of
information about the light.

612
00:34:09,280 --> 00:34:11,670
You don't care about
the light levels.

613
00:34:11,670 --> 00:34:13,790
What you care about is what's
going on in the world.

614
00:34:13,790 --> 00:34:17,370
Now Ted Adelson in the brain and
cogs department here has

615
00:34:17,370 --> 00:34:20,430
exploited this fact brilliantly
in a variety of

616
00:34:20,430 --> 00:34:22,580
gorgeous demos.

617
00:34:22,580 --> 00:34:24,660
One of which is this.

618
00:34:24,660 --> 00:34:28,390
It is extremely difficult,
even if you've seen this

619
00:34:28,390 --> 00:34:32,600
before, to convince yourself
that the gray levels of A and

620
00:34:32,600 --> 00:34:34,610
B are identical.

621
00:34:34,610 --> 00:34:35,980
Which they are.

622
00:34:35,980 --> 00:34:38,650
In fact, it's so difficult that
even if I stick a bar

623
00:34:38,650 --> 00:34:42,560
across it that's clearly the
same thing, it's still kind of

624
00:34:42,560 --> 00:34:45,450
hard to see that as identical.

625
00:34:45,450 --> 00:34:48,760
Your brain wants to do all sorts
of things to deny that

626
00:34:48,760 --> 00:34:51,070
possibility.

627
00:34:51,070 --> 00:34:52,820
What's going on here?

628
00:34:52,820 --> 00:34:56,660
What's going on is that virtual
cylinder is casting a

629
00:34:56,660 --> 00:35:02,130
virtual shadow, that you
are busy discounting.

630
00:35:02,130 --> 00:35:05,290
You're saying, I don't care
about the shadow.

631
00:35:05,290 --> 00:35:07,370
There's a checker board.

632
00:35:07,370 --> 00:35:08,410
I know about checker boards.

633
00:35:08,410 --> 00:35:11,030
Checker boards to go dark
square, light square.

634
00:35:11,030 --> 00:35:12,290
I can figure this out.

635
00:35:12,290 --> 00:35:14,460
That means B is a light
square, because it's

636
00:35:14,460 --> 00:35:18,050
surrounded by dark squares, and
A is dark square because

637
00:35:18,050 --> 00:35:20,370
it's surrounded by
light squares.

638
00:35:20,370 --> 00:35:26,050
And, it's a particularly lovely
example of, among other

639
00:35:26,050 --> 00:35:31,580
things, this ability to get rid
of information about the

640
00:35:31,580 --> 00:35:32,400
illumination.

641
00:35:32,400 --> 00:35:35,270
This isn't to say that what
you do is somehow just run

642
00:35:35,270 --> 00:35:37,890
some sort of, throw away the
light source information, and

643
00:35:37,890 --> 00:35:41,660
don't do anything with it.

644
00:35:41,660 --> 00:35:42,910
Get back there.

645
00:35:45,230 --> 00:35:46,220
What's it say?

646
00:35:46,220 --> 00:35:47,470
Cow.

647
00:35:49,760 --> 00:35:53,200
There's only white and
black on that screen.

648
00:35:53,200 --> 00:35:59,480
Look at the C and ask where that
contour is coming from.

649
00:35:59,480 --> 00:36:03,640
That contour, particularly the
outside of that C, has

650
00:36:03,640 --> 00:36:06,970
extremely little support
in the image.

651
00:36:06,970 --> 00:36:09,900
What you are doing is
making an inference.

652
00:36:09,900 --> 00:36:12,650
This time, you're using the
shadow information.

653
00:36:12,650 --> 00:36:16,120
You don't want to
see the shadow.

654
00:36:16,120 --> 00:36:18,990
How many people see it not
only as cow, also as an

655
00:36:18,990 --> 00:36:22,540
embossed word cow, sticking
out a little bit?

656
00:36:22,540 --> 00:36:25,210
The reason you're seeing the cow
at all, is you're assuming

657
00:36:25,210 --> 00:36:27,130
that those black things
are shadows.

658
00:36:27,130 --> 00:36:30,100
If those black things are
shadows, it follows the light

659
00:36:30,100 --> 00:36:31,810
is coming from the upper left.

660
00:36:31,810 --> 00:36:35,030
If the light is coming from the
upper left, well, we can

661
00:36:35,030 --> 00:36:41,280
go and figure out what shape
object must have been

662
00:36:41,280 --> 00:36:46,560
producing those shadows, and
the answer spells cow.

663
00:36:46,560 --> 00:36:48,990
But, if you were to take
a look at the C,

664
00:36:48,990 --> 00:36:50,560
is about the clearest.

665
00:36:50,560 --> 00:36:53,090
Well, I don't know, the O is
pretty good too, and so the W,

666
00:36:53,090 --> 00:36:53,920
for that matter.

667
00:36:53,920 --> 00:36:56,740
Anyway, looking any of those
guys, and look at the shapes

668
00:36:56,740 --> 00:37:00,040
of the black bits, which is the
only thing that stands out

669
00:37:00,040 --> 00:37:01,070
from the background,
of course.

670
00:37:01,070 --> 00:37:02,900
There's nothing else there,
except the black bits on a

671
00:37:02,900 --> 00:37:04,450
white background.

672
00:37:04,450 --> 00:37:11,150
None of those bits say C or O
or W. It's a construction

673
00:37:11,150 --> 00:37:14,930
based on the assumption that
the black bits are shadows.

674
00:37:14,930 --> 00:37:20,310
You use that sort of information
all the time to do

675
00:37:20,310 --> 00:37:22,660
things like see faces.

676
00:37:22,660 --> 00:37:24,990
These are so called Mooney
faces, named

677
00:37:24,990 --> 00:37:26,950
after a guy named Mooney.

678
00:37:30,640 --> 00:37:36,370
You can tell me a lot about the
curvature of these faces,

679
00:37:36,370 --> 00:37:39,610
even though, again, there's
nothing on the screen except

680
00:37:39,610 --> 00:37:42,140
for black regions and
white regions.

681
00:37:42,140 --> 00:37:47,040
None of which are, themselves,
particularly faced shape.

682
00:37:47,040 --> 00:37:49,060
I mean look at the eye.

683
00:37:49,060 --> 00:37:51,860
Are any of those eyes actually
shaped -- look at the guy on

684
00:37:51,860 --> 00:37:53,070
the right--

685
00:37:53,070 --> 00:37:58,720
I mean his eye, he's only got
one apparently, that eye looks

686
00:37:58,720 --> 00:38:03,180
like, I don't know, a mutant
bunny or something.

687
00:38:03,180 --> 00:38:10,810
If I just presented the eye
piece on the on the guy on the

688
00:38:10,810 --> 00:38:14,160
right, that black glob that's
defining his eye, if I just

689
00:38:14,160 --> 00:38:16,970
presented that in isolation,
nobody would

690
00:38:16,970 --> 00:38:18,430
say, oh yeah, sure.

691
00:38:18,430 --> 00:38:20,180
That's an eye.

692
00:38:20,180 --> 00:38:23,350
You'd all be saying dalmatian
dog before, mutant bunny this

693
00:38:23,350 --> 00:38:25,670
time, or something.

694
00:38:25,670 --> 00:38:32,230
And, it relies on these
assumptions about shadow, is

695
00:38:32,230 --> 00:38:33,640
what is what you're
doing here.

696
00:38:36,510 --> 00:38:40,730
It survives inversion
reasonably well.

697
00:38:40,730 --> 00:38:46,630
But, those don't look like
faces very much.

698
00:38:46,630 --> 00:38:50,050
What went wrong?

699
00:38:50,050 --> 00:38:53,690
You might think, I know that
shadows aren't red and blue.

700
00:38:53,690 --> 00:38:55,110
But that's actually
not the problem.

701
00:38:55,110 --> 00:38:57,170
Here they look pretty
good, right.

702
00:38:57,170 --> 00:39:02,780
Those faces look ok, but
these faces look lousy.

703
00:39:02,780 --> 00:39:04,230
Why do they look lousy?

704
00:39:07,590 --> 00:39:08,850
Somebody raise a hand,
or something,

705
00:39:08,850 --> 00:39:09,650
yeah there's a theory.

706
00:39:09,650 --> 00:39:11,280
AUDIENCE: The shadows are lighte
than the [INAUDIBLE]

707
00:39:11,280 --> 00:39:13,760
PROFESSOR: Yes, if you make
the shadow regions lighter

708
00:39:13,760 --> 00:39:16,990
than the lit regions,
the brain says,

709
00:39:16,990 --> 00:39:19,080
that's not a shadow.

710
00:39:19,080 --> 00:39:20,590
I don't care about shadows.

711
00:39:20,590 --> 00:39:24,190
I don't want to see shadows as
entities in their own right

712
00:39:24,190 --> 00:39:25,720
particularly, most
of the time.

713
00:39:25,720 --> 00:39:27,980
But I'll tell you one thing
I know about shadows.

