1 00:00:01,680 --> 00:00:04,080 The following content is provided under a Creative 2 00:00:04,080 --> 00:00:05,620 Commons license. 3 00:00:05,620 --> 00:00:07,920 Your support will help MIT OpenCourseWare 4 00:00:07,920 --> 00:00:12,280 continue to offer high quality educational resources for free. 5 00:00:12,280 --> 00:00:14,910 To make a donation or view additional materials 6 00:00:14,910 --> 00:00:18,870 from hundreds of MIT courses, visit MIT OpenCourseWare 7 00:00:18,870 --> 00:00:21,969 at ocw.mit.edu. 8 00:00:21,969 --> 00:00:23,760 NANCY KANWISHER: So I'm going to talk today 9 00:00:23,760 --> 00:00:25,572 about a couple of things. 10 00:00:25,572 --> 00:00:27,030 I had a hell of a time constructing 11 00:00:27,030 --> 00:00:29,890 a nice clean narrative arc to everything I wanted to say. 12 00:00:29,890 --> 00:00:31,890 And so I finally just decided, the hell with it. 13 00:00:31,890 --> 00:00:33,150 I'm just going to be honest. 14 00:00:33,150 --> 00:00:34,280 There's several different pieces. 15 00:00:34,280 --> 00:00:35,613 They don't make a narrative arc. 16 00:00:35,613 --> 00:00:36,750 That's life. 17 00:00:36,750 --> 00:00:41,520 I want to address what I see as a sort of macroscopic view 18 00:00:41,520 --> 00:00:43,110 of the organization of the human brain 19 00:00:43,110 --> 00:00:46,704 is giving us a kind of picture of what 20 00:00:46,704 --> 00:00:48,120 I'm going to call the architecture 21 00:00:48,120 --> 00:00:49,361 of human intelligence. 22 00:00:49,361 --> 00:00:51,610 We're trying to understand intelligence in this class. 23 00:00:51,610 --> 00:00:54,630 And so I think the overall organization of the human 24 00:00:54,630 --> 00:00:57,360 brain-- in which we've made a lot of progress in the last 20 25 00:00:57,360 --> 00:00:58,080 years-- 26 00:00:58,080 --> 00:01:01,260 gives us a kind of macro picture of what 27 00:01:01,260 --> 00:01:03,060 the pieces of the system are. 28 00:01:03,060 --> 00:01:04,180 So I'll talk about that. 29 00:01:04,180 --> 00:01:07,200 And then I'll also-- if I talk fast enough-- 30 00:01:07,200 --> 00:01:10,860 do a kind of whirlwind introduction through the basic 31 00:01:10,860 --> 00:01:14,550 methods of human cognitive neuroscience using face 32 00:01:14,550 --> 00:01:17,490 recognition as an example to illustrate what each 33 00:01:17,490 --> 00:01:19,270 of the methods can and cannot do. 34 00:01:19,270 --> 00:01:20,781 So that's the agenda. 35 00:01:20,781 --> 00:01:22,030 It's going to be pretty basic. 36 00:01:22,030 --> 00:01:23,550 So if you've heard me speak before, 37 00:01:23,550 --> 00:01:26,010 you've probably heard a lot of this. 38 00:01:26,010 --> 00:01:27,600 Anyway, the key question we're trying 39 00:01:27,600 --> 00:01:29,970 to address in this course is, how does the brain 40 00:01:29,970 --> 00:01:32,250 produce intelligent behavior? 41 00:01:32,250 --> 00:01:34,530 And how may we be able to replicate that 42 00:01:34,530 --> 00:01:36,390 intelligence in machines? 43 00:01:36,390 --> 00:01:39,090 So there's, of course, a million different ways 44 00:01:39,090 --> 00:01:42,090 to go at that question. 45 00:01:42,090 --> 00:01:45,660 And you can go at it from a kind of computational angle, 46 00:01:45,660 --> 00:01:49,140 a coding perspective, from a fine-grained neural circuit 47 00:01:49,140 --> 00:01:49,740 perspective. 48 00:01:49,740 --> 00:01:52,380 But I'm going to do something that's kind of in between. 49 00:01:52,380 --> 00:01:55,500 because those are the things we can approach in human brains. 50 00:01:55,500 --> 00:01:57,180 And it's really human intelligence 51 00:01:57,180 --> 00:01:58,970 we want to understand. 52 00:01:58,970 --> 00:02:01,010 It's a sum of human intelligences. 53 00:02:01,010 --> 00:02:04,020 A lot of it are things that we share with animals, 54 00:02:04,020 --> 00:02:05,280 but some of it is not. 55 00:02:05,280 --> 00:02:09,120 And so I think it's important to be able to approach this 56 00:02:09,120 --> 00:02:14,310 not just from the perspectives of animal research, magnificent 57 00:02:14,310 --> 00:02:16,230 as those methods are, but to also see what we 58 00:02:16,230 --> 00:02:18,040 can learn about human brains. 59 00:02:18,040 --> 00:02:18,540 OK. 60 00:02:18,540 --> 00:02:21,870 So I'll talk a bit about the overall functional architecture 61 00:02:21,870 --> 00:02:22,810 of the human brain. 62 00:02:22,810 --> 00:02:24,476 What are the basic pieces of the system? 63 00:02:24,476 --> 00:02:27,960 And then I'll get into some different methods 64 00:02:27,960 --> 00:02:30,500 and what they tell us about face perception. 65 00:02:30,500 --> 00:02:31,000 OK. 66 00:02:31,000 --> 00:02:34,590 So at the most general level, we can ask whether human 67 00:02:34,590 --> 00:02:37,230 intelligence-- as people have been asking for centuries, 68 00:02:37,230 --> 00:02:38,340 actually-- 69 00:02:38,340 --> 00:02:40,590 whether human intelligence is the product 70 00:02:40,590 --> 00:02:43,980 of a bunch of very special purpose components, 71 00:02:43,980 --> 00:02:47,640 each optimized to solve a specific problem, 72 00:02:47,640 --> 00:02:50,010 kind of like this device here, where 73 00:02:50,010 --> 00:02:54,270 you have a saw for cutting wood, scissors for cutting paper. 74 00:02:54,270 --> 00:02:56,010 Saws don't work that well on paper, 75 00:02:56,010 --> 00:02:59,430 and scissors don't work that well on wood. 76 00:02:59,430 --> 00:03:03,450 Or whether human intelligence is a product of some more 77 00:03:03,450 --> 00:03:07,080 generic, all-purpose computational power 78 00:03:07,080 --> 00:03:10,650 that makes us generically smart without optimizing us 79 00:03:10,650 --> 00:03:13,340 for any particular task. 80 00:03:13,340 --> 00:03:15,030 And just to foreshadow the answer, 81 00:03:15,030 --> 00:03:18,240 as in all questions in psychology, the answer is both. 82 00:03:18,240 --> 00:03:20,550 But we'll do that in some detail. 83 00:03:20,550 --> 00:03:23,370 Before we get into that, who cares? 84 00:03:23,370 --> 00:03:25,800 And I'd say, first of all, this kind of macro level 85 00:03:25,800 --> 00:03:28,380 question about functional components of the human mind 86 00:03:28,380 --> 00:03:31,155 and brain matters for a bunch of reasons. 87 00:03:31,155 --> 00:03:33,780 First of all, I just think it's one of the most basic questions 88 00:03:33,780 --> 00:03:37,890 we can ask about ourselves-- about who we are-- 89 00:03:37,890 --> 00:03:42,810 is to ask what the basic pieces are of our minds. 90 00:03:42,810 --> 00:03:47,340 Second, more pragmatically, this kind of divide and conquer 91 00:03:47,340 --> 00:03:49,710 research strategy has been effective in lots 92 00:03:49,710 --> 00:03:51,270 of different fields that are trying 93 00:03:51,270 --> 00:03:52,950 to understand a complex system. 94 00:03:52,950 --> 00:03:55,537 What do you do with this incredibly complex system, 95 00:03:55,537 --> 00:03:57,870 where you just can't even figure out how to get started? 96 00:03:57,870 --> 00:03:59,411 Well, one sensible way to get started 97 00:03:59,411 --> 00:04:02,004 is first figure out what its pieces are and then 98 00:04:02,004 --> 00:04:04,170 maybe try to figure out how each of the pieces work. 99 00:04:04,170 --> 00:04:06,829 And then maybe some day, maybe not in my lifetime, 100 00:04:06,829 --> 00:04:08,370 figure out how they all work together 101 00:04:08,370 --> 00:04:12,300 in some coordinated fashion. 102 00:04:12,300 --> 00:04:15,750 And third, somewhat more subtly, of course, we 103 00:04:15,750 --> 00:04:17,640 want to know not just what the pieces are, 104 00:04:17,640 --> 00:04:20,940 but what the computations that are performed in each 105 00:04:20,940 --> 00:04:23,250 of those pieces and what the representations extracted 106 00:04:23,250 --> 00:04:25,200 in each piece are. 107 00:04:25,200 --> 00:04:28,710 And I think even just a functional characterization 108 00:04:28,710 --> 00:04:31,399 of the scope of a particular brain region 109 00:04:31,399 --> 00:04:32,940 already gives us some important clues 110 00:04:32,940 --> 00:04:35,590 about the kinds of computations that go on there. 111 00:04:35,590 --> 00:04:37,920 So if we find that there's a part of the brain that's 112 00:04:37,920 --> 00:04:40,200 primarily involved in face recognition, 113 00:04:40,200 --> 00:04:42,660 not in reading visually presented words, 114 00:04:42,660 --> 00:04:45,560 recognizing scenes, or recognizing objects, 115 00:04:45,560 --> 00:04:47,700 that already gives us some clues about the kinds 116 00:04:47,700 --> 00:04:51,570 of computations that would be appropriate for that scope 117 00:04:51,570 --> 00:04:52,336 of task. 118 00:04:52,336 --> 00:04:54,210 So if you tried to write the code to do that, 119 00:04:54,210 --> 00:04:56,760 you'd be writing very different code if it only 120 00:04:56,760 --> 00:04:59,430 had to do face recognition versus if it also 121 00:04:59,430 --> 00:05:02,110 had to be able to recognize words and scenes 122 00:05:02,110 --> 00:05:04,040 and objects presented visually. 123 00:05:04,040 --> 00:05:04,540 OK. 124 00:05:04,540 --> 00:05:07,750 So that's my list of the main reasons. 125 00:05:07,750 --> 00:05:10,330 And of course, there are heaps of different ways 126 00:05:10,330 --> 00:05:15,760 to investigate this question, and I'll 127 00:05:15,760 --> 00:05:17,600 mention some of those in the second half. 128 00:05:17,600 --> 00:05:20,890 But I want to start with Spearman, 129 00:05:20,890 --> 00:05:23,740 who published a paper in 1904 in the American 130 00:05:23,740 --> 00:05:25,570 Journal of Psychology. 131 00:05:25,570 --> 00:05:28,510 This article was sandwiched between a discussion 132 00:05:28,510 --> 00:05:31,330 of the soul and an article on the psychology 133 00:05:31,330 --> 00:05:33,040 of the English sparrow. 134 00:05:33,040 --> 00:05:37,330 And in this article, Spearman did the following low tech 135 00:05:37,330 --> 00:05:39,760 but fascinating thing. 