1 00:00:02,464 --> 00:00:02,964 [SQUEAKING] 2 00:00:02,964 --> 00:00:04,446 [RUSTLING] 3 00:00:04,446 --> 00:00:07,410 [CLICKING] 4 00:00:10,880 --> 00:00:13,580 NANCY KANWISHER: Before we get on to the topic for today, 5 00:00:13,580 --> 00:00:16,940 I felt like last Monday's lecture was not my best. 6 00:00:16,940 --> 00:00:17,780 I don't know why. 7 00:00:17,780 --> 00:00:19,363 It's not that I didn't put time on it. 8 00:00:19,363 --> 00:00:22,733 I looked and I had the wrong lecture numbers on slides. 9 00:00:22,733 --> 00:00:23,900 There was all kind of chaos. 10 00:00:23,900 --> 00:00:24,710 I'm sorry about that. 11 00:00:24,710 --> 00:00:26,252 Sometimes you put in a lot of effort, 12 00:00:26,252 --> 00:00:28,670 and you still give a lecture that isn't all that clear. 13 00:00:28,670 --> 00:00:31,620 So let me try to tell you what I thought were the main points. 14 00:00:31,620 --> 00:00:34,730 I started off by saying why it's really fundamentally important 15 00:00:34,730 --> 00:00:37,280 to be able to understand not just what people look 16 00:00:37,280 --> 00:00:40,820 like from the outside, but what we really care about people is 17 00:00:40,820 --> 00:00:43,880 what's going on the inside about their thoughts 18 00:00:43,880 --> 00:00:44,660 and their beliefs. 19 00:00:44,660 --> 00:00:46,460 And we are constantly making inferences 20 00:00:46,460 --> 00:00:50,570 about what people know, and believe, and want, and think. 21 00:00:50,570 --> 00:00:52,940 And we do that all the time to understand 22 00:00:52,940 --> 00:00:54,710 why they're doing something and to predict 23 00:00:54,710 --> 00:00:55,680 what they'll do next. 24 00:00:55,680 --> 00:00:57,530 And so it's fundamentally important. 25 00:00:57,530 --> 00:01:00,570 It's the essence of being a human being in many ways. 26 00:01:00,570 --> 00:01:02,180 It's the essence of literature. 27 00:01:02,180 --> 00:01:05,810 And we do it all the time. 28 00:01:05,810 --> 00:01:09,410 And a classic way that people have studied false beliefs is-- 29 00:01:09,410 --> 00:01:12,590 or thinking about other people's thoughts is with 30 00:01:12,590 --> 00:01:14,060 the false-belief task-- 31 00:01:14,060 --> 00:01:16,310 the Sally-Anne task that I described. 32 00:01:16,310 --> 00:01:18,590 And the reason people use the false-belief task 33 00:01:18,590 --> 00:01:21,810 rather than just a belief task is if the beliefs are true, 34 00:01:21,810 --> 00:01:24,200 you can answer what somebody will do based on the world, 35 00:01:24,200 --> 00:01:25,770 not based on their mind. 36 00:01:25,770 --> 00:01:28,125 And so to unconfound the two, we use a false belief 37 00:01:28,125 --> 00:01:30,000 that's different than the state of the world. 38 00:01:30,000 --> 00:01:32,390 So we can be sure we're asking people 39 00:01:32,390 --> 00:01:35,750 what will happen next based on what that person believes. 40 00:01:35,750 --> 00:01:40,850 And through decades of use of the Sally-Anne false-belief 41 00:01:40,850 --> 00:01:44,330 task and variations of it, it's clear 42 00:01:44,330 --> 00:01:46,820 that there's a very distinctive developmental time 43 00:01:46,820 --> 00:01:49,520 course and ability to solve this problem. 44 00:01:49,520 --> 00:01:51,170 Five-year-olds solve it no problem. 45 00:01:51,170 --> 00:01:54,170 Three-year-olds systematically fail. 46 00:01:54,170 --> 00:01:58,610 And people with autism typically get-- 47 00:01:58,610 --> 00:02:01,550 pass this task late or not at all. 48 00:02:01,550 --> 00:02:03,650 High-functioning people with autism 49 00:02:03,650 --> 00:02:06,930 pass the task, just later, like 7, 8, 9 years, 50 00:02:06,930 --> 00:02:08,270 not five years old. 51 00:02:08,270 --> 00:02:09,020 OK? 52 00:02:09,020 --> 00:02:11,840 So that's the kind of behavioral background evidence 53 00:02:11,840 --> 00:02:14,300 that there's something distinctive about thinking 54 00:02:14,300 --> 00:02:16,123 about other people's minds. 55 00:02:16,123 --> 00:02:17,540 Then we considered whether there's 56 00:02:17,540 --> 00:02:19,550 special brain mechanisms. 57 00:02:19,550 --> 00:02:20,990 And I argued that, yes, there are. 58 00:02:20,990 --> 00:02:21,990 There's a bunch of them. 59 00:02:21,990 --> 00:02:24,980 But the most impressively selective one 60 00:02:24,980 --> 00:02:27,585 is the TBJ shown up there. 