714
00:39:27,980 --> 00:39:30,130
Shadows are darker than
the other stuff.

715
00:39:30,130 --> 00:39:31,400
If the shadows are lighter
than the other

716
00:39:31,400 --> 00:39:33,480
stuff, it's not shadow.

717
00:39:33,480 --> 00:39:37,430
It's something else, and
something weird.

718
00:39:37,430 --> 00:39:38,740
So this doesn't work.

719
00:39:38,740 --> 00:39:39,800
But that works.

720
00:39:39,800 --> 00:39:42,030
Oh, I suppose the fact that it
said on this one, shadow must

721
00:39:42,030 --> 00:39:44,040
be darker, might have tipped
some people off.

722
00:39:47,090 --> 00:39:48,020
OK.

723
00:39:48,020 --> 00:39:50,070
Let's see here.

724
00:39:50,070 --> 00:39:54,620
I think what I will do -- this
makes it very natural break

725
00:39:54,620 --> 00:39:57,970
point -- where it says Mooney
face and we'll go on to this

726
00:39:57,970 --> 00:40:00,280
question about making the best
guess you can make in the

727
00:40:00,280 --> 00:40:03,570
context of going
from 2D to 3D.

728
00:40:03,570 --> 00:40:08,790
D But before we go on to that,
let's us lets take our brief,

729
00:40:08,790 --> 00:40:11,260
stretch your limbs kind
of break, here.

730
00:42:03,370 --> 00:42:03,970
OK.

731
00:42:03,970 --> 00:42:07,490
Let us gather back
together here.

732
00:42:11,830 --> 00:42:14,460
By the way, it looks like it's
getting to be that time in the

733
00:42:14,460 --> 00:42:24,550
term, where people are abusing
their natural sleep mechanism

734
00:42:24,550 --> 00:42:27,860
that I will talk about
later on in the term.

735
00:42:27,860 --> 00:42:33,200
But looking around at this
crowd, I would say that you're

736
00:42:33,200 --> 00:42:39,200
not getting the seven
to eight you need.

737
00:42:39,200 --> 00:42:47,610
Or, if you are, maybe you really
need ten and you're

738
00:42:47,610 --> 00:42:49,900
catching the extra to 2 here.

739
00:42:57,390 --> 00:43:01,720
I've been talking about sort of
little, almost like atomic

740
00:43:01,720 --> 00:43:05,320
small scale examples,
of these sort of

741
00:43:05,320 --> 00:43:06,840
inferences that you make.

742
00:43:06,840 --> 00:43:10,840
And, now what I want to do, is
sort of head for the larger

743
00:43:10,840 --> 00:43:12,530
picture of how you
make an inference

744
00:43:12,530 --> 00:43:14,630
about the whole scene.

745
00:43:14,630 --> 00:43:19,960
I'm not going to get all the way
there, and there are many

746
00:43:19,960 --> 00:43:21,440
realm I could talk
about this in.

747
00:43:21,440 --> 00:43:24,270
So I'm going to talk about it in
one restricted area, which

748
00:43:24,270 --> 00:43:29,600
is this question of going from
a 2D image to 3D inferences

749
00:43:29,600 --> 00:43:30,710
about the world.

750
00:43:30,710 --> 00:43:35,770
Something that you do
automatically, all the time.

751
00:43:35,770 --> 00:43:38,000
I want to explain a bit
about how you do it.

752
00:43:38,000 --> 00:43:42,070
You will see, on your hand out,
this is very useless,

753
00:43:42,070 --> 00:43:45,600
blank region, that says one
two, three, four, five.

754
00:43:45,600 --> 00:43:48,070
I'm going to go through a
series of depth cues.

755
00:43:48,070 --> 00:43:50,970
Probably more than
five of them.

756
00:43:50,970 --> 00:43:52,890
And, that's what's supposed
to go in there.

757
00:43:52,890 --> 00:43:56,000
They're all lots of different
sources of information that

758
00:43:56,000 --> 00:44:03,170
you use to go from the 2D world
to the 3D world, many of

759
00:44:03,170 --> 00:44:06,050
them seen here in this lovely
piece of renaissance art that

760
00:44:06,050 --> 00:44:07,900
we will come up back to.

761
00:44:07,900 --> 00:44:11,360
But, here's a much more
boring piece of art.

762
00:44:11,360 --> 00:44:12,610
What do you see?

763
00:44:15,530 --> 00:44:16,900
Circle, square and a diamond.

764
00:44:16,900 --> 00:44:19,310
And, then there's the clever
person, who's trying to figure

765
00:44:19,310 --> 00:44:21,030
out, I've describe
them as pac-man

766
00:44:21,030 --> 00:44:22,540
before, but these aren't.

767
00:44:22,540 --> 00:44:25,310
But, you see a circle, square,
and a diamond.

768
00:44:25,310 --> 00:44:29,100
You don't have any serious
difficulty inferring that --

769
00:44:29,100 --> 00:44:30,860
triangle, sorry.

770
00:44:30,860 --> 00:44:33,050
You don't see this.

771
00:44:33,050 --> 00:44:37,240
For present purposes, the
important point here is, you

772
00:44:37,240 --> 00:44:41,190
can also tell me their
depth order.

773
00:44:41,190 --> 00:44:44,400
It's in some sense so obvious
that you never think about it,

774
00:44:44,400 --> 00:44:49,710
but it's a very important source
of information about

775
00:44:49,710 --> 00:44:54,040
depth order, that you get simply
because you know that

776
00:44:54,040 --> 00:44:55,950
solid object occlude
each other.

777
00:44:55,950 --> 00:44:59,850
So you firmly believe
that I am standing

778
00:44:59,850 --> 00:45:02,000
in front the screen.

779
00:45:02,000 --> 00:45:05,130
I could had, well no, I
couldn't have, it'd be

780
00:45:05,130 --> 00:45:09,520
theoretically possible that I
had suddenly cut a cunningly

781
00:45:09,520 --> 00:45:16,050
wolf shaped hole in the screen
and I'm A, very large, and B,

782
00:45:16,050 --> 00:45:20,350
standing over towards east
campus somewhere.

783
00:45:20,350 --> 00:45:23,920
And, you're looking at me
through this set of holes.

784
00:45:23,920 --> 00:45:27,910
No.

785
00:45:27,910 --> 00:45:32,080
You automatically leap to the
assumption that if A looks

786
00:45:32,080 --> 00:45:37,040
like it's occluding B,
A in front of B.

787
00:45:37,040 --> 00:45:38,210
Ah, my bunnies.

788
00:45:38,210 --> 00:45:39,340
I don't know what happened
to them.

789
00:45:39,340 --> 00:45:40,540
They got kind of pixellated.

790
00:45:40,540 --> 00:45:50,780
But, all right, which bunnies
are closer to you?

791
00:45:50,780 --> 00:45:53,530
The big bunnies are
closer to you.

792
00:45:53,530 --> 00:45:54,030
Right.

793
00:45:54,030 --> 00:45:56,170
Why do you think the big bunnies
are closer to you?

794
00:45:56,170 --> 00:45:58,690
Because, you're making an
inference that bunnies are

795
00:45:58,690 --> 00:46:01,810
roughly bunny sized.

796
00:46:01,810 --> 00:46:07,370
And, in the same way, I'm
currently making the

797
00:46:07,370 --> 00:46:11,370
assumption that you guys are all
more or less people sized.

798
00:46:11,370 --> 00:46:14,400
If I did not make that
assumption, I would come to

799
00:46:14,400 --> 00:46:18,030
some very odd inferences about
the current view that I'm

800
00:46:18,030 --> 00:46:19,050
looking at.

801
00:46:19,050 --> 00:46:22,890
So, the people in the front row
-- there's a person in the

802
00:46:22,890 --> 00:46:26,560
front row -- her head is about
two degrees of visual angle.

803
00:46:26,560 --> 00:46:31,840
Remember 360 degrees around my
head each degree it's about my

804
00:46:31,840 --> 00:46:34,540
thumb, so her head takes
up about two degrees.

805
00:46:34,540 --> 00:46:38,840
And, let's see, there's this
guy in the cheap seats back

806
00:46:38,840 --> 00:46:44,020
there, his head is only
about half a degree.

807
00:46:44,020 --> 00:46:48,020
I could make the assumption
that he's a pinhead.

808
00:46:48,020 --> 00:46:49,910
A guy with a real small
had sitting out the

809
00:46:49,910 --> 00:46:51,530
deep back out there.

810
00:46:51,530 --> 00:46:53,190
Well actually, I wouldn't make
the assumption that he was

811
00:46:53,190 --> 00:46:54,170
sitting in the back.

812
00:46:54,170 --> 00:46:57,780
He's a pin head sitting at the
same distance as large head

813
00:46:57,780 --> 00:47:00,820
woman here in the front.

814
00:47:00,820 --> 00:47:02,340
But that's dumb.