136 00:05:39,760 --> 00:05:43,060 He tested a whole bunch of kids in two different schools 137 00:05:43,060 --> 00:05:45,610 on a wide variety of different tasks. 138 00:05:45,610 --> 00:05:48,550 And this included scholastic achievement type things. 139 00:05:48,550 --> 00:05:50,560 He got exam grades from each student 140 00:05:50,560 --> 00:05:52,420 in a bunch of different classes. 141 00:05:52,420 --> 00:05:54,340 And he measured a whole bunch of other kinds 142 00:05:54,340 --> 00:05:56,740 of psychological abilities, including 143 00:05:56,740 --> 00:05:59,440 some very psychophysical perceptual discrimination 144 00:05:59,440 --> 00:06:00,310 abilities. 145 00:06:00,310 --> 00:06:02,740 How well could people discriminate 146 00:06:02,740 --> 00:06:04,540 the loudness of two different tones, 147 00:06:04,540 --> 00:06:06,970 the brightness of two different flashes of light, 148 00:06:06,970 --> 00:06:12,190 the weight of two different pieces of stuff? 149 00:06:12,190 --> 00:06:14,650 And what he found-- well, before I tell you what he found, 150 00:06:14,650 --> 00:06:16,210 what would you expect with this? 151 00:06:16,210 --> 00:06:19,300 Should we expect a correlation between your ability 152 00:06:19,300 --> 00:06:21,580 to discriminate two different loudnesses 153 00:06:21,580 --> 00:06:27,520 and, say, your math score in grade five on a math exam? 154 00:06:27,520 --> 00:06:31,480 Spearman's main result is that most pairs of tasks 155 00:06:31,480 --> 00:06:32,890 were correlated with each other. 156 00:06:32,890 --> 00:06:34,480 That is, if you were good at one, 157 00:06:34,480 --> 00:06:35,710 you're good at the others-- 158 00:06:35,710 --> 00:06:37,690 even tasks that seemingly had very little 159 00:06:37,690 --> 00:06:39,220 to do with each other. 160 00:06:39,220 --> 00:06:41,290 And this is the basis of the whole idea 161 00:06:41,290 --> 00:06:43,690 of g, which is the general factor, which 162 00:06:43,690 --> 00:06:48,400 is what led to the whole idea of IQ and IQ testing. 163 00:06:48,400 --> 00:06:51,297 And in America, we're very uptight about the idea of IQ. 164 00:06:51,297 --> 00:06:53,380 Brits don't seem to have a problem with this idea. 165 00:06:53,380 --> 00:06:57,620 They're very enthusiastic about the idea and always have been. 166 00:06:57,620 --> 00:07:03,640 But aside from all the social uses and misuses of IQ tests, 167 00:07:03,640 --> 00:07:06,610 the point is there's actually a deep discovery about psychology 168 00:07:06,610 --> 00:07:09,250 that Spearman made from the fact that all of these tasks 169 00:07:09,250 --> 00:07:10,810 were correlated with each other. 170 00:07:10,810 --> 00:07:13,750 He didn't know what it was, kind of like Gregor Mendel inferring 171 00:07:13,750 --> 00:07:16,750 genes without knowing anything about molecular biology. 172 00:07:16,750 --> 00:07:19,570 Spearman just inferred there's something general 173 00:07:19,570 --> 00:07:21,700 about the human intellect such that there 174 00:07:21,700 --> 00:07:25,060 are these strong correlations across tasks. 175 00:07:25,060 --> 00:07:27,060 OK, so that's g. 176 00:07:27,060 --> 00:07:29,980 But less well known about Spearman's work 177 00:07:29,980 --> 00:07:34,240 is he also talked about the specific factor, s. 178 00:07:34,240 --> 00:07:37,960 And s was the fact that although the broad result 179 00:07:37,960 --> 00:07:40,985 of his experiments was that most pairs of tasks were correlated, 180 00:07:40,985 --> 00:07:43,360 there were some tasks that weren't so strongly correlated 181 00:07:43,360 --> 00:07:46,510 with others, and that you could factor those out and discover 182 00:07:46,510 --> 00:07:49,960 some mental abilities that weren't just broadly shared 183 00:07:49,960 --> 00:07:52,060 across subjects. 184 00:07:52,060 --> 00:07:54,550 And I think this kind of foreshadows everything 185 00:07:54,550 --> 00:07:56,860 that we see with functional MRI. 186 00:07:56,860 --> 00:08:00,135 There's a lot of specific s's, and there's also some g. 187 00:08:00,135 --> 00:08:02,440 And you can see those in different brain regions, 188 00:08:02,440 --> 00:08:04,540 as I will detail next. 189 00:08:04,540 --> 00:08:08,920 Another method was invented by Franz Joseph Gall. 190 00:08:08,920 --> 00:08:11,560 And he argued that there are distinct mental faculties that 191 00:08:11,560 --> 00:08:13,990 live in different parts of the brain, which I think 192 00:08:13,990 --> 00:08:16,120 is more or less right, as I'll argue. 193 00:08:16,120 --> 00:08:19,990 But Gall lived in the 1700s, and he didn't have an MRI machine. 194 00:08:19,990 --> 00:08:23,290 So he did the best he could, which wasn't so hot. 195 00:08:23,290 --> 00:08:25,450 He invented the infamous method of phrenology. 196 00:08:25,450 --> 00:08:27,250 He felt the bumps on the skull and tried 197 00:08:27,250 --> 00:08:29,950 to relate those to specific abilities 198 00:08:29,950 --> 00:08:31,810 of different individuals, and from this, 199 00:08:31,810 --> 00:08:35,950 inferred 27 mental faculties. 200 00:08:35,950 --> 00:08:39,549 My favorites are and amativeness, filial piety, 201 00:08:39,549 --> 00:08:41,919 and veneration. 202 00:08:41,919 --> 00:08:44,330 And so there's a kernel of the right idea, 203 00:08:44,330 --> 00:08:47,680 but kind of the wrong method. 204 00:08:47,680 --> 00:08:52,750 And another method that was a very early one 205 00:08:52,750 --> 00:08:55,270 was the method of studying the loss 206 00:08:55,270 --> 00:08:59,030 of specific mental abilities after brain damage. 207 00:08:59,030 --> 00:09:01,600 And so Flourens, who's often credited 208 00:09:01,600 --> 00:09:05,890 as being the first experimental neuroscientist, 209 00:09:05,890 --> 00:09:11,196 went around making lesions in pigeons and rabbits 210 00:09:11,196 --> 00:09:12,820 and then tested them on various things. 211 00:09:12,820 --> 00:09:15,100 And he didn't really find much difference 212 00:09:15,100 --> 00:09:17,140 in what parts of the brain he took out 213 00:09:17,140 --> 00:09:18,310 for their mental abilities. 214 00:09:18,310 --> 00:09:21,590 Maybe that's because he wasn't such a hot experimental-- 215 00:09:21,590 --> 00:09:24,150 he didn't have great experimental methods. 216 00:09:24,150 --> 00:09:27,910 In any case, he argued that all sensory and volitional 217 00:09:27,910 --> 00:09:30,880 faculties exist in the cerebral hemispheres 218 00:09:30,880 --> 00:09:34,060 and must be regarded as occupying concurrently 219 00:09:34,060 --> 00:09:36,059 the same seat in those structures. 220 00:09:36,059 --> 00:09:38,350 In other words, everything is on top of everything else 221 00:09:38,350 --> 00:09:39,490 in the brain. 222 00:09:39,490 --> 00:09:44,770 So that was a sort of dominant view for a while. 223 00:09:44,770 --> 00:09:47,260 People thought Gall was kind of a crackpot, 224 00:09:47,260 --> 00:09:49,270 even though he wrote very popular books 225 00:09:49,270 --> 00:09:51,750 and went around Europe giving popular lectures 226 00:09:51,750 --> 00:09:54,640 that huge numbers of people attended. 227 00:09:54,640 --> 00:09:56,830 The respectable intellectual society 228 00:09:56,830 --> 00:09:59,420 didn't take him seriously. 229 00:09:59,420 --> 00:10:01,730 In fact, the whole idea of localization of function 230 00:10:01,730 --> 00:10:05,090 wasn't taken seriously until Paul Broca, 231 00:10:05,090 --> 00:10:08,750 a member of the French Academy, stood up 232 00:10:08,750 --> 00:10:14,399 in front of the Society of Anthropology in Paris in 1861 233 00:10:14,399 --> 00:10:16,190 and announced that the left frontal lobe is 234 00:10:16,190 --> 00:10:17,870 the seat of speech. 235 00:10:17,870 --> 00:10:20,090 And this was based on his patient 236 00:10:20,090 --> 00:10:23,210 Tan, whose brain is shown here. 237 00:10:23,210 --> 00:10:24,830 Tan was named Tan because that was 238 00:10:24,830 --> 00:10:29,240 all he could say after damage to his left inferior frontal lobe. 239 00:10:29,240 --> 00:10:31,400 And Broca pointed out that Tan had 240 00:10:31,400 --> 00:10:33,690 lots of other mental faculties preserved, 241 00:10:33,690 --> 00:10:36,210 and it was simply speech that was disrupted. 242 00:10:36,210 --> 00:10:40,020 And from this was one of the first respectable people 243 00:10:40,020 --> 00:10:43,860 to argue for localization of function. 244 00:10:43,860 --> 00:10:44,360 OK. 245 00:10:44,360 --> 00:10:47,240 So this research program goes on. 246 00:10:47,240 --> 00:10:48,810 And by the end of the 20th century, 247 00:10:48,810 --> 00:10:52,550 there's pretty much agreement that basic sensory and motor 248 00:10:52,550 --> 00:10:55,199 functions do exhibit localization 249 00:10:55,199 --> 00:10:56,240 of function in the brain. 250 00:10:56,240 --> 00:10:59,210 There are different regions for basic visual processing, 251 00:10:59,210 --> 00:11:00,830 auditory processing, and so forth. 252 00:11:00,830 --> 00:11:03,780 And that was no longer controversial. 253 00:11:03,780 --> 00:11:05,810 But the whole question of whether higher level 254 00:11:05,810 --> 00:11:10,430 mental functions were localized and distinct parts of the brain 255 00:11:10,430 --> 00:11:14,450 was controversial then and remains controversial now. 256 00:11:14,450 --> 00:11:19,700 And so the method I'll focus on is functional MRI, 257 00:11:19,700 --> 00:11:22,550 because I think it's played a huge role in addressing 258 00:11:22,550 --> 00:11:26,430 this question at this macroscopic level. 259 00:11:26,430 --> 00:11:29,300 And I think you guys know what an MRI machine is. 260 00:11:29,300 --> 00:11:33,326 In case anybody has been on Mars for a while, 261 00:11:33,326 --> 00:11:34,700 the important part is its measure 262 00:11:34,700 --> 00:11:38,450 is a very indirect measure of neural activity by way 263 00:11:38,450 --> 00:11:40,160 of a long causal chain. 264 00:11:40,160 --> 00:11:46,910 Neurons fire, you incur metabolic cost, and blood flow 265 00:11:46,910 --> 00:11:48,080 changes to that region. 