61 00:02:27,585 --> 00:02:29,210 And the evidence that it's specifically 62 00:02:29,210 --> 00:02:31,490 involved in thinking about other people's thoughts 63 00:02:31,490 --> 00:02:33,560 comes from the fact that that activation 64 00:02:33,560 --> 00:02:36,320 is a greater activation when you think about-- 65 00:02:36,320 --> 00:02:38,360 when you solve the false-belief type problems, 66 00:02:38,360 --> 00:02:40,700 when you think about another person's thoughts 67 00:02:40,700 --> 00:02:43,010 compared to when you think about a representation, 68 00:02:43,010 --> 00:02:46,730 a physical representation, like a photograph or a map, OK? 69 00:02:46,730 --> 00:02:50,420 So those are logically isomorphic problems. 70 00:02:50,420 --> 00:02:52,550 We have to always answer a question 71 00:02:52,550 --> 00:02:54,180 about a representation. 72 00:02:54,180 --> 00:02:56,420 It's just the representation is in somebody's head 73 00:02:56,420 --> 00:02:58,800 or it's a physical representation in the world. 74 00:02:58,800 --> 00:03:02,270 And in that difference, you get that region of the brain, OK? 75 00:03:02,270 --> 00:03:07,637 So that's cool because it's a very nicely designed-- 76 00:03:07,637 --> 00:03:09,470 it's not quite a-- nothing's a minimal pair, 77 00:03:09,470 --> 00:03:11,470 but it's a minimal pair in some respects, right? 78 00:03:11,470 --> 00:03:13,670 It carves out whether the representation 79 00:03:13,670 --> 00:03:17,840 is mental or physical, but it doesn't solve everything. 80 00:03:17,840 --> 00:03:19,730 And a suite of other tasks have shown 81 00:03:19,730 --> 00:03:21,980 that that region is actually specific in a whole bunch 82 00:03:21,980 --> 00:03:23,240 of other respects. 83 00:03:23,240 --> 00:03:26,150 It doesn't respond just whenever you think about a person. 84 00:03:26,150 --> 00:03:27,800 There are external properties. 85 00:03:27,800 --> 00:03:29,660 And most impressively, it does not even 86 00:03:29,660 --> 00:03:32,480 respond when you think about their visceral body 87 00:03:32,480 --> 00:03:35,550 sensations, like thirst, and hunger, and pain. 88 00:03:35,550 --> 00:03:38,480 So the TBJ doesn't respond to just thinking 89 00:03:38,480 --> 00:03:41,830 about any mental states of another person. 90 00:03:41,830 --> 00:03:44,330 It's specifically thinking about their thoughts and beliefs. 91 00:03:44,330 --> 00:03:46,520 And that's pretty damn remarkable, right? 92 00:03:46,520 --> 00:03:48,590 It's, how abstract can you get? 93 00:03:48,590 --> 00:03:50,390 And yet, here's a specific brain region 94 00:03:50,390 --> 00:03:54,057 for that very abstract, very specific thing. 95 00:03:54,057 --> 00:03:55,640 And the final little bit of evidence I 96 00:03:55,640 --> 00:03:58,143 showed you is that it also generalizes. 97 00:03:58,143 --> 00:03:59,810 It's not just about language because you 98 00:03:59,810 --> 00:04:03,860 can show people movies that have no words in them 99 00:04:03,860 --> 00:04:06,770 but that clearly show characters who must be thinking 100 00:04:06,770 --> 00:04:07,937 about each other's thoughts. 101 00:04:07,937 --> 00:04:10,145 And in those moments when the characters are thinking 102 00:04:10,145 --> 00:04:11,840 about each other's thoughts, that region 103 00:04:11,840 --> 00:04:15,920 turns on, more, for example, than when they're thinking 104 00:04:15,920 --> 00:04:18,980 about each other's pain, OK? 105 00:04:18,980 --> 00:04:19,850 Yeah. 106 00:04:19,850 --> 00:04:22,160 AUDIENCE: Did somebody look at what happens 107 00:04:22,160 --> 00:04:24,440 if other people think about me? 108 00:04:24,440 --> 00:04:26,240 Or if I think they thinking about me? 109 00:04:28,562 --> 00:04:30,020 NANCY KANWISHER: You mean if you're 110 00:04:30,020 --> 00:04:32,390 thinking about other people thinking about you? 111 00:04:32,390 --> 00:04:33,940 AUDIENCE: Yeah, exactly. 112 00:04:33,940 --> 00:04:36,440 NANCY KANWISHER: Yeah, I would assume that that region would 113 00:04:36,440 --> 00:04:37,412 be engaged. 114 00:04:37,412 --> 00:04:38,870 I'm sure there are studies on that. 115 00:04:38,870 --> 00:04:41,660 Because you're thinking about their thoughts, right? 116 00:04:41,660 --> 00:04:42,980 AUDIENCE: But more? 117 00:04:42,980 --> 00:04:44,130 Is it more interesting? 118 00:04:44,130 --> 00:04:45,922 NANCY KANWISHER: Probably just because it's 119 00:04:45,922 --> 00:04:46,950 more salient, right? 120 00:04:46,950 --> 00:04:48,920 I actually, oddly, in this class, 121 00:04:48,920 --> 00:04:50,610 I talk about attention at the end, 122 00:04:50,610 --> 00:04:54,180 which is weird because attention is an issue with every study. 