815
00:47:02,340 --> 00:47:02,640
Right?

816
00:47:02,640 --> 00:47:04,510
Your visual system knows
that's dumb.

817
00:47:04,510 --> 00:47:05,940
Your visual system
knows people are

818
00:47:05,940 --> 00:47:06,870
roughly people's sized.

819
00:47:06,870 --> 00:47:09,660
Not exactly people sized, but
roughly people sized, and if I

820
00:47:09,660 --> 00:47:11,890
see a bunch of small things
there and a bunch of big

821
00:47:11,890 --> 00:47:15,200
things here, odds are that
this is closer than that.

822
00:47:15,200 --> 00:47:18,320
And that's part of what's giving
me my current inference

823
00:47:18,320 --> 00:47:21,430
that I'm looking at a tilted
plane of people in purple

824
00:47:21,430 --> 00:47:25,940
seats is this information
about size.

825
00:47:25,940 --> 00:47:28,820
If I organize the bunnies the
way you guys are organized, I

826
00:47:28,820 --> 00:47:31,500
got a much clearer sensation
of depth from

827
00:47:31,500 --> 00:47:32,990
this texture radiant.

828
00:47:32,990 --> 00:47:36,560
So, now you should be able to
see sort of tilted rabbit

829
00:47:36,560 --> 00:47:38,140
plane, right?

830
00:47:38,140 --> 00:47:40,330
Even though you know
objectively, it's just sitting

831
00:47:40,330 --> 00:47:42,280
flat on the screen,
it looks tilted.

832
00:47:42,280 --> 00:47:43,830
Is that a hand up there?

833
00:47:43,830 --> 00:47:45,080
That was a hand.

834
00:47:54,860 --> 00:47:56,860
Sorry I missed that.

835
00:47:56,860 --> 00:47:57,720
My previous image.

836
00:47:57,720 --> 00:48:00,458
We can do that.

837
00:48:00,458 --> 00:48:01,708
AUDIENCE: [INAUDIBLE]

838
00:48:07,430 --> 00:48:07,760
PROFESSOR: Yeah.

839
00:48:07,760 --> 00:48:09,380
That's another possibility.

840
00:48:09,380 --> 00:48:13,280
It could would be that.

841
00:48:13,280 --> 00:48:18,450
And, in fact, it is,
just a flat image.

842
00:48:18,450 --> 00:48:20,200
And you actually sound like
you're getting a sort of a

843
00:48:20,200 --> 00:48:24,830
hybrid of a tilted plane with
bunnies of a range of sizes.

844
00:48:24,830 --> 00:48:28,140
Actually, we can see that
combination here.

845
00:48:28,140 --> 00:48:31,810
So, we got a whole bunch
of big bunnies,

846
00:48:31,810 --> 00:48:36,720
and two little bunnies.

847
00:48:36,720 --> 00:48:41,560
Which is the smallest bunny
in this display?

848
00:48:41,560 --> 00:48:43,890
The bottom right bunny, right?

849
00:48:43,890 --> 00:48:47,420
These guys are identical
in size.

850
00:48:47,420 --> 00:48:48,850
It's a very minimal display.

851
00:48:48,850 --> 00:48:51,100
There's a much more vivid
version of this illusion in

852
00:48:51,100 --> 00:48:53,260
the book as I recall.

853
00:48:53,260 --> 00:48:58,070
It's a very minimal version
of the illusion.

854
00:48:58,070 --> 00:49:03,984
Because you assume,
if you are --

855
00:49:08,700 --> 00:49:09,620
All right.

856
00:49:09,620 --> 00:49:12,520
So I still can't draw. --

857
00:49:12,520 --> 00:49:16,770
If I'm looking at two bunnies,
if I'm looking at a ground

858
00:49:16,770 --> 00:49:26,300
plane, closer is also lower
in the visual field.

859
00:49:26,300 --> 00:49:28,660
So, you make an automatic
assumption that was what was

860
00:49:28,660 --> 00:49:32,050
giving this woman that notion
of a fairly tilted plane in

861
00:49:32,050 --> 00:49:34,160
the first bunny example.

862
00:49:34,160 --> 00:49:38,200
You make the assumption that
the bottom of the image is

863
00:49:38,200 --> 00:49:40,730
closer to you then the
top of the image.

864
00:49:40,730 --> 00:49:42,390
Well, if the bottom
of the image --

865
00:49:42,390 --> 00:49:45,260
let's go a back here--

866
00:49:45,260 --> 00:49:52,050
if the bottom of the image is
closer then the top, and these

867
00:49:52,050 --> 00:49:55,820
two bunny images are the same
size, if this guy's closer, it

868
00:49:55,820 --> 00:49:57,290
must be really small.

869
00:50:00,000 --> 00:50:03,530
Suppose I took the guy from
the back row here.

870
00:50:03,530 --> 00:50:08,050
His whole upper body fills sort
of the top joint of my

871
00:50:08,050 --> 00:50:09,900
thumb in visual angle terms.

872
00:50:09,900 --> 00:50:13,970
If I moved that image to the
front, and sat him in a seat

873
00:50:13,970 --> 00:50:18,330
here, he would be about
the size of this

874
00:50:18,330 --> 00:50:20,280
woman's upper arms.

875
00:50:20,280 --> 00:50:23,160
And, I would think, that's
a really small guy.

876
00:50:25,850 --> 00:50:29,360
She could wear him on her
--instead of wearing your

877
00:50:29,360 --> 00:50:31,610
heart on your sleeve, you could
wear the whole guy on

878
00:50:31,610 --> 00:50:35,260
your sleeve.

879
00:50:35,260 --> 00:50:36,600
So, that's what's
going on here.

880
00:50:36,600 --> 00:50:40,590
You're making the assumption
that small bunny one is closer

881
00:50:40,590 --> 00:50:41,990
than small bunny two.

882
00:50:41,990 --> 00:50:43,710
They're the same image size.

883
00:50:43,710 --> 00:50:46,490
You therefore infer that out in
the world, this must be a

884
00:50:46,490 --> 00:50:49,090
really small bunny, and
that one is just a

885
00:50:49,090 --> 00:50:50,310
reasonably small bunny.

886
00:50:50,310 --> 00:50:54,680
Now the bunny part turns out
to be not that critical.

887
00:50:54,680 --> 00:51:00,200
Here you can also see a nice
plane going off into the

888
00:51:00,200 --> 00:51:04,290
distance with objects that are
clearly not meaningful.

889
00:51:04,290 --> 00:51:04,750
Right.

890
00:51:04,750 --> 00:51:06,590
Big stuff, front and low.

891
00:51:06,590 --> 00:51:13,390
Small stuff high in the image
and, you simply infer this is

892
00:51:13,390 --> 00:51:16,530
close, and stuff up there is
far, and you get a nice

893
00:51:16,530 --> 00:51:19,250
impression of a tilted plane.

894
00:51:21,910 --> 00:51:28,060
This is one of the cues that
is interestingly variable

895
00:51:28,060 --> 00:51:29,640
depending on where
you're from.

896
00:51:33,330 --> 00:51:36,690
The atmosphere scatters light,
particularly water in the

897
00:51:36,690 --> 00:51:39,740
atmosphere scatters light,
that's why the sky is blue.

898
00:51:39,740 --> 00:51:42,650
The result is that objects that
are far away tend to be

899
00:51:42,650 --> 00:51:44,600
both hazier and bluer.

900
00:51:44,600 --> 00:51:49,000
Something that you can see in
all sorts of works of art.

901
00:51:49,000 --> 00:51:51,810
Go to the museum, you can see
artists take advantage of this

902
00:51:51,810 --> 00:51:54,050
right, left and center.

903
00:51:54,050 --> 00:51:59,420
And, you can probably get even
in my pathetically reduced

904
00:51:59,420 --> 00:52:00,190
version of it.

905
00:52:00,190 --> 00:52:02,830
You probably get a sensation
of depth here that

906
00:52:02,830 --> 00:52:04,770
you don't get here.

907
00:52:04,770 --> 00:52:07,740
The geographic aspect
of it, --

908
00:52:07,740 --> 00:52:09,870
anybody here from Arizona?

909
00:52:09,870 --> 00:52:12,020
Doesn't work well in Arizona.

910
00:52:12,020 --> 00:52:16,610
I know this because I went to a
meeting in Tucson and it was

911
00:52:16,610 --> 00:52:19,450
boring, so I went out for a
walk, and I saw this hill, and

912
00:52:19,450 --> 00:52:22,640
I said to the guy who was at the
street corner with me, how

913
00:52:22,640 --> 00:52:25,690
long would it take me to
walk to that hill?

914
00:52:25,690 --> 00:52:29,770
And he said, three
days, four days?

915
00:52:32,520 --> 00:52:37,670
It's like 50 miles away and
it's like 6000 feet high.

916
00:52:37,670 --> 00:52:41,380
But it was extremely crisp.