266 00:11:48,080 --> 00:11:52,190 Blood flow increase more than compensates for oxygen use, 267 00:11:52,190 --> 00:11:56,210 producing a local decrease rather than 268 00:11:56,210 --> 00:11:59,270 the expected increase in deoxyhemoglobin 269 00:11:59,270 --> 00:12:01,059 relative to oxyhemoglobin. 270 00:12:01,059 --> 00:12:02,600 Those two are magnetically different. 271 00:12:02,600 --> 00:12:05,060 That's what the MRI machine detects. 272 00:12:05,060 --> 00:12:06,860 It's very indirect, so it's remarkable it 273 00:12:06,860 --> 00:12:08,990 works as well as it does. 274 00:12:08,990 --> 00:12:11,870 And it's currently the best noninvasive method 275 00:12:11,870 --> 00:12:18,380 we have in humans in terms of spatial resolution, not 276 00:12:18,380 --> 00:12:19,600 temporal resolution. 277 00:12:19,600 --> 00:12:20,600 OK. 278 00:12:20,600 --> 00:12:26,299 So many of you are already diving into the details of some 279 00:12:26,299 --> 00:12:27,340 of the data we collected. 280 00:12:27,340 --> 00:12:28,839 But in case you're on other projects 281 00:12:28,839 --> 00:12:31,150 and are coming from other fields, 282 00:12:31,150 --> 00:12:35,000 the basic format of the data in a typical functional MRI study 283 00:12:35,000 --> 00:12:38,630 is you have tens of thousands of three dimensional pixels 284 00:12:38,630 --> 00:12:41,840 or voxels that you scan. 285 00:12:41,840 --> 00:12:46,520 And typically, you sample the whole set 286 00:12:46,520 --> 00:12:48,950 once every two seconds or so. 287 00:12:48,950 --> 00:12:51,140 You can push it and do it every second or less 288 00:12:51,140 --> 00:12:53,390 under special circumstances. 289 00:12:53,390 --> 00:12:56,000 You can have more voxels by sampling at higher resolution, 290 00:12:56,000 --> 00:12:59,930 but that's a ballpark of the format of the kind of movie 291 00:12:59,930 --> 00:13:02,460 you can get of brain activity. 292 00:13:02,460 --> 00:13:02,960 OK. 293 00:13:02,960 --> 00:13:06,710 So a few things about the method and its limitations, 294 00:13:06,710 --> 00:13:09,470 because they're really important in terms of what you can learn 295 00:13:09,470 --> 00:13:11,690 from functional MRI and what you can't. 296 00:13:11,690 --> 00:13:13,630 So first of all, this is a timeline. 297 00:13:13,630 --> 00:13:17,760 My x-axis, even though it's invisible, is time in seconds. 298 00:13:17,760 --> 00:13:22,280 And so if you imagine looking at V1 and presenting a brief, say, 299 00:13:22,280 --> 00:13:26,420 tenth of a second high contrast flash of a checkerboard, what 300 00:13:26,420 --> 00:13:30,290 we know from neurophysiology is that neurons fire 301 00:13:30,290 --> 00:13:33,230 within 100 milliseconds of a visual onset. 302 00:13:33,230 --> 00:13:36,650 The information gets right up there really fast. 303 00:13:36,650 --> 00:13:39,920 The BOLD, which stands for Blood Oxygenation Level 304 00:13:39,920 --> 00:13:42,530 Dependent, or functional MRI response, 305 00:13:42,530 --> 00:13:44,930 is way lagged behind this. 306 00:13:44,930 --> 00:13:48,890 So the neurons are firing way over here in this graph, 307 00:13:48,890 --> 00:13:50,380 essentially at time zero-- 308 00:13:50,380 --> 00:13:51,780 a tenth of a second. 309 00:13:51,780 --> 00:13:56,180 But the MRI response is six seconds later, OK? 310 00:13:56,180 --> 00:13:59,000 So it's really slow. 311 00:13:59,000 --> 00:14:01,670 And that has a bunch of implications about what we can 312 00:14:01,670 --> 00:14:03,680 and cannot learn from it. 313 00:14:03,680 --> 00:14:07,450 So first of all, because it's so slow, 314 00:14:07,450 --> 00:14:14,060 we can't resolve the steps in a computation for fast systems 315 00:14:14,060 --> 00:14:17,660 like vision and hearing and language understanding-- 316 00:14:17,660 --> 00:14:20,600 systems for which we have dedicated machinery that's 317 00:14:20,600 --> 00:14:24,230 highly efficient where you can recognize the gist of a scene 318 00:14:24,230 --> 00:14:25,730 within a quarter of a second of one 319 00:14:25,730 --> 00:14:28,290 it flashes on a screen in front of you. 320 00:14:28,290 --> 00:14:30,900 And similarly, you understand the meaning of a sentence 321 00:14:30,900 --> 00:14:33,530 so rapidly that you've already parsed 322 00:14:33,530 --> 00:14:37,880 much of the sentence well before the sentence is over. 323 00:14:37,880 --> 00:14:41,580 So these are extremely efficient rapid mental processes. 324 00:14:41,580 --> 00:14:44,270 That means the component steps in those mental processes 325 00:14:44,270 --> 00:14:49,100 happen over a few tens of milliseconds. 326 00:14:49,100 --> 00:14:52,430 And we're way off in temporal resolution with functional MRI. 327 00:14:52,430 --> 00:14:55,920 All of those things are squashed together on top of each other. 328 00:14:55,920 --> 00:14:56,570 That's a drag. 329 00:14:56,570 --> 00:14:57,290 That's just life. 330 00:14:57,290 --> 00:14:59,677 We can't see those individual components steps 331 00:14:59,677 --> 00:15:00,510 with functional MRI. 332 00:15:03,340 --> 00:15:05,350 The second thing is that the spatial resolution 333 00:15:05,350 --> 00:15:08,770 is the best that we have in humans noninvasively right now, 334 00:15:08,770 --> 00:15:10,870 but it's absolutely awful compared 335 00:15:10,870 --> 00:15:12,950 to what you can do in animals. 336 00:15:12,950 --> 00:15:15,940 So I missed Jim DiCarlo's talk yesterday, 337 00:15:15,940 --> 00:15:18,130 but those methods are spectacular. 338 00:15:18,130 --> 00:15:19,930 You can record from individual neurons, 339 00:15:19,930 --> 00:15:22,990 record their precise activity with beautiful time 340 00:15:22,990 --> 00:15:23,920 information. 341 00:15:23,920 --> 00:15:29,110 In contrast, functional MRI is like the dark ages. 342 00:15:29,110 --> 00:15:31,420 We have, typically, hundreds of thousands 343 00:15:31,420 --> 00:15:33,880 of neurons in each voxel. 344 00:15:33,880 --> 00:15:36,850 So the real miracle of functional MRI 345 00:15:36,850 --> 00:15:40,060 is that we ever see anything at all rather than just garbage, 346 00:15:40,060 --> 00:15:43,120 because you're summing over so many neurons at once. 347 00:15:43,120 --> 00:15:46,057 And it's just a lucky fact of the organization 348 00:15:46,057 --> 00:15:47,890 of the human brain that you have clusterings 349 00:15:47,890 --> 00:15:51,400 of neurons with similar responsal activities 350 00:15:51,400 --> 00:15:54,040 and similar functions at such a macro grain 351 00:15:54,040 --> 00:15:56,950 that you can see some stuff with functional MRI, 352 00:15:56,950 --> 00:15:58,420 although you miss a lot as well. 353 00:16:01,940 --> 00:16:04,690 The third important limit of functional MRI that comes out 354 00:16:04,690 --> 00:16:08,800 of just a consideration of what the method measures 355 00:16:08,800 --> 00:16:12,910 is that you can only really see differences between conditions 356 00:16:12,910 --> 00:16:14,320 with functional MRI. 357 00:16:14,320 --> 00:16:17,410 The magnitude of the MRI response in a voxel at a time 358 00:16:17,410 --> 00:16:19,180 point is meaningless. 359 00:16:19,180 --> 00:16:23,190 It might be 563, and that's all it means. 360 00:16:23,190 --> 00:16:23,852 Nothing, right? 361 00:16:23,852 --> 00:16:24,560 It means nothing. 362 00:16:24,560 --> 00:16:27,640 It's just the intensity of the MRI signal. 363 00:16:27,640 --> 00:16:29,590 The only way to make it mean something 364 00:16:29,590 --> 00:16:31,510 is to compare it to something else-- 365 00:16:31,510 --> 00:16:34,510 usually two different tasks or two different stimuli. 366 00:16:34,510 --> 00:16:36,580 And so you can go far with that, but it's 367 00:16:36,580 --> 00:16:38,770 important to realize you can't train translate it 368 00:16:38,770 --> 00:16:41,950 into any kind of absolute measure of neural activity. 369 00:16:41,950 --> 00:16:45,190 It's only a relative measure of strength 370 00:16:45,190 --> 00:16:49,520 of neural activity between two or more different conditions. 371 00:16:49,520 --> 00:16:50,020 OK. 372 00:16:50,020 --> 00:16:54,070 And the final deep limitation of functional MRI 373 00:16:54,070 --> 00:16:57,880 is that we use this convenient phrase "neural activity." 374 00:16:57,880 --> 00:17:01,054 It's very convenient, because it's extremely vague. 375 00:17:01,054 --> 00:17:02,470 And fittingly so, because we don't 376 00:17:02,470 --> 00:17:04,300 know exactly what kind of neural activity 377 00:17:04,300 --> 00:17:06,099 is driving the BOLD response. 378 00:17:06,099 --> 00:17:08,500 It could be spikes or action potentials. 379 00:17:08,500 --> 00:17:12,220 It could be synaptic activity that doesn't lead to spikes. 380 00:17:12,220 --> 00:17:14,260 It could be tonic inhibition. 381 00:17:14,260 --> 00:17:16,270 It could be all kinds of different things. 382 00:17:16,270 --> 00:17:18,640 Anything that's metabolically expensive 383 00:17:18,640 --> 00:17:22,540 is likely to increase the blood flow response. 384 00:17:22,540 --> 00:17:25,329 In practice, when people have looked at it, 385 00:17:25,329 --> 00:17:28,230 it's very nicely correlated with firing rate-- 386 00:17:28,230 --> 00:17:32,660 with some bumps and caveats, so you can never be totally sure. 387 00:17:32,660 --> 00:17:35,667 But it's a pretty good proxy for firing rate. 388 00:17:35,667 --> 00:17:37,750 You just need to remember in the back of your mind 389 00:17:37,750 --> 00:17:39,550 that it could be other stuff too. 390 00:17:39,550 --> 00:17:43,630 The final, very important caveat is that functional MRI-- 391 00:17:43,630 --> 00:17:46,720 like most other methods where you're just 392 00:17:46,720 --> 00:17:49,240 recording neural activity in a variety of different ways-- 393 00:17:49,240 --> 00:17:50,860 you're just watching. 394 00:17:50,860 --> 00:17:52,520 You're not intervening. 395 00:17:52,520 --> 00:17:55,020 And that means you're not measuring the causal role 396 00:17:55,020 --> 00:17:56,260 of the things you measure. 