123 00:04:54,180 --> 00:04:56,630 But most of the brain regions we've talked about, 124 00:04:56,630 --> 00:04:59,720 surely, including this one, are modulable by how strongly 125 00:04:59,720 --> 00:05:01,380 you're attending to something. 126 00:05:01,380 --> 00:05:03,440 So if something's really salient, or important, 127 00:05:03,440 --> 00:05:06,320 or you're really paying a lot of attention to it, 128 00:05:06,320 --> 00:05:07,820 you're going to get more activation. 129 00:05:07,820 --> 00:05:09,362 And if it's more interesting to think 130 00:05:09,362 --> 00:05:11,210 about what other people think about me 131 00:05:11,210 --> 00:05:14,150 than to think about what other people think about each other, 132 00:05:14,150 --> 00:05:18,510 you'll find some modulation in here, I'm sure. 133 00:05:18,510 --> 00:05:21,720 OK, and then finally, I talked about moral reasoning 134 00:05:21,720 --> 00:05:22,830 as a test case. 135 00:05:22,830 --> 00:05:25,230 It's not that the TPJ is selectively 136 00:05:25,230 --> 00:05:26,790 involved in moral reasoning. 137 00:05:26,790 --> 00:05:30,390 It's that many of the critical aspects of moral reasoning 138 00:05:30,390 --> 00:05:33,490 depend on what a person knew at the time. 139 00:05:33,490 --> 00:05:36,450 And so to use that information in moral reasoning, 140 00:05:36,450 --> 00:05:38,100 you need to pull that region in. 141 00:05:38,100 --> 00:05:40,860 So I realized I wrote that question ambiguously 142 00:05:40,860 --> 00:05:41,470 on the quiz. 143 00:05:41,470 --> 00:05:45,530 I meant to ask, is the TPJ engaged specifically or only 144 00:05:45,530 --> 00:05:47,280 in moral reasoning, to which they answer-- 145 00:05:47,280 --> 00:05:49,498 the correct answer would be no. 146 00:05:49,498 --> 00:05:51,040 But I didn't put the "only" in there. 147 00:05:51,040 --> 00:05:53,190 And I decided it was ambiguous, so if you said "yes," 148 00:05:53,190 --> 00:05:53,982 you got the points. 149 00:05:56,610 --> 00:06:00,630 Anyway, I gave several bits of evidence that, 150 00:06:00,630 --> 00:06:04,687 using the moral-reasoning case, that the TPJ is-- 151 00:06:08,400 --> 00:06:09,930 it's stronger evidence it's involved 152 00:06:09,930 --> 00:06:11,680 in thinking about other people's thoughts, 153 00:06:11,680 --> 00:06:14,658 first, that we showed that people with autism 154 00:06:14,658 --> 00:06:16,200 will have this difficulty in thinking 155 00:06:16,200 --> 00:06:17,520 about each other's thoughts. 156 00:06:17,520 --> 00:06:20,010 Even once they can pass the false-belief task, 157 00:06:20,010 --> 00:06:23,850 they put less emphasis, less weight on what the person knew 158 00:06:23,850 --> 00:06:27,900 at the time when evaluating the moral status of their actions, 159 00:06:27,900 --> 00:06:28,470 OK? 160 00:06:28,470 --> 00:06:30,122 It's not that they make a mistake 161 00:06:30,122 --> 00:06:31,830 or that they're unable to morally reason. 162 00:06:31,830 --> 00:06:33,120 It's a pretty subtle thing. 163 00:06:33,120 --> 00:06:35,430 It's just a small difference in how much 164 00:06:35,430 --> 00:06:40,710 they weigh what another person knew at the time, OK? 165 00:06:40,710 --> 00:06:44,040 That's also known as less forgiveness or less exoneration 166 00:06:44,040 --> 00:06:46,290 for accidental harm, right? 167 00:06:46,290 --> 00:06:49,260 You kick somebody by accident, and they go ouch, 168 00:06:49,260 --> 00:06:50,970 well, maybe you get a little bit of blame 169 00:06:50,970 --> 00:06:52,350 because you were a klutz and you should 170 00:06:52,350 --> 00:06:53,433 have thought or something. 171 00:06:53,433 --> 00:06:55,800 But it was an accident, so you should be exonerated. 172 00:06:55,800 --> 00:07:00,700 People with autism exonerate slightly less, right? 173 00:07:00,700 --> 00:07:01,200 OK. 174 00:07:01,200 --> 00:07:04,290 So we then talked about the fact that if you 175 00:07:04,290 --> 00:07:07,980 zap the TBJ with TMS, stick a coil there, 176 00:07:07,980 --> 00:07:09,120 you do the same thing. 177 00:07:09,120 --> 00:07:11,490 You slightly reduce the weight people 178 00:07:11,490 --> 00:07:14,370 put on what the person knew at the time 179 00:07:14,370 --> 00:07:19,830 in their moral evaluation of the person's actions. 180 00:07:19,830 --> 00:07:22,350 And then I showed this bizarre fact 181 00:07:22,350 --> 00:07:25,500 that, as I think Gisella asked, well, 182 00:07:25,500 --> 00:07:28,410 shouldn't the TBJ be different in people with autism? 