917
00:52:41,380 --> 00:52:44,010
And, there's no water
in Arizona.

918
00:52:44,010 --> 00:52:45,650
I don't know why people
live there.

919
00:52:45,650 --> 00:52:47,000
It's like hot all the time.

920
00:52:47,000 --> 00:52:48,960
And there's no water.

921
00:52:48,960 --> 00:52:54,410
But anyway, aerial perspective
cues don't work, so not only

922
00:52:54,410 --> 00:52:58,520
are you thirsty, put your
short one depth cue.

923
00:52:58,520 --> 00:53:01,380
Anyway, this works much better
in a humid setting that in a

924
00:53:01,380 --> 00:53:04,110
non humid setting, but
the point is that

925
00:53:04,110 --> 00:53:07,290
again you know about--

926
00:53:07,290 --> 00:53:09,780
this is your visual system
making use of the physics of

927
00:53:09,780 --> 00:53:12,510
the situation in order to infer
something about the

928
00:53:12,510 --> 00:53:13,810
depth of the situation.

929
00:53:16,660 --> 00:53:20,440
You also know about the
geometry of the world.

930
00:53:20,440 --> 00:53:25,050
So, this is an extremely
limited picture.

931
00:53:25,050 --> 00:53:30,410
If I say, this is a highway
going off to infinity

932
00:53:30,410 --> 00:53:34,725
somewhere in Arizona, or
something like that, that's

933
00:53:34,725 --> 00:53:36,260
not a great picture.

934
00:53:36,260 --> 00:53:37,380
But you can believe that.

935
00:53:37,380 --> 00:53:43,190
Because you know implicitly
that parallel lines in the

936
00:53:43,190 --> 00:53:47,820
world, if they're in depth,
will look like they are

937
00:53:47,820 --> 00:53:51,010
converging towards a vanishing
point somewhere.

938
00:53:51,010 --> 00:53:54,680
Now it is sometimes claimed that
this is known as linear

939
00:53:54,680 --> 00:53:59,620
perspective, that linear
perspective was discovered by

940
00:53:59,620 --> 00:54:02,100
artists during the
renaissance.

941
00:54:02,100 --> 00:54:03,460
That's only sort of true.

942
00:54:03,460 --> 00:54:06,100
What happened in the
renaissance, was that they

943
00:54:06,100 --> 00:54:10,260
made this knowledge explicit,
and became able to use it for

944
00:54:10,260 --> 00:54:13,020
instance to make their
art works.

945
00:54:13,020 --> 00:54:18,240
But, your cat and your lizard
and stuff know about linear

946
00:54:18,240 --> 00:54:19,000
perspective.

947
00:54:19,000 --> 00:54:20,380
They just know it implicitly.

948
00:54:20,380 --> 00:54:26,270
The same way they knew about
arial perspective and size

949
00:54:26,270 --> 00:54:29,220
clues and occlusion cues,
and things like that.

950
00:54:29,220 --> 00:54:33,880
It wasn't that we woke up one
day in Renaissance, Italy and

951
00:54:33,880 --> 00:54:36,690
suddenly we could use
this depth cue.

952
00:54:36,690 --> 00:54:41,360
What we figured out was how to
paint with this depth cue.

953
00:54:41,360 --> 00:54:43,070
And you could do all sorts
of amusing things

954
00:54:43,070 --> 00:54:44,230
with the depth cue.

955
00:54:44,230 --> 00:54:46,920
So, for instance.

956
00:54:46,920 --> 00:54:49,280
Lines look more or less
the same size, right.

957
00:54:54,940 --> 00:54:58,360
Let me see if I can
change that here.

958
00:55:13,710 --> 00:55:14,310
All right.

959
00:55:14,310 --> 00:55:21,290
Even though we're using crude
materials, let's do a forced

960
00:55:21,290 --> 00:55:23,210
choice vote here.

961
00:55:23,210 --> 00:55:26,280
If you know they're at the same
size, so it's boring to

962
00:55:26,280 --> 00:55:27,470
ask if they're the same size.

963
00:55:27,470 --> 00:55:31,790
But if you had to vote bigger
or smaller, how many people

964
00:55:31,790 --> 00:55:36,470
would vote that the bottom
line now looks bigger?

965
00:55:36,470 --> 00:55:39,960
How many would vote that the
bottom line now look smaller.

966
00:55:39,960 --> 00:55:43,240
Well, I guess that worked,
cheap chalk and all.

967
00:55:43,240 --> 00:55:49,090
This is an illusion known
as the Ponzo illusion.

968
00:55:49,090 --> 00:55:53,200
There are a number of ways to
account for it, but one of the

969
00:55:53,200 --> 00:55:56,980
intuitively appealing ones, at
least, is to say, what this is

970
00:55:56,980 --> 00:56:00,670
doing, even though you're not
particularly seeing it as a

971
00:56:00,670 --> 00:56:03,130
depth cue, is it's telling the
chunks of your brain that are

972
00:56:03,130 --> 00:56:09,040
trying to figure out 3D, I see
these two converging lines.

973
00:56:09,040 --> 00:56:11,250
If they're parallel lines in the
world, they must be going

974
00:56:11,250 --> 00:56:12,210
off into depth.

975
00:56:12,210 --> 00:56:14,810
If they're going off into depth,
then this thing is

976
00:56:14,810 --> 00:56:17,500
further away than this one.

977
00:56:17,500 --> 00:56:19,410
Well, if this one is further
away, and they are the same

978
00:56:19,410 --> 00:56:21,960
size on my retina, this
must be bigger.

979
00:56:21,960 --> 00:56:25,360
If that's not intuitively
obvious to you, think about

980
00:56:25,360 --> 00:56:28,530
this as train tracks.

981
00:56:28,530 --> 00:56:32,040
So here are train tracks going
off into the distance, and ask

982
00:56:32,040 --> 00:56:40,830
yourself, which maiden here,
tied of to the tracks is in

983
00:56:40,830 --> 00:56:43,370
more distress?

984
00:56:43,370 --> 00:56:47,550
It's obvious that this must
be the bigger person if we

985
00:56:47,550 --> 00:56:51,770
interpret this as train tracks
going off into the distance.

986
00:56:51,770 --> 00:56:52,650
Even though we know
that they're

987
00:56:52,650 --> 00:56:55,260
essentially the same size.

988
00:56:55,260 --> 00:56:58,470
You're getting the results
of the inference are then

989
00:56:58,470 --> 00:57:02,580
influencing what you see.

990
00:57:02,580 --> 00:57:04,980
They are influencing other
inferences that you

991
00:57:04,980 --> 00:57:08,370
make about the image.

992
00:57:08,370 --> 00:57:13,380
Now you can exploit these rules
of linear perspective in

993
00:57:13,380 --> 00:57:18,250
much more elaborate
fashion then that.

994
00:57:18,250 --> 00:57:24,570
And the great master of
that game is Escher.

995
00:57:24,570 --> 00:57:29,560
Here is one of Escher's
pictures.

996
00:57:29,560 --> 00:57:33,620
What you want to do again, I
think the image looks rather

997
00:57:33,620 --> 00:57:37,600
sharper up on these sides guys,
but ask yourself where

998
00:57:37,600 --> 00:57:40,480
the vanishing point is.

999
00:57:40,480 --> 00:57:44,260
So, look at the first floor
of that structure.

1000
00:57:44,260 --> 00:57:46,880
And, it's pretty clear that
those railings are converging

1001
00:57:46,880 --> 00:57:49,640
to a vanishing point off
to the right somewhere.

1002
00:57:49,640 --> 00:57:52,060
Well, now look at
the top floor.

1003
00:57:52,060 --> 00:57:54,310
That's converting to
vanishing point

1004
00:57:54,310 --> 00:57:57,600
off to the left somewhere.

1005
00:57:57,600 --> 00:58:00,600
And, then when you tried it put
the whole thing together,

1006
00:58:00,600 --> 00:58:03,960
you've got a structure the
doesn't quite hang together

1007
00:58:03,960 --> 00:58:05,070
quite right.

1008
00:58:05,070 --> 00:58:06,720
This tells you a couple
of things.

1009
00:58:06,720 --> 00:58:10,590
Thing one it tells you Escher
was a very clever draftsman.

1010
00:58:10,590 --> 00:58:14,140
Thing two that it tells you is
that you do a lot of these

1011
00:58:14,140 --> 00:58:18,600
calculations about perspective
very locally.

1012
00:58:18,600 --> 00:58:23,800
What you do is you say, your
attending, let's say, to the

1013
00:58:23,800 --> 00:58:26,740
lower floor there, and you say,
yeah, this all adds up.

1014
00:58:26,740 --> 00:58:28,200
It makes sense.

1015
00:58:28,200 --> 00:58:30,410
And, then you direct your
attention to the upper floor.

1016
00:58:30,410 --> 00:58:31,630
And it all adds up.