397 00:17:56,260 --> 00:17:57,460 And that's very important, because it 398 00:17:57,460 --> 00:17:59,410 could be that everything you measure is just 399 00:17:59,410 --> 00:18:02,200 completely epiphenomenal and has absolutely nothing to do 400 00:18:02,200 --> 00:18:04,194 with behavior. 401 00:18:04,194 --> 00:18:05,860 So in practice, that's unlikely that you 402 00:18:05,860 --> 00:18:08,710 have all this systematic stuff for no reason, 403 00:18:08,710 --> 00:18:13,510 but you need to keep in mind that functional MRI affords 404 00:18:13,510 --> 00:18:16,150 no window at all into the causal role of different regions. 405 00:18:16,150 --> 00:18:20,200 For that, you need to complement it with other methods. 406 00:18:20,200 --> 00:18:22,450 So despite all these limitations, 407 00:18:22,450 --> 00:18:25,870 I think functional MRI has had a huge impact on the field. 408 00:18:25,870 --> 00:18:27,281 And admittedly, I'm biased, but I 409 00:18:27,281 --> 00:18:28,780 think it's one of these things where 410 00:18:28,780 --> 00:18:32,500 as it happens, we get so used to a result the minute 411 00:18:32,500 --> 00:18:33,370 it gets published. 412 00:18:33,370 --> 00:18:34,740 It was like, oh, yeah, right. 413 00:18:34,740 --> 00:18:37,446 One of these, one of those, so what? 414 00:18:37,446 --> 00:18:39,190 But I think it's important to step back, 415 00:18:39,190 --> 00:18:42,460 so I made a bunch of pictures to show you 416 00:18:42,460 --> 00:18:43,720 why I think this is important. 417 00:18:43,720 --> 00:18:44,220 OK. 418 00:18:44,220 --> 00:18:47,710 Here is Penfield's functional map of the human brain, 419 00:18:47,710 --> 00:18:51,180 published in 1957, a year before I was born. 420 00:18:51,180 --> 00:18:52,330 And he has six-- 421 00:18:52,330 --> 00:18:54,944 count them, six-- functional regions labeled in there. 422 00:18:54,944 --> 00:18:56,110 You probably can't see them. 423 00:18:56,110 --> 00:19:00,430 But it's the basic sensory and motor regions, visual cortex, 424 00:19:00,430 --> 00:19:05,920 auditory cortex, motor cortex, speech appear in Broca's area, 425 00:19:05,920 --> 00:19:10,270 and then my favorite is this word that says interpretive. 426 00:19:10,270 --> 00:19:10,870 Nice. 427 00:19:10,870 --> 00:19:11,590 OK. 428 00:19:11,590 --> 00:19:15,310 Anyway, this was based on electrical recording 429 00:19:15,310 --> 00:19:18,230 and stimulation in patients with epilepsy who 430 00:19:18,230 --> 00:19:20,950 were undergoing brain surgery. 431 00:19:20,950 --> 00:19:23,550 Actually a very powerful method, but that's where it got him. 432 00:19:23,550 --> 00:19:25,870 He published this near the end of his career. 433 00:19:25,870 --> 00:19:28,660 And that's nice, but it's pretty rudimentary. 434 00:19:28,660 --> 00:19:32,170 OK, now, cut to 1990, immediately 435 00:19:32,170 --> 00:19:34,810 before the advent of functional MRI. 436 00:19:34,810 --> 00:19:37,510 And this is really crude-- the black outlines 437 00:19:37,510 --> 00:19:40,410 are the basic sensory and motor regions. 438 00:19:40,410 --> 00:19:42,940 And I've added a couple of big colored blobs 439 00:19:42,940 --> 00:19:45,490 for regions that had been identified by studying patients 440 00:19:45,490 --> 00:19:46,940 with brain damage. 441 00:19:46,940 --> 00:19:49,240 So even from Broca and Wernicke, it 442 00:19:49,240 --> 00:19:51,605 was known that approximately these regions were involved 443 00:19:51,605 --> 00:19:53,230 in language, because people with damage 444 00:19:53,230 --> 00:19:55,400 there lost their language abilities. 445 00:19:55,400 --> 00:19:57,890 You get whacked in your parietal lobe, 446 00:19:57,890 --> 00:19:59,410 you have weird attentional problems, 447 00:19:59,410 --> 00:20:02,737 like neglecting the left half of space and stuff like that. 448 00:20:02,737 --> 00:20:04,570 If you have damage somewhere to the back end 449 00:20:04,570 --> 00:20:05,986 of the right hemisphere, you might 450 00:20:05,986 --> 00:20:07,350 lose face recognition ability. 451 00:20:07,350 --> 00:20:10,740 These things were known by around 1990, not much else. 452 00:20:10,740 --> 00:20:13,890 That's basically the functional map of the brain in 1990. 453 00:20:13,890 --> 00:20:16,350 That probably seems like ancient history to a lot of you, 454 00:20:16,350 --> 00:20:18,311 but not to me. 455 00:20:18,311 --> 00:20:18,810 OK. 456 00:20:18,810 --> 00:20:20,820 Here we are today. 457 00:20:20,820 --> 00:20:22,920 There's a lot of stuff we've learned, right? 458 00:20:22,920 --> 00:20:29,070 There a lot of particular parts of the human brain 459 00:20:29,070 --> 00:20:32,670 whose function has been characterized quite precisely. 460 00:20:32,670 --> 00:20:36,300 Not in the sense that we know the precise circuits in there 461 00:20:36,300 --> 00:20:38,610 or that we can very precisely characterize 462 00:20:38,610 --> 00:20:40,930 the representations or computations, 463 00:20:40,930 --> 00:20:43,920 but to the sense that we know that a region may be very 464 00:20:43,920 --> 00:20:46,230 selectively involved, for example, 465 00:20:46,230 --> 00:20:49,140 in thinking about what other people are thinking. 466 00:20:49,140 --> 00:20:52,062 A totally remarkable result that Rebecca Saxe 467 00:20:52,062 --> 00:20:54,270 who discovered it will tell you about when she's here 468 00:20:54,270 --> 00:20:55,470 next week. 469 00:20:55,470 --> 00:20:59,920 So that was completely unknown even 15 years ago, 470 00:20:59,920 --> 00:21:02,010 let alone back in 1990. 471 00:21:02,010 --> 00:21:03,810 And likewise, most of these other regions 472 00:21:03,810 --> 00:21:06,630 were either known in the blurriest sense 473 00:21:06,630 --> 00:21:08,140 or not with this precision. 474 00:21:08,140 --> 00:21:10,860 So I think even though this is very limited, 475 00:21:10,860 --> 00:21:12,750 and it's kind of step zero in trying 476 00:21:12,750 --> 00:21:16,500 to understand the human brain, I think it's important progress. 477 00:21:16,500 --> 00:21:19,500 And I think to push a little farther, 478 00:21:19,500 --> 00:21:21,630 I'd like to see this as an admittedly very 479 00:21:21,630 --> 00:21:24,690 blurry but still a picture of the architecture 480 00:21:24,690 --> 00:21:25,740 of human intelligence. 481 00:21:25,740 --> 00:21:27,390 What are the basic pieces? 482 00:21:27,390 --> 00:21:30,150 What is it we have in here to work with when we think? 483 00:21:30,150 --> 00:21:31,500 We have these basic pieces-- 484 00:21:31,500 --> 00:21:34,530 a bunch more that haven't been discovered yet, and a lot more 485 00:21:34,530 --> 00:21:36,277 that we need to know about each of these 486 00:21:36,277 --> 00:21:37,860 and how they interact and all of that, 487 00:21:37,860 --> 00:21:42,760 but a reasonable beginning. 488 00:21:42,760 --> 00:21:46,020 So that's my story here for fun. 489 00:21:46,020 --> 00:21:49,290 This is me with a bunch of functional regions identified 490 00:21:49,290 --> 00:21:50,010 in my brain. 491 00:21:50,010 --> 00:21:52,230 And so the argument I'm making here 492 00:21:52,230 --> 00:21:54,270 is that the human mind and brain contains 493 00:21:54,270 --> 00:21:56,880 a set of highly specialized components, 494 00:21:56,880 --> 00:22:00,640 each solving a different specific problem, 495 00:22:00,640 --> 00:22:03,660 and that each of these regions is present in essentially 496 00:22:03,660 --> 00:22:04,800 every normal person. 497 00:22:04,800 --> 00:22:07,860 It's just part of the basic architecture of the human mind 498 00:22:07,860 --> 00:22:09,880 and brain. 499 00:22:09,880 --> 00:22:12,550 Now, this view is pretty simple. 500 00:22:12,550 --> 00:22:15,540 But nonetheless, it's often confused 501 00:22:15,540 --> 00:22:17,190 with a whole bunch of other things 502 00:22:17,190 --> 00:22:19,240 that people think are the same thing 503 00:22:19,240 --> 00:22:22,025 and that aren't, so it's starting to drive me insane. 504 00:22:22,025 --> 00:22:23,400 So I'm going to take five minutes 505 00:22:23,400 --> 00:22:26,490 and go through the things this does not mean. 506 00:22:26,490 --> 00:22:30,310 And I hope this doesn't insult your intelligence, 507 00:22:30,310 --> 00:22:33,210 but it's amazing how in the current literature in the field 508 00:22:33,210 --> 00:22:34,650 people conflate these things. 509 00:22:34,650 --> 00:22:37,500 So I'm talking about functional specificity, which 510 00:22:37,500 --> 00:22:40,260 is the question of whether this particular region right here 511 00:22:40,260 --> 00:22:42,570 is engaged in pretty selectively in just 512 00:22:42,570 --> 00:22:45,360 that particular mental process and not 513 00:22:45,360 --> 00:22:46,830 lots of other mental process. 514 00:22:46,830 --> 00:22:49,050 That's what I mean by functional specificity. 515 00:22:49,050 --> 00:22:52,830 That's a different idea than anatomical specificity. 516 00:22:52,830 --> 00:22:55,770 Anatomical specificity would say it is only 517 00:22:55,770 --> 00:22:58,500 this region that's involved, and nothing else is involved. 518 00:22:58,500 --> 00:23:00,210 That's a different question. 519 00:23:00,210 --> 00:23:03,840 How specific is this region versus are there other regions 520 00:23:03,840 --> 00:23:05,340 that do something similar? 521 00:23:05,340 --> 00:23:08,974 Also an interesting question, but a different one. 522 00:23:08,974 --> 00:23:10,390 I'm going to go through this fast. 523 00:23:10,390 --> 00:23:12,723 So if any of it doesn't make sense, just raise your hand 524 00:23:12,723 --> 00:23:15,810 and I'll explain it more. 525 00:23:15,810 --> 00:23:17,760 Yet another idea is the necessity 526 00:23:17,760 --> 00:23:20,592 of a brain region for a particular function. 527 00:23:20,592 --> 00:23:22,050 That's actually what we really want 528 00:23:22,050 --> 00:23:24,424 to know with the functional specificity question-- is not 529 00:23:24,424 --> 00:23:26,430 just does it only turn on when you do x, 530 00:23:26,430 --> 00:23:29,760 but do you absolutely need it for x? 531 00:23:29,760 --> 00:23:31,860 And so that's actually a central question 532 00:23:31,860 --> 00:23:33,420 that's closely connected. 