183 00:07:28,410 --> 00:07:31,140 Yes, absolutely, according to all of this, it should be. 184 00:07:31,140 --> 00:07:34,180 But the basic univariate measures-- how big it is, 185 00:07:34,180 --> 00:07:36,210 how selective it is, where it is-- 186 00:07:36,210 --> 00:07:38,430 do not find a difference in that region 187 00:07:38,430 --> 00:07:40,590 with that contrast in high-functioning people 188 00:07:40,590 --> 00:07:41,400 with autism. 189 00:07:41,400 --> 00:07:43,350 That's surprising. 190 00:07:43,350 --> 00:07:46,410 But one possible answer to that is even though it's there 191 00:07:46,410 --> 00:07:51,097 and it's just as big and strong and all of that, it's-- 192 00:07:51,097 --> 00:07:52,680 that doesn't mean it doesn't represent 193 00:07:52,680 --> 00:07:53,790 different information. 194 00:07:53,790 --> 00:07:57,450 And I showed you an example that in typical people, 195 00:07:57,450 --> 00:08:01,440 you can decode from the TPJ whether the person is reading 196 00:08:01,440 --> 00:08:04,860 about another person's intentional harm or accidental 197 00:08:04,860 --> 00:08:05,790 harm. 198 00:08:05,790 --> 00:08:08,530 And you can't in people with ASD. 199 00:08:08,530 --> 00:08:09,030 OK? 200 00:08:09,030 --> 00:08:12,210 So that's my summary of last time. 201 00:08:12,210 --> 00:08:14,640 And then all of that was focusing very particularly 202 00:08:14,640 --> 00:08:18,120 on the most fancy, quintessentially human aspect 203 00:08:18,120 --> 00:08:20,730 of social cognition, which is this business of representing 204 00:08:20,730 --> 00:08:22,440 each other's thoughts and beliefs. 205 00:08:22,440 --> 00:08:26,100 But I pointed out at the end that there are also 206 00:08:26,100 --> 00:08:28,680 lots and lots of other facets of social perception 207 00:08:28,680 --> 00:08:30,660 and social cognition, many of which 208 00:08:30,660 --> 00:08:33,179 have somewhat selective brain regions, lots of them 209 00:08:33,179 --> 00:08:35,429 other parts of the brain, and we just didn't have time 210 00:08:35,429 --> 00:08:38,909 to filtrate in to that, OK? 211 00:08:38,909 --> 00:08:42,440 Hopefully that was a little bit clearer than I was last time. 212 00:08:42,440 --> 00:08:44,810 OK, so, so far, in this course, we've 213 00:08:44,810 --> 00:08:46,850 been focusing on all these bits of brain 214 00:08:46,850 --> 00:08:49,610 that seem to do very distinctive, often 215 00:08:49,610 --> 00:08:51,453 very selective things, OK? 216 00:08:51,453 --> 00:08:53,120 So the one we've just been talking about 217 00:08:53,120 --> 00:08:55,130 is that little guy right there. 218 00:08:55,130 --> 00:08:57,720 But we've talked about a lot of these things in here. 219 00:08:57,720 --> 00:09:00,650 And the field of human-cognitive neuroscience 220 00:09:00,650 --> 00:09:03,290 has invested lots of effort to find these things 221 00:09:03,290 --> 00:09:06,110 and try to characterize what each of them does. 222 00:09:06,110 --> 00:09:07,730 And that's pretty cool, right? 223 00:09:07,730 --> 00:09:10,370 This is all stuff we didn't know 20 years ago, and it's nice, 224 00:09:10,370 --> 00:09:13,940 and it counts as real progress, I think. 225 00:09:13,940 --> 00:09:19,640 But it leaves lots of things woefully unanswered. 226 00:09:19,640 --> 00:09:21,800 None of these regions can act alone, 227 00:09:21,800 --> 00:09:25,790 even though I've depicted them in a somewhat silly fashion, 228 00:09:25,790 --> 00:09:29,150 as nice little M&Ms on the brain. 229 00:09:29,150 --> 00:09:30,560 None of them act alone. 230 00:09:30,560 --> 00:09:32,190 None of them could act alone. 231 00:09:32,190 --> 00:09:33,650 They need information to process, 232 00:09:33,650 --> 00:09:35,710 so there has to be input to each region. 233 00:09:35,710 --> 00:09:38,210 They need to be able to tell other regions what they figured 234 00:09:38,210 --> 00:09:39,860 out or there's no point. 235 00:09:39,860 --> 00:09:42,230 And probably, as they solve a problem, 236 00:09:42,230 --> 00:09:44,060 as they conduct their computations, 237 00:09:44,060 --> 00:09:45,800 they're probably interacting all the time 238 00:09:45,800 --> 00:09:47,550 with lots of other regions. 239 00:09:47,550 --> 00:09:50,630 So we desperately need to understand not just 240 00:09:50,630 --> 00:09:52,903 that this patch does faces. 