1017
00:58:31,630 --> 00:58:32,560
It makes sense.

1018
00:58:32,560 --> 00:58:35,880
It's only when you try to
combine all of that across

1019
00:58:35,880 --> 00:58:38,630
whole image, that you realize
that the whole image somehow

1020
00:58:38,630 --> 00:58:40,380
doesn't quite make sense.

1021
00:58:40,380 --> 00:58:43,090
And, that's how Escher can do
things like, have staircases

1022
00:58:43,090 --> 00:58:48,380
that always go up, and water
falls that fall apparently in

1023
00:58:48,380 --> 00:58:50,820
an infinite loop and
things like that.

1024
00:58:50,820 --> 00:58:53,840
Grab yourself your favorite
Escher website and or your

1025
00:58:53,840 --> 00:58:57,020
favorite Escher book, and you
can watch him manipulating

1026
00:58:57,020 --> 00:58:59,400
these depth cues endlessly.

1027
00:58:59,400 --> 00:59:01,460
It's great entertainment.

1028
00:59:04,370 --> 00:59:15,030
Now what I want to do is to
bring together the themes of

1029
00:59:15,030 --> 00:59:19,960
the lectures to this point
in single demo.

1030
00:59:19,960 --> 00:59:23,280
At the moment, that doesn't
look like much of nothing,

1031
00:59:23,280 --> 00:59:27,690
except that, well, I don't know,
what does it look like?

1032
00:59:27,690 --> 00:59:28,680
Cubes.

1033
00:59:28,680 --> 00:59:29,720
All right.

1034
00:59:29,720 --> 00:59:35,180
That's interesting, because
I don't see no cubes!

1035
00:59:35,180 --> 00:59:39,240
Where's that inference
coming from?

1036
00:59:39,240 --> 00:59:42,390
What you're really have is
a bunch of Y junctions.

1037
00:59:42,390 --> 00:59:44,100
And you know about
Y junctions.

1038
00:59:44,100 --> 00:59:46,650
Those are probably corners.

1039
00:59:46,650 --> 00:59:49,140
And, they look like they might
be kind of cube-y corners.

1040
00:59:49,140 --> 00:59:51,770
But, what I'm going to do
is rotate each of those.

1041
00:59:56,750 --> 01:00:03,900
Now, this is cool stimulus
for a variety of reasons.

1042
01:00:03,900 --> 01:00:06,380
First of all, now you're
really seeing a cube.

1043
01:00:06,380 --> 01:00:07,670
Right?

1044
01:00:07,670 --> 01:00:08,610
No problem.

1045
01:00:08,610 --> 01:00:10,500
Second of all, there
are two cubes.

1046
01:00:13,320 --> 01:00:17,250
There's the cube, with
its face pointing.

1047
01:00:17,250 --> 01:00:18,450
Let's try this.

1048
01:00:18,450 --> 01:00:24,210
There's that face, pointing
down and to the right.

1049
01:00:30,690 --> 01:00:35,510
There's that face pointing
up and to the left.

1050
01:00:35,510 --> 01:00:36,270
So, you've got two cubes.

1051
01:00:36,270 --> 01:00:38,890
This is an ambigious
by stable figure.

1052
01:00:38,890 --> 01:00:40,540
It's known as a Necker cube.

1053
01:00:43,530 --> 01:00:46,180
We can quickly draw
one of those.

1054
01:00:49,090 --> 01:00:50,730
Endless fun for doodling
again.

1055
01:00:50,730 --> 01:00:55,610
You can make yourself ambiguous
figures instantly.

1056
01:00:55,610 --> 01:00:58,920
That figure by itself is known
as the Necker cube after a guy

1057
01:00:58,920 --> 01:01:00,170
named Necker.

1058
01:01:02,150 --> 01:01:05,740
So, you're inferring
two cubes.

1059
01:01:05,740 --> 01:01:08,120
You're inferring one
of two cubes.

1060
01:01:08,120 --> 01:01:11,270
The cube isn't particularly
there.

1061
01:01:11,270 --> 01:01:13,670
The black stuff --

1062
01:01:13,670 --> 01:01:16,060
all you're seeing is the
verticies of this cube--

1063
01:01:16,060 --> 01:01:22,080
but you're still managing to
infer the rest of the cube You

1064
01:01:22,080 --> 01:01:24,940
can probably see the lines
of the cube in the

1065
01:01:24,940 --> 01:01:27,500
black region, right.

1066
01:01:27,500 --> 01:01:30,620
In fact, you can see the
intersection, if you straight

1067
01:01:30,620 --> 01:01:34,020
up from here, you can see the
intersection of two lines that

1068
01:01:34,020 --> 01:01:35,270
aren't there.

1069
01:01:39,340 --> 01:01:42,570
I can make those
lines go away.

1070
01:01:42,570 --> 01:01:46,770
Now take a look at that cube,
and imagine that what you're

1071
01:01:46,770 --> 01:01:50,470
looking at is a cube --

1072
01:01:50,470 --> 01:01:54,700
sort of a wire frame cube--
that's behind a sheet of sort

1073
01:01:54,700 --> 01:01:56,040
of black swiss cheese.

1074
01:01:56,040 --> 01:01:57,950
You're looking at it
through holes.

1075
01:01:57,950 --> 01:02:00,680
Can you get it to
go back there?

1076
01:02:00,680 --> 01:02:03,120
If you get it to go back there,
you can hold it back

1077
01:02:03,120 --> 01:02:05,910
there, you probably noticed that
the subjective contours

1078
01:02:05,910 --> 01:02:08,790
pretty much disappear on you.

1079
01:02:08,790 --> 01:02:09,720
Why is that?

1080
01:02:09,720 --> 01:02:13,380
Well, if it's behind, there's
no reason that you should be

1081
01:02:13,380 --> 01:02:14,620
seeing those contour.

1082
01:02:14,620 --> 01:02:15,880
They would be invisible.

1083
01:02:15,880 --> 01:02:18,580
And so the invisible contours
become invisible.

1084
01:02:18,580 --> 01:02:21,610
Now if you bring the cube back
in front in your perception,

1085
01:02:21,610 --> 01:02:24,600
you'll see, oh yeah, now if that
cube's floating in front.

1086
01:02:24,600 --> 01:02:26,240
I ought to be able to
see the whole cube.

1087
01:02:26,240 --> 01:02:28,250
And now, I can see the
subjective contours.

1088
01:02:28,250 --> 01:02:34,610
So you can make the subjective
contours contingent on which

1089
01:02:34,610 --> 01:02:39,160
particular interpretation you
care to give to the image.

1090
01:02:39,160 --> 01:02:44,450
So, I think this illustrates
very nicely the notion that

1091
01:02:44,450 --> 01:02:51,140
what you are seeing is your
current hypothesis about what

1092
01:02:51,140 --> 01:02:55,760
might be generating the
image on the screen.

1093
01:02:55,760 --> 01:02:58,700
Another thing that it points
out, is that you are only

1094
01:02:58,700 --> 01:03:02,450
willing to entertain at limited
set of hypothesis.

1095
01:03:02,450 --> 01:03:08,140
It is extremely hard to look at
this and see it as however

1096
01:03:08,140 --> 01:03:08,780
many it is.

1097
01:03:08,780 --> 01:03:14,430
Eight little disks with
chicken feed in them.

1098
01:03:14,430 --> 01:03:16,660
With little Y's in them
of some sort.

1099
01:03:16,660 --> 01:03:18,820
It's very, very hard
to see that.

1100
01:03:18,820 --> 01:03:21,810
Even though that is perfectly
consistent with the image

1101
01:03:21,810 --> 01:03:22,960
hypothesis.

1102
01:03:22,960 --> 01:03:28,590
You are out there trying to make
a guess about the world.

1103
01:03:28,590 --> 01:03:31,300
And, you're not willing to
entertain all of them.

1104
01:03:31,300 --> 01:03:33,910
At least most of them are
immediately relegated to the

1105
01:03:33,910 --> 01:03:35,910
realm of the very unlikely.

1106
01:03:38,900 --> 01:03:43,070
And, normally out in the world,
what happens is that a

1107
01:03:43,070 --> 01:03:46,240
single hypothesis immediately
pops to the four,

1108
01:03:46,240 --> 01:03:47,370
and you accept it.

1109
01:03:47,370 --> 01:03:49,830
In weird situations
like this, you can

1110
01:03:49,830 --> 01:03:51,260
entertain a few of them.

1111
01:03:55,710 --> 01:03:58,680
You immediately narrow down the
realm of possibilities to

1112
01:03:58,680 --> 01:04:01,680
a few, not to an infinity.

1113
01:04:01,680 --> 01:04:03,720
Even though there's an infinite
number of possible

1114
01:04:03,720 --> 01:04:07,020
ways to generate this thing.

1115
01:04:07,020 --> 01:04:10,140
Shadows I've already shown
are a depth cue.