533 00:23:33,420 --> 00:23:36,470 It's really part of functional specificities. 534 00:23:36,470 --> 00:23:39,840 It's the causal question. 535 00:23:39,840 --> 00:23:42,900 It's different from the question of sufficiency. 536 00:23:42,900 --> 00:23:45,060 Is a given brain region sufficient 537 00:23:45,060 --> 00:23:46,334 for a mental process? 538 00:23:46,334 --> 00:23:48,750 Well, I think that's just kind of a wrong headed question, 539 00:23:48,750 --> 00:23:50,166 because nothing's ever sufficient. 540 00:23:50,166 --> 00:23:52,115 It's just kind of a confused idea. 541 00:23:52,115 --> 00:23:52,990 What would that mean? 542 00:23:52,990 --> 00:23:56,670 That would mean we excise my face area, we put it in a dish, 543 00:23:56,670 --> 00:23:58,020 keep all the neurons alive. 544 00:23:58,020 --> 00:23:59,310 Let's pretend we can do that. 545 00:23:59,310 --> 00:24:00,768 I'm sure Ed Boyden could figure out 546 00:24:00,768 --> 00:24:03,030 how to do that in a weekend. 547 00:24:03,030 --> 00:24:04,990 And so we have this thing alive in a dish, 548 00:24:04,990 --> 00:24:06,310 can it do face recognition? 549 00:24:06,310 --> 00:24:07,531 Well, of course not. 550 00:24:07,531 --> 00:24:10,030 You got to get the information in there in the right format. 551 00:24:10,030 --> 00:24:11,904 And if information doesn't get out and inform 552 00:24:11,904 --> 00:24:15,330 the rest of a brain, it doesn't house a face percept, right? 553 00:24:15,330 --> 00:24:19,417 So you need things to be connected up, 554 00:24:19,417 --> 00:24:21,750 and you need lots of other brain regions to be involved. 555 00:24:21,750 --> 00:24:25,170 So let's distinguish whether this brain region is 556 00:24:25,170 --> 00:24:27,696 functionally specific for a process from whether it's 557 00:24:27,696 --> 00:24:29,070 sufficient for the whole process. 558 00:24:29,070 --> 00:24:31,500 Of course it's not sufficient. 559 00:24:31,500 --> 00:24:32,100 All right. 560 00:24:32,100 --> 00:24:35,700 I know you guys would never say anything so dumb. 561 00:24:35,700 --> 00:24:36,350 OK. 562 00:24:36,350 --> 00:24:38,610 A question of connectivity-- 563 00:24:38,610 --> 00:24:43,730 so people often say, oh, well, this region 564 00:24:43,730 --> 00:24:47,040 is part of a network, period. 565 00:24:47,040 --> 00:24:49,260 And my reaction is, duh. 566 00:24:49,260 --> 00:24:51,090 Of course it's part of a network. 567 00:24:51,090 --> 00:24:52,530 Everything's part of a network. 568 00:24:52,530 --> 00:24:54,840 In no way does that engage with the question 569 00:24:54,840 --> 00:24:57,500 of whether that region is functionally specific. 570 00:24:57,500 --> 00:25:00,380 A functionally specific region of course is part of a network. 571 00:25:00,380 --> 00:25:02,130 It talks to other brain regions. 572 00:25:02,130 --> 00:25:04,500 Those other brain regions may play an important role 573 00:25:04,500 --> 00:25:06,430 in its processing, sure. 574 00:25:06,430 --> 00:25:08,430 At the very least, they're necessary for getting 575 00:25:08,430 --> 00:25:10,810 the information in and out and using it. 576 00:25:10,810 --> 00:25:11,930 OK? 577 00:25:11,930 --> 00:25:14,250 OK. 578 00:25:14,250 --> 00:25:15,690 All right. 579 00:25:15,690 --> 00:25:17,750 The final thing that people confuse it 580 00:25:17,750 --> 00:25:20,820 with functional specificity is innateness. 581 00:25:20,820 --> 00:25:23,040 This is a very different concept. 582 00:25:23,040 --> 00:25:24,840 Just because we have some particular part 583 00:25:24,840 --> 00:25:27,780 of the brain for which we make it really strong evidence 584 00:25:27,780 --> 00:25:30,750 that it's very specifically involved in mental process x, 585 00:25:30,750 --> 00:25:31,470 that's cool. 586 00:25:31,470 --> 00:25:32,820 That's important. 587 00:25:32,820 --> 00:25:35,880 That's completely orthogonal to how it got wired up 588 00:25:35,880 --> 00:25:39,240 and whether it's innately specified in the genome 589 00:25:39,240 --> 00:25:40,290 that whole circuit-- 590 00:25:40,290 --> 00:25:42,780 or whether that circuit is instructed 591 00:25:42,780 --> 00:25:45,060 by experience over development, or as 592 00:25:45,060 --> 00:25:47,790 in the usual case, very complicated combinations 593 00:25:47,790 --> 00:25:49,200 of those two. 594 00:25:49,200 --> 00:25:51,810 So just to remind you that functional specificity 595 00:25:51,810 --> 00:25:54,270 is a different question from innateness. 596 00:25:54,270 --> 00:25:56,970 And one way you can see that very clearly 597 00:25:56,970 --> 00:25:59,310 is to consider the case of the visual word form 598 00:25:59,310 --> 00:26:02,880 area, about which I'll show you some data in a moment. 599 00:26:02,880 --> 00:26:05,880 The visual word form area responds selectively 600 00:26:05,880 --> 00:26:10,350 to words and letter strings in an orthography you know, 601 00:26:10,350 --> 00:26:12,630 not an orthography you don't know. 602 00:26:12,630 --> 00:26:14,490 It's very anatomically stereotyped. 603 00:26:14,490 --> 00:26:17,910 Mine is approximately right there, 604 00:26:17,910 --> 00:26:20,760 and so is yours in your brain. 605 00:26:20,760 --> 00:26:23,100 And it responds to orthographies you know. 606 00:26:23,100 --> 00:26:25,020 If you can read Arabic and Hebrew, 607 00:26:25,020 --> 00:26:26,640 yours also responds when you look 608 00:26:26,640 --> 00:26:28,230 at words in Arabic and Hebrew. 609 00:26:28,230 --> 00:26:31,500 If you can't, it doesn't, or it responds a whole lot less. 610 00:26:31,500 --> 00:26:34,650 So that's a function of your individual experience, not 611 00:26:34,650 --> 00:26:36,240 your ancestor's experience. 612 00:26:36,240 --> 00:26:38,580 It has strong functional specificity, 613 00:26:38,580 --> 00:26:41,430 and yet, its functional specificity is not innate. 614 00:26:41,430 --> 00:26:43,700 So this idea that I'm staking out here 615 00:26:43,700 --> 00:26:45,090 has become kind of unpopular. 616 00:26:45,090 --> 00:26:46,510 It's very trendy to say, of course 617 00:26:46,510 --> 00:26:49,390 we know the brain doesn't have specialized components. 618 00:26:49,390 --> 00:26:52,150 So for example, here's from a textbook. 619 00:26:52,150 --> 00:26:54,420 Scott Huettel-- unlike the phrenologists 620 00:26:54,420 --> 00:26:57,810 who believe this very stupid idea that very complex traits 621 00:26:57,810 --> 00:26:59,910 are associated with discrete brain regions, 622 00:26:59,910 --> 00:27:02,940 modern researchers recognize that a single brain 623 00:27:02,940 --> 00:27:05,234 region may participate in more than one function. 624 00:27:05,234 --> 00:27:07,650 Well, he built in the hedge word "may," so we can't really 625 00:27:07,650 --> 00:27:08,191 have a fight. 626 00:27:08,191 --> 00:27:12,180 But he's trying to stake out this different view . 627 00:27:12,180 --> 00:27:14,490 Lisa Feldman Barrett-- I haven't met her, 628 00:27:14,490 --> 00:27:17,430 but she's driving me insane, most recently 629 00:27:17,430 --> 00:27:20,370 by proclaiming all kinds of things in The New York Times 630 00:27:20,370 --> 00:27:21,510 just a few weeks ago. 631 00:27:21,510 --> 00:27:24,630 Quote, "in general, the workings of the brain 632 00:27:24,630 --> 00:27:26,400 are not one to one, whereby a given 633 00:27:26,400 --> 00:27:30,870 region has a distinct psychological purpose." 634 00:27:30,870 --> 00:27:32,691 Well, she's got hedge words "in general." 635 00:27:32,691 --> 00:27:33,690 We all have hedge words. 636 00:27:33,690 --> 00:27:36,090 But basically, what she's reasoning from 637 00:27:36,090 --> 00:27:40,230 is the fact that her data suggests that specific emotions 638 00:27:40,230 --> 00:27:43,530 don't inhabit specific brain regions from the idea 639 00:27:43,530 --> 00:27:45,990 that the whole brain has no localization of function. 640 00:27:45,990 --> 00:27:47,640 Well, that's idiotic. 641 00:27:47,640 --> 00:27:50,070 It's just idiotic, right? 642 00:27:50,070 --> 00:27:53,610 So I hope that people will stop these fast and lose arguments. 643 00:27:53,610 --> 00:27:55,110 But here's my favorite-- 644 00:27:55,110 --> 00:27:56,730 this old coot Uttal. 645 00:27:56,730 --> 00:27:58,440 I know this is going to be on the web, 646 00:27:58,440 --> 00:28:02,550 and here I am carrying on as if we are-- anyway, whatever. 647 00:28:02,550 --> 00:28:04,045 This guy cracks me up. 648 00:28:04,045 --> 00:28:04,920 He's been publishing. 649 00:28:04,920 --> 00:28:07,440 Every year, he publishes another book 650 00:28:07,440 --> 00:28:09,120 going after functional MRI. 651 00:28:09,120 --> 00:28:11,010 Any studies using brain images that 652 00:28:11,010 --> 00:28:13,530 report single areas of activation exclusively 653 00:28:13,530 --> 00:28:15,960 associated with a particular cognitive process 654 00:28:15,960 --> 00:28:18,900 should be a priori considered to be artifacts of the arbitrary 655 00:28:18,900 --> 00:28:20,700 threshold set by the investigators 656 00:28:20,700 --> 00:28:22,540 and seriously questioned. 657 00:28:22,540 --> 00:28:24,970 You go. 658 00:28:24,970 --> 00:28:27,840 So anyway, that's fun. 659 00:28:27,840 --> 00:28:31,221 Anyway, my point is just that we should engage in the data, 660 00:28:31,221 --> 00:28:31,720 right? 661 00:28:31,720 --> 00:28:34,640 This isn't like an ideology, where we can just 662 00:28:34,640 --> 00:28:36,100 proclaim our opinions. 663 00:28:36,100 --> 00:28:37,840 There are data that speak to it. 664 00:28:37,840 --> 00:28:40,541 So let me show you some of mine. 665 00:28:40,541 --> 00:28:41,040 OK. 666 00:28:41,040 --> 00:28:44,152 So what would be evidence of functional specificity? 667 00:28:44,152 --> 00:28:45,610 There are lots of ways of doing it. 668 00:28:45,610 --> 00:28:48,810 The way I like to do it is something called a functional 669 00:28:48,810 --> 00:28:51,400 region of interest method. 