241 00:09:52,903 --> 00:09:54,320 There's lots more we need to know. 242 00:09:54,320 --> 00:09:56,153 And one of the things we really need to know 243 00:09:56,153 --> 00:09:59,840 is, what is it connected to, and who is it interacting with? 244 00:09:59,840 --> 00:10:01,130 OK? 245 00:10:01,130 --> 00:10:02,635 OK. 246 00:10:02,635 --> 00:10:03,760 So that's what I just said. 247 00:10:03,760 --> 00:10:05,780 And so that means looking at not just 248 00:10:05,780 --> 00:10:07,580 the cortex that we've been focusing 249 00:10:07,580 --> 00:10:11,450 on through this whole course, this dark matter that's 250 00:10:11,450 --> 00:10:14,030 on the surface of the brain up there. 251 00:10:14,030 --> 00:10:16,072 But today, we're going to do a figure-ground flip 252 00:10:16,072 --> 00:10:18,363 on the brain, and we're going to start paying attention 253 00:10:18,363 --> 00:10:20,990 to all that stuff that used to be background down there, 254 00:10:20,990 --> 00:10:23,180 all that white matter underneath, which 255 00:10:23,180 --> 00:10:26,300 is like a big heap of myelinated fibers 256 00:10:26,300 --> 00:10:30,770 that connect long-range regions of the brain to each other, OK? 257 00:10:30,770 --> 00:10:32,930 So you might say, OK, just wires-- 258 00:10:32,930 --> 00:10:34,430 who cares about the wires? 259 00:10:34,430 --> 00:10:36,290 That was my attitude for a long time. 260 00:10:36,290 --> 00:10:37,940 I've gotten over it. 261 00:10:37,940 --> 00:10:39,860 We desperately need to know about the wires 262 00:10:39,860 --> 00:10:41,053 for all kinds of reasons. 263 00:10:41,053 --> 00:10:43,220 So I'm going to go through a whole bunch of reasons. 264 00:10:43,220 --> 00:10:44,630 And there's a lot of little details, 265 00:10:44,630 --> 00:10:45,880 and I don't want you to panic. 266 00:10:45,880 --> 00:10:48,500 I just want to give you the gist of why this is worth 267 00:10:48,500 --> 00:10:50,000 paying attention to. 268 00:10:50,000 --> 00:10:53,330 OK, so first of all, white matter 269 00:10:53,330 --> 00:10:55,860 makes up 45% of the human brain. 270 00:10:55,860 --> 00:10:58,700 So that alone tells you it's not like some trivial thing. 271 00:10:58,700 --> 00:11:00,770 It's a big part of your brain. 272 00:11:00,770 --> 00:11:04,400 This is all the more interesting because that's not 273 00:11:04,400 --> 00:11:05,730 true in other animals. 274 00:11:05,730 --> 00:11:08,570 So I think white matter makes up a higher 275 00:11:08,570 --> 00:11:10,910 percent of the human brain than any other animal, 276 00:11:10,910 --> 00:11:12,470 or at least we're way up there. 277 00:11:12,470 --> 00:11:14,570 In mice, it's only 10%. 278 00:11:14,570 --> 00:11:17,090 And maybe that's a relatively uninteresting thing 279 00:11:17,090 --> 00:11:19,160 about scaling with brain size, but maybe it's 280 00:11:19,160 --> 00:11:22,130 something deeper about human brain-- what's 281 00:11:22,130 --> 00:11:24,770 special about the human brain and nobody knows. 282 00:11:24,770 --> 00:11:26,010 And here's a fun fact. 283 00:11:26,010 --> 00:11:28,520 If you took all the myelinated fibers in the human brain 284 00:11:28,520 --> 00:11:30,260 and you laid them out end to end, 285 00:11:30,260 --> 00:11:32,220 you could go around the world three times. 286 00:11:32,220 --> 00:11:36,820 So we've got lots of cableage sitting in here. 287 00:11:36,820 --> 00:11:37,900 OK. 288 00:11:37,900 --> 00:11:42,520 So I briefly argued before that you simply cannot understand 289 00:11:42,520 --> 00:11:46,660 the cortex without understanding its connections of one region 290 00:11:46,660 --> 00:11:47,360 to another. 291 00:11:47,360 --> 00:11:50,560 It's just crazy to study one little patch of the brain 292 00:11:50,560 --> 00:11:52,960 and not know who it's talking to and where 293 00:11:52,960 --> 00:11:54,700 it gets its inputs from. 294 00:11:54,700 --> 00:11:59,350 And as I just said, the pressing need for that knowledge 295 00:11:59,350 --> 00:12:01,660 is heightened by the presence of this map, which 296 00:12:01,660 --> 00:12:02,620 we didn't use to have. 297 00:12:02,620 --> 00:12:04,750 Now that we have this map, it's all the more 298 00:12:04,750 --> 00:12:07,090 important and pressing to know what the connections 299 00:12:07,090 --> 00:12:09,010 of those regions are. 300 00:12:09,010 --> 00:12:09,880 OK. 301 00:12:09,880 --> 00:12:12,130 So here's a nice quote making this point. 302 00:12:12,130 --> 00:12:14,890 This is Heidi Johansen-Berg and Matt Rushworth. 303 00:12:14,890 --> 00:12:19,840 They say, "Connectivity patterns define functional networks. 