1116
01:04:10,140 --> 01:04:14,550
The reason for putting this nice
piece of renaissance art

1117
01:04:14,550 --> 01:04:17,530
up there, is to point out that,
while shadows are a

1118
01:04:17,530 --> 01:04:21,320
depth cue, you're not actually
terribly picky about the

1119
01:04:21,320 --> 01:04:24,520
physics of the situation.

1120
01:04:24,520 --> 01:04:26,600
At least not the global physics
of the situation.

1121
01:04:26,600 --> 01:04:29,100
So where's the sun here?

1122
01:04:29,100 --> 01:04:30,840
This is an outdoor scene.

1123
01:04:30,840 --> 01:04:33,430
Where's the sun coming from?

1124
01:04:33,430 --> 01:04:36,120
Well, if you look at the people
in the lower left --

1125
01:04:36,120 --> 01:04:38,490
the people on the ground plane
there -- it's pretty clear

1126
01:04:38,490 --> 01:04:41,460
that the sun must be down and
to the left somewhere.

1127
01:04:41,460 --> 01:04:42,730
Right?

1128
01:04:42,730 --> 01:04:46,620
Well, look at the shadow
underneath that portico.

1129
01:04:46,620 --> 01:04:51,720
Well, the sun must be up and to
the right there somewhere.

1130
01:04:51,720 --> 01:04:55,860
There's no consistent source of
illumination in this image,

1131
01:04:55,860 --> 01:05:00,680
but your perfectly willing to
use the shadow information to

1132
01:05:00,680 --> 01:05:01,950
give you depth information.

1133
01:05:01,950 --> 01:05:04,050
It's giving it to you locally.

1134
01:05:04,050 --> 01:05:06,300
The fact that it doesn't
add up globally,

1135
01:05:06,300 --> 01:05:07,430
doesn't bother you.

1136
01:05:07,430 --> 01:05:11,110
And it doesn't even bother you
to the extent that it bothers

1137
01:05:11,110 --> 01:05:14,450
you in the Escher picture
where the building was

1138
01:05:14,450 --> 01:05:15,370
actually impossible.

1139
01:05:15,370 --> 01:05:20,240
Here, you don't recognize the
impossibility at all, unless

1140
01:05:20,240 --> 01:05:23,040
it's pointed out to
you directly.

1141
01:05:23,040 --> 01:05:27,420
And, finally I should add to the
list, three more than are

1142
01:05:27,420 --> 01:05:29,560
rather hard to demonstrate.

1143
01:05:29,560 --> 01:05:35,010
That are hard to put up just
as Powerpoint slides.

1144
01:05:35,010 --> 01:05:38,180
When people think about depth
perception, if they think

1145
01:05:38,180 --> 01:05:41,710
about depth perception, it's
stereopsis binocular vision

1146
01:05:41,710 --> 01:05:44,470
that they typically think of.

1147
01:05:44,470 --> 01:05:47,770
Your two eyes are in two
different places in your head

1148
01:05:47,770 --> 01:05:50,790
for most of us.

1149
01:05:50,790 --> 01:05:53,960
If you blink from eye to eye,
or just cover your one eye

1150
01:05:53,960 --> 01:05:55,950
after the other, it's actually
better if you hold one finger

1151
01:05:55,950 --> 01:05:58,910
out in front of you, and
blink from eye to eye.

1152
01:05:58,910 --> 01:06:02,520
You'll see the image in the
two eyes is not the same.

1153
01:06:02,520 --> 01:06:04,480
The difference in those
two images is highly

1154
01:06:04,480 --> 01:06:06,460
geometrically regular.

1155
01:06:06,460 --> 01:06:09,530
And you make use of that
regularity as a depth cue.

1156
01:06:09,530 --> 01:06:11,230
It's called binocular
disparity.

1157
01:06:11,230 --> 01:06:12,490
Very useful depth cue.

1158
01:06:12,490 --> 01:06:17,480
It's what's giving you the magic
eye demos that you get.

1159
01:06:17,480 --> 01:06:19,940
Those posters that if you cross
your eyes just right,

1160
01:06:19,940 --> 01:06:21,200
they jump out in depth.

1161
01:06:21,200 --> 01:06:22,450
Those still around?

1162
01:06:25,590 --> 01:06:27,240
It's a very useful depth cue.

1163
01:06:27,240 --> 01:06:29,590
It's a little on the overrated
side, because it's a lot of

1164
01:06:29,590 --> 01:06:31,340
fun to study.

1165
01:06:31,340 --> 01:06:34,480
People who think that binocular
vision is the be all

1166
01:06:34,480 --> 01:06:39,100
and end all depth cue, should
cover one eye, and ask if the

1167
01:06:39,100 --> 01:06:41,990
world suddenly looks
very flat.

1168
01:06:41,990 --> 01:06:44,310
It looks a little flatter, but
I can still perfectly well

1169
01:06:44,310 --> 01:06:46,460
tell who's in front of who.

1170
01:06:46,460 --> 01:06:49,590
On the other hand, if you want
to see what's stereo is doing

1171
01:06:49,590 --> 01:06:53,360
for you, on a beautiful day like
today, go outside, lie

1172
01:06:53,360 --> 01:06:58,000
under a tree, close one eye, and
look up into the branches,

1173
01:06:58,000 --> 01:06:59,750
and try to figure out
which twigs are in

1174
01:06:59,750 --> 01:07:00,660
front of other twigs.

1175
01:07:00,660 --> 01:07:02,610
You'll have very hard
time doing it.

1176
01:07:02,610 --> 01:07:05,270
Open the eye, and the whole
thing will jump

1177
01:07:05,270 --> 01:07:06,650
out at you in depth.

1178
01:07:06,650 --> 01:07:09,780
That's the sort of information
that stereo is giving you.

1179
01:07:09,780 --> 01:07:11,480
You don't have to do
it with stereo.

1180
01:07:11,480 --> 01:07:15,170
Motion parralax to jump to the
bottom of that list, is a

1181
01:07:15,170 --> 01:07:20,990
similar sort of geometric
clue.

1182
01:07:20,990 --> 01:07:24,700
If I'm here, and then I'm here,
the image changes in a

1183
01:07:24,700 --> 01:07:26,460
geometrically regular way.

1184
01:07:26,460 --> 01:07:29,930
You guys are sliding around on
my retina in such a way that

1185
01:07:29,930 --> 01:07:33,110
these guys are moving more than
you guys out in the back

1186
01:07:33,110 --> 01:07:34,310
on my retina.

1187
01:07:34,310 --> 01:07:37,870
And, I know that, again
implicitly, like I know linear

1188
01:07:37,870 --> 01:07:40,810
praralax and I can use that
to inferred depth.

1189
01:07:40,810 --> 01:07:43,680
Try this under the tree, and
you can see the same thing.

1190
01:07:43,680 --> 01:07:46,320
Close one eye, look up
into the branches.

1191
01:07:46,320 --> 01:07:47,890
The branches look
relatively flat.

1192
01:07:47,890 --> 01:07:50,340
Now rather than opening this
eye, just move your head back

1193
01:07:50,340 --> 01:07:53,570
and forth, and the tree will
jump out at you in depth.

1194
01:07:53,570 --> 01:07:55,770
Actually quite striking, you
should try this sometime.

1195
01:07:55,770 --> 01:07:59,210
I don't know if anybody ever
does take me up on this

1196
01:07:59,210 --> 01:07:59,990
suggestion.

1197
01:07:59,990 --> 01:08:03,940
So if you actually try it, let
me know so that I know that

1198
01:08:03,940 --> 01:08:05,220
somebody actually tried it.

1199
01:08:05,220 --> 01:08:11,670
Oh and vergence is another one
of these geometrical cues.

1200
01:08:11,670 --> 01:08:12,890
Hold your finger out
in front of you.

1201
01:08:12,890 --> 01:08:14,600
Look at your finger.

1202
01:08:14,600 --> 01:08:18,470
Now move the finger towards you,
holding it as a single

1203
01:08:18,470 --> 01:08:20,830
finger as long as you can.

1204
01:08:20,830 --> 01:08:23,790
So I can watch you
go cross eyed.

1205
01:08:23,790 --> 01:08:27,570
What you are doing is converging
your eyes.

1206
01:08:27,570 --> 01:08:30,380
And, if you could move your
finger further out, you would

1207
01:08:30,380 --> 01:08:32,030
be diverging your eyes.

1208
01:08:32,030 --> 01:08:35,930
Well, what you can think of that
as doing, is taking like

1209
01:08:35,930 --> 01:08:40,540
a pair of sticks and pointing
them at the object.

1210
01:08:40,540 --> 01:08:44,520
It's your visual axis in the
sense, you're pointing at the

1211
01:08:44,520 --> 01:08:46,400
object there.

1212
01:08:46,400 --> 01:08:49,510
And the angle formed by those
sticks is narrower if you're

1213
01:08:49,510 --> 01:08:52,250
looking far away, then it is
if you're looking close up.