670 00:28:51,400 --> 00:28:54,000 The problem is that although there 671 00:28:54,000 --> 00:28:57,450 are very systematic regularities in the functional organization 672 00:28:57,450 --> 00:29:00,210 of the brain, each of these regions that I'm talking about 673 00:29:00,210 --> 00:29:02,250 is in approximately the same location 674 00:29:02,250 --> 00:29:03,900 in each normal subject. 675 00:29:03,900 --> 00:29:07,260 Their actual location varies a bit from subject to subject. 676 00:29:07,260 --> 00:29:09,930 So if you do the standard thing of aligning brains 677 00:29:09,930 --> 00:29:12,990 and averaging across them, you get a lot of mush, 678 00:29:12,990 --> 00:29:16,290 and yet there isn't much mush in each subject individually. 679 00:29:16,290 --> 00:29:17,820 And so to deal with that problem-- 680 00:29:17,820 --> 00:29:19,654 and to deal with a bunch of other problems-- 681 00:29:19,654 --> 00:29:21,153 we use something called a functional 682 00:29:21,153 --> 00:29:22,290 region of interest method. 683 00:29:22,290 --> 00:29:25,290 And that means if you want to study a given region, 684 00:29:25,290 --> 00:29:27,660 you find it in that subject individually. 685 00:29:27,660 --> 00:29:30,630 And then once you've found it with a simple contrast-- you 686 00:29:30,630 --> 00:29:32,520 want to find a face region, you find a region 687 00:29:32,520 --> 00:29:34,290 that responds more when people look at faces 688 00:29:34,290 --> 00:29:35,581 than when they look at objects. 689 00:29:35,581 --> 00:29:37,500 Now you found it in that subject. 690 00:29:37,500 --> 00:29:40,420 It's these 85 voxels right there in that subject. 691 00:29:40,420 --> 00:29:43,800 Now we run a new experiment to test more interesting questions 692 00:29:43,800 --> 00:29:46,590 about it, and we measure the response in those voxels. 693 00:29:46,590 --> 00:29:47,580 OK? 694 00:29:47,580 --> 00:29:50,430 That also has the advantage that the data you plot and look at 695 00:29:50,430 --> 00:29:53,070 is independent of the way you found those voxels-- 696 00:29:53,070 --> 00:29:55,530 a very important problem in a lot of functional 697 00:29:55,530 --> 00:29:58,350 neuroimaging, where people have non-independent statistical 698 00:29:58,350 --> 00:30:00,250 problems with their data analysis. 699 00:30:00,250 --> 00:30:03,370 If you have a functional region of interest that's localized 700 00:30:03,370 --> 00:30:05,740 independently of the data you look at in it, 701 00:30:05,740 --> 00:30:07,370 you get out of that problem. 702 00:30:07,370 --> 00:30:09,400 It's also a huge benefit, because one 703 00:30:09,400 --> 00:30:12,286 of the central problems with functional brain imaging, 704 00:30:12,286 --> 00:30:13,660 which I think has led to the fact 705 00:30:13,660 --> 00:30:17,320 that a large percent of the published neuroimaging findings 706 00:30:17,320 --> 00:30:20,110 are probably noise, is that there are just 707 00:30:20,110 --> 00:30:21,520 too many degrees of freedom. 708 00:30:21,520 --> 00:30:23,230 You have tens of thousands of voxels. 709 00:30:23,230 --> 00:30:26,380 You have loads of different places to look and ways 710 00:30:26,380 --> 00:30:27,844 to analyze your data. 711 00:30:27,844 --> 00:30:30,010 One of the things I love dearly about the functional 712 00:30:30,010 --> 00:30:33,760 region of interest method is that you tie your hands 713 00:30:33,760 --> 00:30:35,920 in a really good way, right? 714 00:30:35,920 --> 00:30:38,870 So you specify in advance exactly where 715 00:30:38,870 --> 00:30:41,500 you're going to look, and you specify exactly how you're 716 00:30:41,500 --> 00:30:43,190 going to quantify the response. 717 00:30:43,190 --> 00:30:45,270 And so you have no degrees of freedom, 718 00:30:45,270 --> 00:30:47,720 and that gives you a huge statistical advantage. 719 00:30:47,720 --> 00:30:50,860 And it means you're less likely to be inadvertently publishing 720 00:30:50,860 --> 00:30:52,080 papers on noise. 721 00:30:52,080 --> 00:30:52,900 OK. 722 00:30:52,900 --> 00:30:56,080 So that's the functional region of interest method. 723 00:30:56,080 --> 00:30:57,790 We've done loads of these experiments. 724 00:30:57,790 --> 00:31:00,010 Here's just from a current experiment in my lab 725 00:31:00,010 --> 00:31:02,406 being conducted by Zeynep Saygin. 726 00:31:02,406 --> 00:31:04,780 She's actually looking at connectivity of different brain 727 00:31:04,780 --> 00:31:06,430 regions using a different method I probably 728 00:31:06,430 --> 00:31:07,660 won't have time to talk about. 729 00:31:07,660 --> 00:31:08,285 It's very cool. 730 00:31:08,285 --> 00:31:10,270 But in the process, she's run a whole bunch 731 00:31:10,270 --> 00:31:11,810 of functional localizers. 732 00:31:11,810 --> 00:31:15,100 And so we can look in her data at the response of the fusiform 733 00:31:15,100 --> 00:31:17,860 face area to a whole bunch of different conditions. 734 00:31:17,860 --> 00:31:20,120 So these are a bunch of auditory language conditions, 735 00:31:20,120 --> 00:31:21,940 so, OK, not too surprising. 736 00:31:21,940 --> 00:31:24,040 It doesn't respond very much to those. 737 00:31:24,040 --> 00:31:26,260 They're presented auditorily, but these are all 738 00:31:26,260 --> 00:31:27,490 visual stimuli here. 739 00:31:27,490 --> 00:31:29,530 The two yellow bars are faces. 740 00:31:29,530 --> 00:31:31,420 This is line drawings of faces. 741 00:31:31,420 --> 00:31:34,480 This is color video clips of faces-- 742 00:31:34,480 --> 00:31:36,280 strong responses to both. 743 00:31:36,280 --> 00:31:38,970 And all of these other conditions-- 744 00:31:38,970 --> 00:31:41,740 line drawings of objects, movies of objects, movies of scenes, 745 00:31:41,740 --> 00:31:44,480 scrambled objects, words, scrambled words, bodies-- 746 00:31:44,480 --> 00:31:46,740 all produce much lower responses. 747 00:31:46,740 --> 00:31:47,680 OK? 748 00:31:47,680 --> 00:31:52,400 So I would say this is pretty strong selectivity. 749 00:31:52,400 --> 00:31:55,150 It's been tested against lots of alternatives, 750 00:31:55,150 --> 00:31:58,720 only a tiny percent of which are shown here. 751 00:31:58,720 --> 00:32:01,024 As I mentioned before, it's present in more or less 752 00:32:01,024 --> 00:32:03,190 the same place and pretty much every normal subject. 753 00:32:03,190 --> 00:32:06,710 I think it's just a basic piece of mental architecture. 754 00:32:06,710 --> 00:32:10,090 Now, this is a very simple univariate measure. 755 00:32:10,090 --> 00:32:11,830 We're just measuring the very crude thing 756 00:32:11,830 --> 00:32:14,830 of the overall magnitude of MRI response 757 00:32:14,830 --> 00:32:17,590 in that region to these conditions. 758 00:32:17,590 --> 00:32:19,510 There are legitimate counter-arguments 759 00:32:19,510 --> 00:32:22,420 to the simple-minded view I'm putting forth, 760 00:32:22,420 --> 00:32:23,680 and we should consider them. 761 00:32:23,680 --> 00:32:25,600 I think the most important one comes 762 00:32:25,600 --> 00:32:28,150 from pattern analysis methods, which I will talk about 763 00:32:28,150 --> 00:32:30,610 if I get there. 764 00:32:30,610 --> 00:32:33,790 And importantly, these data don't tell us 765 00:32:33,790 --> 00:32:37,470 about the causal role of that region. 766 00:32:37,470 --> 00:32:38,900 We'll return to those. 767 00:32:38,900 --> 00:32:42,280 However, the point is, before we blithely 768 00:32:42,280 --> 00:32:45,640 say it's not fashionable to talk about functional specificity, 769 00:32:45,640 --> 00:32:48,844 we need counterarguments to data like this. 770 00:32:48,844 --> 00:32:49,760 They're pretty strong. 771 00:32:49,760 --> 00:32:53,800 And that's just one example, to show you just a few others 772 00:32:53,800 --> 00:32:57,277 from Zeynep's paper. 773 00:32:57,277 --> 00:32:58,860 OK, so this is what I just showed you, 774 00:32:58,860 --> 00:33:01,090 but I'm in the same experiment. 775 00:33:01,090 --> 00:33:02,590 We can look at other brain regions. 776 00:33:02,590 --> 00:33:03,090 OK. 777 00:33:03,090 --> 00:33:06,700 So this is a bottom surface of the brain 778 00:33:06,700 --> 00:33:10,660 there, so this is the occipital pole, front of the head, bottom 779 00:33:10,660 --> 00:33:11,710 of the temporal lobe. 780 00:33:11,710 --> 00:33:14,560 That face area is the region in yellow in this subject. 781 00:33:14,560 --> 00:33:17,470 This purple region is that visual word form area 782 00:33:17,470 --> 00:33:19,900 that I mentioned, and here is its response magnitude 783 00:33:19,900 --> 00:33:21,400 across a whole bunch of subjects, 784 00:33:21,400 --> 00:33:23,890 localizing and then independently testing. 785 00:33:23,890 --> 00:33:27,280 The purple bars are when subjects are looking 786 00:33:27,280 --> 00:33:28,986 at visually presented words. 787 00:33:28,986 --> 00:33:30,610 And again, all these other conditions-- 788 00:33:30,610 --> 00:33:34,215 faces, objects, bodies, scenes, listening to words, 789 00:33:34,215 --> 00:33:35,965 all of those things-- much lower response. 790 00:33:39,140 --> 00:33:42,550 In the same experiment, we can also look at a set of regions 791 00:33:42,550 --> 00:33:43,900 that respond to speech. 792 00:33:43,900 --> 00:33:46,120 I mentioned those very briefly in my introduction 793 00:33:46,120 --> 00:33:47,690 a few days ago. 794 00:33:47,690 --> 00:33:50,350 These are regions a number of people have found. 795 00:33:50,350 --> 00:33:54,940 In this case, they're immediately below 796 00:33:54,940 --> 00:33:59,170 or lateral to primary auditory cortex in humans, 797 00:33:59,170 --> 00:34:03,490 interestingly situated right between primary auditory cortex 798 00:34:03,490 --> 00:34:04,840 and language sensitive regions. 799 00:34:04,840 --> 00:34:07,060 Right between is the set of regions that respond 800 00:34:07,060 --> 00:34:09,250 to the sounds of speech-- 801 00:34:09,250 --> 00:34:12,820 not to the content of language, but the sounds of speech. 