304 00:12:19,840 --> 00:12:22,480 The inputs to a brain region determine the information 305 00:12:22,480 --> 00:12:26,170 available to it, whereas its outputs dictate the influence 306 00:12:26,170 --> 00:12:28,810 that brain region can can have on other areas. 307 00:12:28,810 --> 00:12:31,720 Therefore, simply by knowing the pattern of inputs and outputs 308 00:12:31,720 --> 00:12:34,690 of a brain region, we can begin to make inferences 309 00:12:34,690 --> 00:12:37,088 about its likely functional specialization." 310 00:12:37,088 --> 00:12:38,380 So I think that's a nice quote. 311 00:12:38,380 --> 00:12:40,808 It makes the point that it's not just that we 312 00:12:40,808 --> 00:12:42,850 need to know the connections, but the connections 313 00:12:42,850 --> 00:12:45,640 and the function are bound to be deeply enmeshed. 314 00:12:45,640 --> 00:12:47,470 One constrains the other. 315 00:12:47,470 --> 00:12:48,760 Yeah? 316 00:12:48,760 --> 00:12:51,320 OK. 317 00:12:51,320 --> 00:12:55,040 Further, recall way back, which will quite 318 00:12:55,040 --> 00:12:57,590 possibly return on the final exam-- 319 00:12:57,590 --> 00:12:59,390 how do we define a cortical area? 320 00:12:59,390 --> 00:13:01,770 I gave you criteria for a cortical area. 321 00:13:01,770 --> 00:13:04,430 And one of the criteria was a distinctive pattern 322 00:13:04,430 --> 00:13:05,810 of connectivity, right? 323 00:13:05,810 --> 00:13:10,310 So it's part of the identifying properties of a cortical area 324 00:13:10,310 --> 00:13:13,640 is what it's connected to. 325 00:13:13,640 --> 00:13:16,295 And so that's another reason we should care. 326 00:13:19,290 --> 00:13:23,190 A third reason is if we knew of a given cortical area what 327 00:13:23,190 --> 00:13:26,280 its long-range connections were to lots of other regions, 328 00:13:26,280 --> 00:13:28,680 that connectivity fingerprint-- remember 329 00:13:28,680 --> 00:13:30,960 we talked briefly about connectivity fingerprints 330 00:13:30,960 --> 00:13:32,520 a month or so ago? 331 00:13:32,520 --> 00:13:34,890 That fingerprint, the distinctive set 332 00:13:34,890 --> 00:13:36,570 of connections of that region-- 333 00:13:36,570 --> 00:13:38,400 you can think of it as a signature 334 00:13:38,400 --> 00:13:40,320 of not just how that region differs 335 00:13:40,320 --> 00:13:43,260 from other regions in that same individual brain 336 00:13:43,260 --> 00:13:47,460 but how we might find a homolog of that region 337 00:13:47,460 --> 00:13:49,910 in another species. 338 00:13:49,910 --> 00:13:52,400 And that would be a very interesting thing to do. 339 00:13:52,400 --> 00:13:58,040 Wouldn't it be cool to know, is there a TPJ in macaques? 340 00:13:58,040 --> 00:14:00,140 Well, macaques can't solve an analog 341 00:14:00,140 --> 00:14:03,110 of the theory-of-mind task. 342 00:14:03,110 --> 00:14:04,730 Chimps-- we could debate a little bit. 343 00:14:04,730 --> 00:14:06,950 And narrow domains-- kind of sort of, a little bit, 344 00:14:06,950 --> 00:14:08,030 not really. 345 00:14:08,030 --> 00:14:11,210 Macaques-- no, OK? 346 00:14:11,210 --> 00:14:13,310 So is there a homolog? 347 00:14:13,310 --> 00:14:18,167 Is there a corresponding region that-- maybe we 348 00:14:18,167 --> 00:14:20,000 took that region, and we adapted it and made 349 00:14:20,000 --> 00:14:21,990 it work better so we could do better things with it? 350 00:14:21,990 --> 00:14:24,080 And if so, what is it doing in macaques, right? 351 00:14:24,080 --> 00:14:26,247 I mean, I think that's just a totally cool question. 352 00:14:26,247 --> 00:14:28,940 And in principle, one way to say what 353 00:14:28,940 --> 00:14:31,643 counts as "the same region" across species, which 354 00:14:31,643 --> 00:14:32,810 is kind of a weird question. 355 00:14:32,810 --> 00:14:33,920 They're different species, so what 356 00:14:33,920 --> 00:14:35,420 would "the same region" mean? 357 00:14:35,420 --> 00:14:37,460 One way to say what the same region means 358 00:14:37,460 --> 00:14:40,938 is to have a similar connectivity fingerprint, OK? 359 00:14:40,938 --> 00:14:42,980 So there are several studies that try to do that. 360 00:14:42,980 --> 00:14:44,900 I couldn't cram them into this lecture, 361 00:14:44,900 --> 00:14:47,210 but if you're interested in reading on it-- reading 362 00:14:47,210 --> 00:14:48,750 about it, shoot me an email. 363 00:14:48,750 --> 00:14:51,645 I'll send you some papers. 