1214
01:08:52,250 --> 01:08:55,720
And you have a fairly
impoverished ability to use

1215
01:08:55,720 --> 01:08:57,950
that as a depth cue too.

1216
01:08:57,950 --> 01:09:01,090
If you were a chameleon you'd
be much better at this.

1217
01:09:01,090 --> 01:09:05,130
Chameleons have eyes the move
independently and are very

1218
01:09:05,130 --> 01:09:10,540
sensitive to the angle that
their eyes are pointing.

1219
01:09:10,540 --> 01:09:15,690
And, in fact, you've seen the
Animal Planet kind of videos

1220
01:09:15,690 --> 01:09:19,000
where the chameleon's tongue
goes out the length of its

1221
01:09:19,000 --> 01:09:21,590
body, and it grabs a fly
or something like that.

1222
01:09:21,590 --> 01:09:23,290
How does that it know
where the fly is?

1223
01:09:23,290 --> 01:09:26,690
It knows by measuring the
angle of its eyes.

1224
01:09:26,690 --> 01:09:27,980
How do we know that?

1225
01:09:27,980 --> 01:09:32,680
Well, we know that because,
somebody went and put glasses

1226
01:09:32,680 --> 01:09:36,710
on chameleon that diverged
the eye.

1227
01:09:36,710 --> 01:09:44,410
So, in order to point it's eyes
at the fly, it had the

1228
01:09:44,410 --> 01:09:46,150
angle wrong, basically.

1229
01:09:46,150 --> 01:09:48,450
So, you put a fly on a popsicle
stick you put the

1230
01:09:48,450 --> 01:09:51,580
glasses on the chameleon and
then your film the chameleon's

1231
01:09:51,580 --> 01:09:53,720
tongue, and the chameleon
keeps missing.

1232
01:09:56,320 --> 01:09:59,630
If I put those prisms on your
eyes, you will adapt, and you

1233
01:09:59,630 --> 01:10:02,750
will eventually be able to
catch the fly again.

1234
01:10:02,750 --> 01:10:05,010
Should you be so inclined.

1235
01:10:05,010 --> 01:10:08,910
The chameleon turns out to be a
less adaptable creature than

1236
01:10:08,910 --> 01:10:11,120
you, and will not adapt.

1237
01:10:11,120 --> 01:10:14,280
You can do the same game
with chickens.

1238
01:10:14,280 --> 01:10:17,340
Put a pair of prisms on your
eyes to divert everything off

1239
01:10:17,340 --> 01:10:20,300
say fifteen degrees to the left,
and then if I tell you

1240
01:10:20,300 --> 01:10:22,220
pick this up, you'll reach
fifteen degrees

1241
01:10:22,220 --> 01:10:23,230
in the wrong direction.

1242
01:10:23,230 --> 01:10:25,050
But eventually you'll learn.

1243
01:10:25,050 --> 01:10:27,330
Put the prisms on a chicken.

1244
01:10:27,330 --> 01:10:28,750
Put some grain down.

1245
01:10:28,750 --> 01:10:30,180
Here are the grains here.

1246
01:10:30,180 --> 01:10:31,430
The chicken sees
it over there.

1247
01:10:31,430 --> 01:10:32,370
Chickens pecking over there.

1248
01:10:32,370 --> 01:10:35,120
Chickens not getting
any grain.

1249
01:10:35,120 --> 01:10:41,170
Chicken will do that forever
and apparently not learn to

1250
01:10:41,170 --> 01:10:44,430
get it right.

1251
01:10:44,430 --> 01:10:45,190
Oh, good.

1252
01:10:45,190 --> 01:10:51,620
I left myself with enough time
to talk about inferences in a

1253
01:10:51,620 --> 01:10:55,620
more global sense of combining
information

1254
01:10:55,620 --> 01:10:56,730
from across the senses.

1255
01:10:56,730 --> 01:11:01,720
So, you've got this job to try
to figure out what's going on,

1256
01:11:01,720 --> 01:11:04,360
I've been talking about doing
that specifically with the

1257
01:11:04,360 --> 01:11:06,180
visual system.

1258
01:11:06,180 --> 01:11:11,560
But, you're collecting
information from multiple

1259
01:11:11,560 --> 01:11:15,230
sources, and taking whatever the
best information is, so if

1260
01:11:15,230 --> 01:11:20,140
I show you a movie, you'll get
captured by the visual

1261
01:11:20,140 --> 01:11:25,350
information, and you're
perfectly happy to hear the

1262
01:11:25,350 --> 01:11:29,490
words coming out of the mouth of
the guy on the screen, even

1263
01:11:29,490 --> 01:11:31,910
though it's coming out of some
speaker on the side.

1264
01:11:31,910 --> 01:11:34,450
And it doesn't matter if they
got fancy dolby stereo or

1265
01:11:34,450 --> 01:11:35,140
something like that.

1266
01:11:35,140 --> 01:11:41,040
Use a cheap, simple speaker
sitting off to the side and

1267
01:11:41,040 --> 01:11:43,380
you'll still hear it as coming
out of the guy's mouth if

1268
01:11:43,380 --> 01:11:43,930
you're watching it.

1269
01:11:43,930 --> 01:11:45,140
So you're combining information

1270
01:11:45,140 --> 01:11:46,660
from multiple senses.

1271
01:11:46,660 --> 01:11:51,670
The particular example I thought
I would discuss with

1272
01:11:51,670 --> 01:11:59,610
you, is the ever pleasant
example of motion sickness.

1273
01:11:59,610 --> 01:12:08,830
So, when I first came to
graduate school, my lab was

1274
01:12:08,830 --> 01:12:16,220
doing research for NASA on the
effects of looking at large

1275
01:12:16,220 --> 01:12:18,220
fields that were rotating.

1276
01:12:18,220 --> 01:12:20,030
Sort of Omni theater stuff.

1277
01:12:20,030 --> 01:12:23,370
If you look at a whole field
that's rotating counter clock

1278
01:12:23,370 --> 01:12:25,820
wise around your line of sight,
you feel like you're

1279
01:12:25,820 --> 01:12:27,940
rotating clock wise.

1280
01:12:27,940 --> 01:12:31,190
Meantime guys up at Brandeis
who we were collaborating

1281
01:12:31,190 --> 01:12:34,260
with, were doing the same sort
of thing, but their depended

1282
01:12:34,260 --> 01:12:37,620
measure was how long it
took you to throw up.

1283
01:12:37,620 --> 01:12:39,090
Fortunately, that was not my

1284
01:12:39,090 --> 01:12:42,630
introduction to graduate school.

1285
01:12:42,630 --> 01:12:47,570
But, it will make you sick.

1286
01:12:47,570 --> 01:12:49,230
I might as well collect
some data here.

1287
01:12:49,230 --> 01:12:52,460
How many people have ever
been motion sick here?

1288
01:12:52,460 --> 01:12:53,660
OK.

1289
01:12:53,660 --> 01:12:56,060
What made you sick?

1290
01:12:56,060 --> 01:12:56,520
Motion.

1291
01:12:56,520 --> 01:12:56,740
Yes.

1292
01:12:56,740 --> 01:12:59,840
Thank you.

1293
01:12:59,840 --> 01:13:03,890
Could were get a little more
specific, while keeping within

1294
01:13:03,890 --> 01:13:06,430
the realm of good taste here?

1295
01:13:11,790 --> 01:13:12,050
Oh.

1296
01:13:12,050 --> 01:13:14,470
Reading on a bumpy bus.

1297
01:13:14,470 --> 01:13:15,940
That's one good example.

1298
01:13:15,940 --> 01:13:17,190
Anybody got another good one?

1299
01:13:39,650 --> 01:13:40,900
We'll take one more here.

1300
01:13:43,890 --> 01:13:45,130
Oh, spinning around
in a circle.

1301
01:13:45,130 --> 01:13:46,760
Did you get sick while you're
spinning around in circle, or

1302
01:13:46,760 --> 01:13:49,650
after you stopped?

1303
01:13:49,650 --> 01:13:50,600
After you stopped.

1304
01:13:50,600 --> 01:13:54,170
And, then when you stopped,
you felt like you were

1305
01:13:54,170 --> 01:13:56,660
spinning around in the
other direction.

1306
01:13:56,660 --> 01:13:58,740
So when you're spinning yourself
around in the circle,

1307
01:13:58,740 --> 01:14:02,980
-- you have, in your inner ears,
these tubes filled with

1308
01:14:02,980 --> 01:14:11,320
fluid, -- and quick drawing
here, quick bit of vestibular

1309
01:14:11,320 --> 01:14:14,980
physiology --need a nice
fat piece of chalk --

1310
01:14:14,980 --> 01:14:21,970
oh, it's yellow --

1311
01:14:21,970 --> 01:14:26,820
you've got these tubes inside
your ears that are filled with

1312
01:14:26,820 --> 01:14:34,400
a fluid, and inside
a little space in

1313
01:14:34,400 --> 01:14:37,110
there, are these hairs.