802 00:34:12,820 --> 00:34:17,800 And so this is when people are saying stuff like, 803 00:34:17,800 --> 00:34:18,909 "ba da ga ba da ga." 804 00:34:18,909 --> 00:34:20,450 So they're just lying in the scanner, 805 00:34:20,450 --> 00:34:21,760 saying, "ba da ga ba da ga." 806 00:34:21,760 --> 00:34:24,820 And here's when they're tapping their fingers 807 00:34:24,820 --> 00:34:27,040 in a systematic order. 808 00:34:27,040 --> 00:34:29,776 Here's when they're listening to sentences. 809 00:34:29,776 --> 00:34:31,150 Importantly, this is when they're 810 00:34:31,150 --> 00:34:34,120 listening to jabberwocky gobbledygook 811 00:34:34,120 --> 00:34:35,469 that's meaningless. 812 00:34:35,469 --> 00:34:37,420 So no meaning, but phonemes-- 813 00:34:37,420 --> 00:34:38,600 same response. 814 00:34:38,600 --> 00:34:40,960 That's what tells us that this region is involved 815 00:34:40,960 --> 00:34:42,520 in processing the sounds of speech, 816 00:34:42,520 --> 00:34:45,850 not the content of language, and load everything else. 817 00:34:45,850 --> 00:34:50,080 So other things-- moving outside of perceptual regions, 818 00:34:50,080 --> 00:34:51,520 you might say, OK, fine. 819 00:34:51,520 --> 00:34:53,912 Perception is an inherently modular process. 820 00:34:53,912 --> 00:34:55,870 There's different kinds of perceptual problems, 821 00:34:55,870 --> 00:34:57,220 that make sense. 822 00:34:57,220 --> 00:34:59,470 But high level cognition-- we wouldn't really 823 00:34:59,470 --> 00:35:01,940 have functional specificity for that. 824 00:35:01,940 --> 00:35:03,890 But oh, yes, we do. 825 00:35:03,890 --> 00:35:06,297 Here are some language regions. 826 00:35:06,297 --> 00:35:08,630 There's a bunch of them in the temporal and frontal lobe 827 00:35:08,630 --> 00:35:10,910 that have been known since Wernicke and Broca. 828 00:35:10,910 --> 00:35:13,220 But now, with functional MRI, we can identify them 829 00:35:13,220 --> 00:35:16,460 in individual subjects and go back and repeatedly query 830 00:35:16,460 --> 00:35:18,770 them and say, are they involved in all 831 00:35:18,770 --> 00:35:20,730 of these other mental processes? 832 00:35:20,730 --> 00:35:24,496 So this is now the response in a language region-- 833 00:35:24,496 --> 00:35:26,120 so identified, here's the response when 834 00:35:26,120 --> 00:35:27,950 you're listening to sentences. 835 00:35:27,950 --> 00:35:32,330 This is when you're listening to jabberwocky nonsense strings. 836 00:35:32,330 --> 00:35:34,490 Here's when you're saying "ba da ga ba da ga." 837 00:35:34,490 --> 00:35:36,390 It's not just speech sounds. 838 00:35:36,390 --> 00:35:40,280 Here's when you're listening to synthetically decomposed speech 839 00:35:40,280 --> 00:35:41,690 sounds that are acoustically very 840 00:35:41,690 --> 00:35:44,002 similar to the jabberwocky speech. 841 00:35:44,002 --> 00:35:45,710 It's just not interested in those things. 842 00:35:45,710 --> 00:35:47,293 It seems to be interested in something 843 00:35:47,293 --> 00:35:49,790 more like the meaning of a sentence. 844 00:35:49,790 --> 00:35:52,415 And just to show you some other data we have on this, 845 00:35:52,415 --> 00:35:57,310 this is data from Ev Fedorenko, who has tested this region. 846 00:35:57,310 --> 00:36:00,950 Now, this is sort of roughly Broca's area, 847 00:36:00,950 --> 00:36:05,540 the main mental functions that people have argued 848 00:36:05,540 --> 00:36:08,347 overlap in the brain with language. 849 00:36:08,347 --> 00:36:10,430 Namely-- sorry, this is probably hard to see here, 850 00:36:10,430 --> 00:36:14,270 but arithmetic, so we have difficult and easy 851 00:36:14,270 --> 00:36:16,070 mental arithmetic. 852 00:36:16,070 --> 00:36:19,190 Intact and scrambled music in pink. 853 00:36:19,190 --> 00:36:21,130 A bunch of working memory tasks-- 854 00:36:21,130 --> 00:36:23,660 spatial working memory and verbal working memory-- 855 00:36:23,660 --> 00:36:25,954 and a bunch of cognitive control tasks-- 856 00:36:25,954 --> 00:36:27,620 just kind of an attention demanding task 857 00:36:27,620 --> 00:36:31,140 where you have to switch between tasks and stuff like that. 858 00:36:31,140 --> 00:36:34,010 And here is the response profile in that region. 859 00:36:34,010 --> 00:36:37,850 Reading sentences, reading non-word strings. 860 00:36:37,850 --> 00:36:39,710 All of those other tasks, both the difficult 861 00:36:39,710 --> 00:36:41,390 and the easy version-- 862 00:36:41,390 --> 00:36:43,900 no response at all. 863 00:36:43,900 --> 00:36:46,510 That's extreme functional specificity, right? 864 00:36:46,510 --> 00:36:48,130 It's not that we've tested everything, 865 00:36:48,130 --> 00:36:49,640 there's more to be done. 866 00:36:49,640 --> 00:36:54,310 But the first pass querying of do those language regions 867 00:36:54,310 --> 00:36:56,650 engage in all of these other things that people thought 868 00:36:56,650 --> 00:36:58,120 might overlap with language? 869 00:36:58,120 --> 00:37:01,420 The answer is no, they don't. 870 00:37:01,420 --> 00:37:03,880 And I think that's really deep and interesting, 871 00:37:03,880 --> 00:37:07,540 because it means that this basic question that we all 872 00:37:07,540 --> 00:37:09,850 start asking ourselves when we're young 873 00:37:09,850 --> 00:37:13,480 is, what is the relationship between language and thought? 874 00:37:13,480 --> 00:37:15,749 I know Liz disagrees with me somewhat on this. 875 00:37:15,749 --> 00:37:17,290 That's because she's very articulate, 876 00:37:17,290 --> 00:37:20,080 and she doesn't feel the difference between an idea 877 00:37:20,080 --> 00:37:21,460 and its articulation. 878 00:37:21,460 --> 00:37:22,510 I'm less articulate. 879 00:37:22,510 --> 00:37:25,442 It's very obvious to me they're different things. 880 00:37:25,442 --> 00:37:26,650 No, it's not the only reason. 881 00:37:26,650 --> 00:37:29,200 She has data, too, and it'd be fun to discuss. 882 00:37:29,200 --> 00:37:33,320 But I think there's a vast gulf between the two 883 00:37:33,320 --> 00:37:36,460 in that many different aspects of cognition 884 00:37:36,460 --> 00:37:39,580 can proceed just fine without language regions. 885 00:37:39,580 --> 00:37:41,440 And actually, the stronger evidence for that 886 00:37:41,440 --> 00:37:43,720 comes not from these functional MRI data, 887 00:37:43,720 --> 00:37:47,500 striking as I think they are, but from patient data. 888 00:37:47,500 --> 00:37:50,980 So Rosemary Varley in England has been testing patients 889 00:37:50,980 --> 00:37:51,990 with global aphasia. 890 00:37:51,990 --> 00:37:53,800 This is this very tragic, horrible thing 891 00:37:53,800 --> 00:37:57,040 that happens in patients who have massive left hemisphere 892 00:37:57,040 --> 00:37:59,800 strokes that pretty much take out 893 00:37:59,800 --> 00:38:02,650 essentially all of their language abilities. 894 00:38:02,650 --> 00:38:06,730 Those people she has shown are intact in their navigation 895 00:38:06,730 --> 00:38:10,390 abilities, their arithmetic abilities, their ability 896 00:38:10,390 --> 00:38:12,970 to solve logic problems, their ability 897 00:38:12,970 --> 00:38:15,670 to think about what other people are thinking, 898 00:38:15,670 --> 00:38:19,880 their ability to appreciate music, and so on and so forth. 899 00:38:19,880 --> 00:38:22,810 So I think there's really a very big difference 900 00:38:22,810 --> 00:38:25,390 between a major part of the system 901 00:38:25,390 --> 00:38:28,000 that you need to understand the meaning of a sentence 902 00:38:28,000 --> 00:38:30,910 and all of those other aspects of thought. 903 00:38:30,910 --> 00:38:32,470 This is just showing you what I mean 904 00:38:32,470 --> 00:38:34,540 by functional specificity-- 905 00:38:34,540 --> 00:38:37,247 what the basic first order evidence is. 906 00:38:37,247 --> 00:38:38,830 And these are just the regions that we 907 00:38:38,830 --> 00:38:41,204 happen to have in this study so I could make a new slide. 908 00:38:41,204 --> 00:38:45,280 But for lots of other perceptual and cognitive functions, 909 00:38:45,280 --> 00:38:47,620 people have found quite specific brain regions 910 00:38:47,620 --> 00:38:50,830 for perceiving bodies and scenes, of course, motion. 911 00:38:50,830 --> 00:38:54,280 The area MT has been studied for a long time-- 912 00:38:54,280 --> 00:38:56,140 regions that are quite specifically 913 00:38:56,140 --> 00:38:58,150 involved in processing shape. 914 00:38:58,150 --> 00:39:00,640 We've been studying color processing regions recently. 915 00:39:00,640 --> 00:39:02,080 They're not as selective for color 916 00:39:02,080 --> 00:39:03,871 as some of these other regions, but they're 917 00:39:03,871 --> 00:39:07,830 very anatomically consistent. 918 00:39:07,830 --> 00:39:12,460 And things I mentioned before in my brief introduction-- regions 919 00:39:12,460 --> 00:39:15,700 that are specifically involved in processing pitch information 920 00:39:15,700 --> 00:39:18,010 and music information, and as you'll 921 00:39:18,010 --> 00:39:20,950 hear next week from Rebecca Saxe, theory 922 00:39:20,950 --> 00:39:23,200 of mind or thinking about other people's thoughts. 923 00:39:23,200 --> 00:39:28,870 And so there's quite a litany of mental functions 924 00:39:28,870 --> 00:39:31,570 that have brain regions that are quite specifically engaged 925 00:39:31,570 --> 00:39:32,860 to that mental function. 926 00:39:32,860 --> 00:39:34,030 And each of these-- 927 00:39:34,030 --> 00:39:37,840 to varying degrees, but to some appreciable degree-- 928 00:39:37,840 --> 00:39:40,480 have corroborating evidence from patients 929 00:39:40,480 --> 00:39:42,670 who have that specific deficit. 930 00:39:42,670 --> 00:39:44,830 So that shows that each of these is 931 00:39:44,830 --> 00:39:47,260 likely to be not only activated during, 932 00:39:47,260 --> 00:39:51,610 but causally involved in its mental function. 933 00:39:51,610 --> 00:39:55,157 And as I mentioned, there are actually good counter-arguments 934 00:39:55,157 --> 00:39:56,740 to some of the things I've been making 935 00:39:56,740 --> 00:39:57,970 that are worth discussing. 