364 00:14:51,645 --> 00:14:52,145 OK. 365 00:14:54,680 --> 00:14:57,590 I also mentioned that the specific set of connections 366 00:14:57,590 --> 00:15:00,680 of a cortical region, particularly its inputs, 367 00:15:00,680 --> 00:15:03,050 play an important role in development. 368 00:15:03,050 --> 00:15:04,790 Remember the rewired ferrets? 369 00:15:04,790 --> 00:15:07,100 If you redirect the input to what 370 00:15:07,100 --> 00:15:10,190 would have been primary auditory cortex in a ferret 371 00:15:10,190 --> 00:15:13,610 and you have that input come in from the eyes, 372 00:15:13,610 --> 00:15:16,550 you can get what would have been primary auditory cortex 373 00:15:16,550 --> 00:15:19,980 to become a lot like primary visual cortex. 374 00:15:19,980 --> 00:15:22,100 So connectivity is important, not just 375 00:15:22,100 --> 00:15:24,620 in how a region functions and how 376 00:15:24,620 --> 00:15:26,720 we say what counts is the same across species 377 00:15:26,720 --> 00:15:28,760 but is probably also crucially important 378 00:15:28,760 --> 00:15:31,760 in the development of regions, OK? 379 00:15:31,760 --> 00:15:34,730 I also showed you evidence that the visual word-form area-- 380 00:15:34,730 --> 00:15:36,290 we can pick out exactly where it's 381 00:15:36,290 --> 00:15:38,180 going to land in an individual brain 382 00:15:38,180 --> 00:15:40,670 by the connectivity fingerprint of that region 383 00:15:40,670 --> 00:15:43,820 before kids learn to read, further evidence that 384 00:15:43,820 --> 00:15:47,260 connectivity determines later function. 385 00:15:47,260 --> 00:15:47,920 OK. 386 00:15:47,920 --> 00:15:49,870 As if this is not enough, other reasons 387 00:15:49,870 --> 00:15:52,660 to care about white matter is that disruptions 388 00:15:52,660 --> 00:15:56,950 of white matter are at the root of many clinical disorders-- 389 00:15:56,950 --> 00:16:01,330 dyslexia, autism, developmental prosopagnosia, amusia, 390 00:16:01,330 --> 00:16:05,830 all of these things and others, for all of them, 391 00:16:05,830 --> 00:16:09,250 disruptions in long-range white-matter connections 392 00:16:09,250 --> 00:16:13,240 have been implicated as possibly playing an important role 393 00:16:13,240 --> 00:16:18,820 in the etiology of that disease. 394 00:16:18,820 --> 00:16:22,360 Aging-- most definitely decline in white matter 395 00:16:22,360 --> 00:16:25,480 is prominent in aging. 396 00:16:25,480 --> 00:16:29,080 Sorry to say, there's a 10% decrease in white-matter fibers 397 00:16:29,080 --> 00:16:31,390 per decade starting at age 20. 398 00:16:31,390 --> 00:16:35,140 Use yours now while you have them. 399 00:16:35,140 --> 00:16:36,520 Let's change the topic. 400 00:16:36,520 --> 00:16:38,200 OK. 401 00:16:38,200 --> 00:16:41,050 There's a lot of talk about how white-matter connections may 402 00:16:41,050 --> 00:16:44,710 change with experience, and learning, and plasticity. 403 00:16:44,710 --> 00:16:48,010 And that's a pretty patchy literature. 404 00:16:48,010 --> 00:16:51,100 And it's not a very impressive literature. 405 00:16:51,100 --> 00:16:53,740 The classic thing you probably learned in 9.00, 406 00:16:53,740 --> 00:16:56,290 that when you juggle, you get changes in white-matter 407 00:16:56,290 --> 00:17:00,370 connections from juggling expertise. 408 00:17:00,370 --> 00:17:00,910 Maybe. 409 00:17:00,910 --> 00:17:01,420 Maybe not. 410 00:17:01,420 --> 00:17:03,440 There's some problems with a lot of that literature. 411 00:17:03,440 --> 00:17:05,619 So it's an interesting question, but it's not clear 412 00:17:05,619 --> 00:17:08,770 what the strong answers are. 413 00:17:08,770 --> 00:17:11,650 And finally, I don't know about circuit design. 414 00:17:11,650 --> 00:17:12,950 Probably some of you do. 415 00:17:12,950 --> 00:17:14,440 But I gather that people who think 416 00:17:14,440 --> 00:17:15,849 a lot about circuit design-- 417 00:17:15,849 --> 00:17:18,550 one of the key features you need to take into account 418 00:17:18,550 --> 00:17:19,900 is wiring length. 419 00:17:19,900 --> 00:17:22,810 You want to keep wiring length short, right? 420 00:17:22,810 --> 00:17:24,099 You have conduction delays. 421 00:17:24,099 --> 00:17:24,940 You have heating. 422 00:17:24,940 --> 00:17:26,950 You have space taken up in circuits. 