1314
01:14:37,110 --> 01:14:39,350
If you bend the hairs, it sends
a signal off to your

1315
01:14:39,350 --> 01:14:42,080
brain, that's the transducer to
the equivalent of the photo

1316
01:14:42,080 --> 01:14:43,410
receptors from the eyes.

1317
01:14:50,250 --> 01:14:53,370
You can imagine taking a bucket
and starting to move,

1318
01:14:53,370 --> 01:14:54,950
the fluid stays behind
for a little

1319
01:14:54,950 --> 01:14:56,480
while, and sloshes around.

1320
01:14:56,480 --> 01:14:59,070
That's why its hard to carry
buckets of liquid around if

1321
01:14:59,070 --> 01:14:59,940
they are too full.

1322
01:14:59,940 --> 01:15:03,110
So if you rotate your head, the
fluid tends to stay put.

1323
01:15:03,110 --> 01:15:07,550
And, it moves over the hairs
bending them and telling you

1324
01:15:07,550 --> 01:15:08,400
that you move your head.

1325
01:15:08,400 --> 01:15:10,750
That's what's telling you about
this sort of motion.

1326
01:15:10,750 --> 01:15:14,730
Well, you spin around for
awhile, and the fluid in here

1327
01:15:14,730 --> 01:15:17,530
eventually catches up with
you and starts moving.

1328
01:15:17,530 --> 01:15:19,250
Then you stop.

1329
01:15:19,250 --> 01:15:24,830
And the fluid keeps going,
and you say, oh no.

1330
01:15:24,830 --> 01:15:27,430
The other way you can do this,
by the way-- so it's very

1331
01:15:27,430 --> 01:15:30,670
carefully calibrated system,
turns out that alcohol is

1332
01:15:30,670 --> 01:15:34,720
lighter then this fluid.

1333
01:15:34,720 --> 01:15:38,350
If you drink, the reason you
get dizzy, it's not because

1334
01:15:38,350 --> 01:15:41,120
you pickled your brain, which
is also true, but because

1335
01:15:41,120 --> 01:15:44,520
you've dilute it this fluid with
alcohol and this stuff is

1336
01:15:44,520 --> 01:15:45,600
uncalibrated.

1337
01:15:45,600 --> 01:15:48,510
And, now, you move your
head a little, and the

1338
01:15:48,510 --> 01:15:49,800
brain, says, oh, man.

1339
01:15:49,800 --> 01:15:51,240
We just moved a lot.

1340
01:15:51,240 --> 01:15:52,490
Don't move that head.

1341
01:15:57,080 --> 01:16:00,590
What she was pointing out, is
that what turns out to be the

1342
01:16:00,590 --> 01:16:06,610
great stimulus for making
yourself motion sick, is a

1343
01:16:06,610 --> 01:16:11,690
mismatch between the information
that in particular

1344
01:16:11,690 --> 01:16:16,280
this vesticular system, this
balance system, and your eyes

1345
01:16:16,280 --> 01:16:16,880
are giving you.

1346
01:16:16,880 --> 01:16:22,510
So the example given here of
reading on the bus is a

1347
01:16:22,510 --> 01:16:24,030
marvelous example.

1348
01:16:24,030 --> 01:16:28,680
Because what you're doing, in
order to read, you're holding

1349
01:16:28,680 --> 01:16:30,410
this thing so it's a relatively

1350
01:16:30,410 --> 01:16:31,900
stable visual stimulus.

1351
01:16:31,900 --> 01:16:34,330
Nothing's happening here.

1352
01:16:34,330 --> 01:16:36,340
But your vestibular
system is saying

1353
01:16:36,340 --> 01:16:37,840
[? bumbidingdibingdaeda-- ?]

1354
01:16:37,840 --> 01:16:39,510
lots of stuff is happening.

1355
01:16:39,510 --> 01:16:43,960
And your digestive system is
busy saying [? Blech. ?]

1356
01:16:43,960 --> 01:16:46,080
Same thing happens on
a plane, right?

1357
01:16:46,080 --> 01:16:51,100
The problem on a plane is that
when the plane bounces up and

1358
01:16:51,100 --> 01:16:53,330
down, I mean short of
catastrophic bouncing up and

1359
01:16:53,330 --> 01:16:55,480
down, when a plane hits
turbulence and it's bouncing

1360
01:16:55,480 --> 01:16:58,760
up and down, what do you see?

1361
01:16:58,760 --> 01:17:00,450
Nothing.

1362
01:17:00,450 --> 01:17:01,360
Right.

1363
01:17:01,360 --> 01:17:02,940
There's nothing happening
visually.

1364
01:17:02,940 --> 01:17:03,710
What do you feel?

1365
01:17:03,710 --> 01:17:06,010
You feel [? bompitdumitbomp. ?]

1366
01:17:06,010 --> 01:17:07,260
and you know, [? bloop. ?]

1367
01:17:09,540 --> 01:17:14,280
And, the interesting question
is why'd you get sick?

1368
01:17:14,280 --> 01:17:18,310
It's fine to say that the
mismatch of visual and

1369
01:17:18,310 --> 01:17:22,410
vestibular information turns
out to be nauseating.

1370
01:17:22,410 --> 01:17:25,000
But, why should it
make you sick?

1371
01:17:25,000 --> 01:17:29,410
The answer is, it's
another inference.

1372
01:17:29,410 --> 01:17:31,700
So, the next question then
is what's the inference?

1373
01:17:31,700 --> 01:17:32,990
What are you guessing?

1374
01:17:36,210 --> 01:17:39,140
You're guessing that the
airline food was lousy.

1375
01:17:39,140 --> 01:17:40,960
Yeah.

1376
01:17:40,960 --> 01:17:42,910
Yeah, you're guessing that
you were poisoned.

1377
01:17:42,910 --> 01:17:44,620
Why you guessing that
you were poisoned?

1378
01:17:47,780 --> 01:17:50,710
It's the mismatch
that's critical.

1379
01:17:50,710 --> 01:17:56,330
If you just make your vision
strange, I could show you

1380
01:17:56,330 --> 01:17:59,010
weird stuff you hadn't
seen before, and you

1381
01:17:59,010 --> 01:18:01,590
wouldn't throw up.

1382
01:18:01,590 --> 01:18:05,250
And, I can also bounce you
around, and until I do fairly

1383
01:18:05,250 --> 01:18:07,410
dramatic stuff, you
won't throw up.

1384
01:18:07,410 --> 01:18:11,950
But, if I mismatch what your
visual system is telling you

1385
01:18:11,950 --> 01:18:15,660
about your body, and what your
vestibular system is telling

1386
01:18:15,660 --> 01:18:20,020
you, what this is,is a
protection of sorts against

1387
01:18:20,020 --> 01:18:21,800
neuro-toxins.

1388
01:18:21,800 --> 01:18:23,030
How do poisons work?

1389
01:18:23,030 --> 01:18:27,080
Well some of them work by
attacking your nervous system.

1390
01:18:27,080 --> 01:18:28,960
Lot of work by attack
your nervous system.

1391
01:18:28,960 --> 01:18:33,330
How are you, the owner of the
nervous system, going to know,

1392
01:18:33,330 --> 01:18:37,770
you're going to think, I can't
integrate anymore.

1393
01:18:40,780 --> 01:18:43,780
I can no longer remember
that I love my mother.

1394
01:18:43,780 --> 01:18:46,460
The sorts of things that
immediately present themselves

1395
01:18:46,460 --> 01:18:50,950
to you as you are being
poisoned, are, your senses are

1396
01:18:50,950 --> 01:18:56,140
coming unglued from each other,
and so if your eyes are

1397
01:18:56,140 --> 01:18:59,540
saying, bong-de-bong bong, and
your vestibular system is

1398
01:18:59,540 --> 01:19:03,940
saying that your body infers
that you have been poisoned,

1399
01:19:03,940 --> 01:19:06,860
and a useful idea, if you've
been poisoned, is to get rid

1400
01:19:06,860 --> 01:19:08,850
of whatever you just ate.

1401
01:19:08,850 --> 01:19:13,370
So, why does this happen on
airplanes and stuff like that?

1402
01:19:13,370 --> 01:19:17,700
You weren't built to be flying
around at 30,000 feet bouncing

1403
01:19:17,700 --> 01:19:18,640
around in the clouds.

1404
01:19:18,640 --> 01:19:22,110
It just wasn't what's nature set
you up to do, and so you

1405
01:19:22,110 --> 01:19:24,520
get the unfortunate situation
where even though you haven't

1406
01:19:24,520 --> 01:19:27,110
been poisoned, you get sick.

1407
01:19:27,110 --> 01:19:28,530
OK.

1408
01:19:28,530 --> 01:19:29,970
Enough of that cheery topic.