936 00:39:57,970 --> 00:40:02,880 I think the pattern analysis data is the strongest. 937 00:40:02,880 --> 00:40:04,840 Oh, and I do need to take a few more minutes. 938 00:40:04,840 --> 00:40:06,100 Just like five or something? 939 00:40:06,100 --> 00:40:07,120 OK. 940 00:40:07,120 --> 00:40:09,610 So all of that's to say, so here's 941 00:40:09,610 --> 00:40:10,900 roughly where we are now. 942 00:40:10,900 --> 00:40:14,262 There are counter-arguments, but loose talk about, oh, there's 943 00:40:14,262 --> 00:40:15,970 no localization of function in the brain. 944 00:40:15,970 --> 00:40:17,386 You got to engage with us at first 945 00:40:17,386 --> 00:40:20,620 and give me a serious counter-argument. 946 00:40:20,620 --> 00:40:21,760 OK. 947 00:40:21,760 --> 00:40:24,730 Finally, I want to say that it's not that the whole brain is 948 00:40:24,730 --> 00:40:26,020 like this, right? 949 00:40:26,020 --> 00:40:29,230 There are big gray patches where we haven't figured 950 00:40:29,230 --> 00:40:33,160 out what it's doing, but there are also substantial patches 951 00:40:33,160 --> 00:40:35,590 that have already been shown to be, 952 00:40:35,590 --> 00:40:37,240 in some sense, the opposite of this. 953 00:40:37,240 --> 00:40:40,090 Regions that are engaged in almost any difficult task 954 00:40:40,090 --> 00:40:41,659 you do at all. 955 00:40:41,659 --> 00:40:43,450 And I think this is a very interesting part 956 00:40:43,450 --> 00:40:45,741 of the whole story of the architecture of intelligence, 957 00:40:45,741 --> 00:40:48,130 so I'm going to take five minutes and tell you about it. 958 00:40:48,130 --> 00:40:51,130 This work is primarily the work of John Duncan in England. 959 00:40:51,130 --> 00:40:54,960 And he's been pointing out for about 15 years 960 00:40:54,960 --> 00:40:59,530 that there are regions in the parietal and frontal lobe shown 961 00:40:59,530 --> 00:41:04,270 here that are engaged in pretty much any difficult task you do. 962 00:41:04,270 --> 00:41:06,850 Any time you increase the difficulty of a task-- 963 00:41:06,850 --> 00:41:10,090 whether it's perceptual or high level cognitive-- 964 00:41:10,090 --> 00:41:12,740 those regions turn on differentially. 965 00:41:12,740 --> 00:41:14,740 And so that's why he calls them multiple demand. 966 00:41:14,740 --> 00:41:16,864 They respond to multiple different kinds of demand. 967 00:41:21,460 --> 00:41:23,140 Duncan argues that these regions are 968 00:41:23,140 --> 00:41:24,880 related to fluid intelligence. 969 00:41:24,880 --> 00:41:27,130 So remember Spearman, who I started with, 970 00:41:27,130 --> 00:41:30,930 who talks about the general factor, g. 971 00:41:30,930 --> 00:41:33,070 Well, Duncan thinks that basically, this 972 00:41:33,070 --> 00:41:34,390 is the seat of g-- 973 00:41:34,390 --> 00:41:37,370 these regions here-- to oversimplify his argument. 974 00:41:37,370 --> 00:41:40,130 There's multiple sources of evidence for that. 975 00:41:40,130 --> 00:41:43,520 And one is, well, they're strongly 976 00:41:43,520 --> 00:41:45,846 activated when you do classic g-loading tasks. 977 00:41:45,846 --> 00:41:46,970 That's not that surprising. 978 00:41:46,970 --> 00:41:49,053 They're activated in all different kinds of tasks. 979 00:41:49,053 --> 00:41:52,640 More interestingly, he did a large patient study, 980 00:41:52,640 --> 00:41:56,840 where they found 80 or so neuropsych patients 981 00:41:56,840 --> 00:41:58,340 in their patient database. 982 00:41:58,340 --> 00:42:01,100 And they identified the locus of the brain damage 983 00:42:01,100 --> 00:42:02,810 in each of those patients. 984 00:42:02,810 --> 00:42:06,470 And what they did was they measured post-injury IQ. 985 00:42:06,470 --> 00:42:09,920 They estimated from a variety of sources pre-injury IQ. 986 00:42:09,920 --> 00:42:13,310 And they asked, how much does your IQ go down 987 00:42:13,310 --> 00:42:16,350 after brain damage as a function of one, 988 00:42:16,350 --> 00:42:18,970 the volume of tissue you lost in the brain damage, 989 00:42:18,970 --> 00:42:22,730 and two, the locus of tissue? 990 00:42:22,730 --> 00:42:25,040 And basically, what he finds is if you 991 00:42:25,040 --> 00:42:28,850 lose tissue in these regions, your IQ goes down. 992 00:42:28,850 --> 00:42:31,010 If you lose tissue elsewhere, you 993 00:42:31,010 --> 00:42:36,335 may become paralyzed or aphasic or prosopagnosia. 994 00:42:36,335 --> 00:42:38,840 Your IQ does not go down. 995 00:42:38,840 --> 00:42:42,470 In fact, he made a kind of ghoulish calculation 996 00:42:42,470 --> 00:42:45,560 that you lose 6 and 1/2 IQ points for every 10 997 00:42:45,560 --> 00:42:50,330 cubic centimeters of this region of cortex, 998 00:42:50,330 --> 00:42:53,900 and almost nothing for the rest of the brain. 999 00:42:53,900 --> 00:42:55,700 So this is kind of crude. 1000 00:42:55,700 --> 00:42:59,000 It's very imperfect what you can get from patient study, 1001 00:42:59,000 --> 00:43:00,870 but I think it's intriguing. 1002 00:43:00,870 --> 00:43:04,070 And so his suggestion is that in addition to these highly 1003 00:43:04,070 --> 00:43:05,690 specialized cortical regions that we 1004 00:43:05,690 --> 00:43:08,690 use for these particular important tasks, 1005 00:43:08,690 --> 00:43:12,290 we also have this kind of general purpose machinery that 1006 00:43:12,290 --> 00:43:14,622 makes us generically smart. 1007 00:43:14,622 --> 00:43:15,830 And I'm going to skip around. 1008 00:43:15,830 --> 00:43:17,205 We've tested this more seriously. 1009 00:43:17,205 --> 00:43:19,790 He did group analyses, which I don't like. 1010 00:43:19,790 --> 00:43:21,710 We did it in collaboration with him 1011 00:43:21,710 --> 00:43:23,390 with individual subject analyses, 1012 00:43:23,390 --> 00:43:25,760 the most precise measurements we could make, and boy, 1013 00:43:25,760 --> 00:43:26,550 is he right. 1014 00:43:26,550 --> 00:43:28,580 I mean, even to the voxel you can 1015 00:43:28,580 --> 00:43:30,650 find that these regions are engaged 1016 00:43:30,650 --> 00:43:33,530 in seven or eight very, very different kinds 1017 00:43:33,530 --> 00:43:35,030 of cognitive demand-- 1018 00:43:35,030 --> 00:43:38,750 all activate the same voxel differentially. 1019 00:43:38,750 --> 00:43:40,850 So the basic story I'm putting forth here-- 1020 00:43:40,850 --> 00:43:45,620 without the second half of my talk, I'm sorry about that-- 1021 00:43:45,620 --> 00:43:48,352 is that at a macro scale, the architecture 1022 00:43:48,352 --> 00:43:50,810 of human intelligence is that we have these special purpose 1023 00:43:50,810 --> 00:43:55,280 bits for a smallish number of important mental functions, 1024 00:43:55,280 --> 00:43:57,946 not all of them innate-- maybe some of them. 1025 00:43:57,946 --> 00:44:00,290 In addition, we have some general purpose machinery. 1026 00:44:00,290 --> 00:44:02,930 There's loads more that we don't know 1027 00:44:02,930 --> 00:44:06,830 from the precise computations that go on in these things, 1028 00:44:06,830 --> 00:44:11,897 to their connectivity, to the actual precise representations 1029 00:44:11,897 --> 00:44:14,480 that you can see with the neural code if you could measure it, 1030 00:44:14,480 --> 00:44:17,060 which we can't in humans, to the timing 1031 00:44:17,060 --> 00:44:20,210 of these complex interactions, which of them are uniquely 1032 00:44:20,210 --> 00:44:23,870 human, which of them are also present in monkeys. 1033 00:44:23,870 --> 00:44:26,680 And I don't have time to go find the slide, 1034 00:44:26,680 --> 00:44:28,670 but one of the things we've been doing recently 1035 00:44:28,670 --> 00:44:31,370 is looking in the ventral visual pathway 1036 00:44:31,370 --> 00:44:36,320 at the organization of face, place, and color selectivity. 1037 00:44:36,320 --> 00:44:38,720 And what we see is that-- 1038 00:44:38,720 --> 00:44:43,220 we is me and Bevil Conway and Rosa Lafer-Sousa. 1039 00:44:43,220 --> 00:44:45,260 Bevil and Rosa had previously shown 1040 00:44:45,260 --> 00:44:47,510 that on the lateral surface in the monkey, 1041 00:44:47,510 --> 00:44:50,150 you have three bands of selectivity. 1042 00:44:50,150 --> 00:44:53,600 So it goes face selectivity, color selectivity, 1043 00:44:53,600 --> 00:44:55,130 place selectivity, and three bands 1044 00:44:55,130 --> 00:44:58,160 on the side of the temporal lobe in monkeys. 1045 00:44:58,160 --> 00:44:59,240 We find this in humans. 1046 00:44:59,240 --> 00:45:02,870 You have exactly the same organization in the same order, 1047 00:45:02,870 --> 00:45:05,360 but it's rolled around on the ventral surface of the brain 1048 00:45:05,360 --> 00:45:06,440 in the same order-- 1049 00:45:06,440 --> 00:45:09,560 face, color, place-- on the bottom of the brain. 1050 00:45:09,560 --> 00:45:13,010 So we think that whole broad region is homologous 1051 00:45:13,010 --> 00:45:15,021 between monkeys and humans. 1052 00:45:15,021 --> 00:45:16,520 It just rolled around on the bottom. 1053 00:45:16,520 --> 00:45:19,710 Maybe it got pushed over [AUDIO OUT] something. 1054 00:45:19,710 --> 00:45:21,590 And that's not exactly a novel argument. 1055 00:45:21,590 --> 00:45:24,956 Actually, Winrich wrote a paper suggesting this a while back, 1056 00:45:24,956 --> 00:45:27,080 and I think we're starting to see those homologies. 1057 00:45:27,080 --> 00:45:28,760 And the reason that's important is 1058 00:45:28,760 --> 00:45:31,370 that it means that all these questions we desperately 1059 00:45:31,370 --> 00:45:35,990 want to answer about the causal role, connectivity, 1060 00:45:35,990 --> 00:45:40,650 population codes, [AUDIO OUT] interactions between regions, 1061 00:45:40,650 --> 00:45:42,120 development-- all of that that we 1062 00:45:42,120 --> 00:45:44,460 can't answer very well in humans, 1063 00:45:44,460 --> 00:45:46,280 Winrich can answer in monkeys. 1064 00:45:46,280 --> 00:45:49,490 And after a break, he will tell you about all of that. 1065 00:45:49,490 --> 00:45:52,240 [APPLAUSE]