423 00:17:26,950 --> 00:17:29,680 All of those things are bad in circuit design, 424 00:17:29,680 --> 00:17:31,930 and they're bad in brain design too. 425 00:17:31,930 --> 00:17:36,280 So a lot of reason to think that a major factor in the design 426 00:17:36,280 --> 00:17:39,190 of brains, especially human brains, 427 00:17:39,190 --> 00:17:42,640 is minimizing wiring length because wiring length 428 00:17:42,640 --> 00:17:44,800 is very expensive metabolically. 429 00:17:44,800 --> 00:17:47,530 You've got to maintain ion gradients 430 00:17:47,530 --> 00:17:51,170 across cell membranes. 431 00:17:51,170 --> 00:17:52,660 It's expensive developmentally. 432 00:17:52,660 --> 00:17:54,710 These damn things need to figure out where to go, 433 00:17:54,710 --> 00:17:56,377 and if they don't go to the right place, 434 00:17:56,377 --> 00:17:59,150 you have a developmental disorder. 435 00:17:59,150 --> 00:18:02,830 And so it's probably a real constraint 436 00:18:02,830 --> 00:18:04,480 on designs of brains. 437 00:18:04,480 --> 00:18:06,970 OK, so that was a whirlwind-- lots 438 00:18:06,970 --> 00:18:10,840 of reasons to care about white matter and connectivity. 439 00:18:10,840 --> 00:18:13,030 Oh, plus, at least in animal research, 440 00:18:13,030 --> 00:18:15,220 there's a whole suite of amazing new methods 441 00:18:15,220 --> 00:18:19,397 for looking at connectivity in animal brains. 442 00:18:19,397 --> 00:18:20,980 And [INAUDIBLE] can tell us more about 443 00:18:20,980 --> 00:18:23,403 that than I could because she's working 444 00:18:23,403 --> 00:18:25,570 in a lab that's right at the forefront of developing 445 00:18:25,570 --> 00:18:26,268 those methods. 446 00:18:26,268 --> 00:18:28,060 Someday, we're going to apply those methods 447 00:18:28,060 --> 00:18:28,870 to a human brain-- 448 00:18:28,870 --> 00:18:31,960 I can't wait-- and get the whole wiring diagram. 449 00:18:31,960 --> 00:18:32,590 OK. 450 00:18:32,590 --> 00:18:36,670 Anyway, so what do we know about the connectivity 451 00:18:36,670 --> 00:18:38,900 of these regions? 452 00:18:38,900 --> 00:18:42,560 Well, you may be thinking, don't we already know all this stuff? 453 00:18:42,560 --> 00:18:45,005 After all, I showed you this diagram way back. 454 00:18:45,005 --> 00:18:46,880 And you've probably seen it every damn course 455 00:18:46,880 --> 00:18:49,070 you take in this department. 456 00:18:49,070 --> 00:18:51,590 It's in most textbooks in the field-- 457 00:18:51,590 --> 00:18:54,033 the whole wiring diagram of the visual system. 458 00:18:54,033 --> 00:18:55,700 So don't we already know all this stuff? 459 00:18:55,700 --> 00:18:57,870 So what's the big deal? 460 00:18:57,870 --> 00:18:59,640 Well, here's the big deal. 461 00:18:59,640 --> 00:19:01,370 That's a macaque brain. 462 00:19:01,370 --> 00:19:04,880 And in macaque brain, you can get the actual answer 463 00:19:04,880 --> 00:19:07,310 to what is the actual structural connectivity 464 00:19:07,310 --> 00:19:11,095 of this patch of cortex to that patch of cortex. 465 00:19:11,095 --> 00:19:12,470 There's a whole bunch of methods, 466 00:19:12,470 --> 00:19:15,470 but traditionally, you inject some dye here 467 00:19:15,470 --> 00:19:17,510 that's uptaken by neurons that travels 468 00:19:17,510 --> 00:19:19,520 along axons that goes here. 469 00:19:19,520 --> 00:19:21,410 You kill the animal, slice up the brain, 470 00:19:21,410 --> 00:19:23,240 and find that tracer over here. 471 00:19:23,240 --> 00:19:26,910 And then those two things are absolutely connected. 472 00:19:26,910 --> 00:19:28,440 That's the gold standard. 473 00:19:28,440 --> 00:19:30,950 And that's the basis of most of those studies, that method 474 00:19:30,950 --> 00:19:33,340 and variations thereof. 475 00:19:33,340 --> 00:19:36,180 But we can't do that in human brains. 476 00:19:36,180 --> 00:19:38,760 And so we do not have anything like this information 477 00:19:38,760 --> 00:19:39,570 in human brains. 478 00:19:39,570 --> 00:19:40,200 Yes, David. 479 00:19:40,200 --> 00:19:41,910 AUDIENCE: When was this done? 480 00:19:41,910 --> 00:19:43,702 NANCY KANWISHER: Oh, that is a compilation. 481 00:19:43,702 --> 00:19:46,140 This was published, in, I think, 1991. 482 00:19:46,140 --> 00:19:48,765 But that was a compilation of heaps of studies that 483 00:19:48,765 --> 00:19:49,890 have been done before that. 484 00:19:49,890 --> 00:19:51,348 It was a big review article looking 485 00:19:51,348 --> 00:19:55,855 at all of this literature, where lots of classic neuroanatomy 486 00:19:55,855 --> 00:19:57,480 people would do these things where they 487 00:19:57,480 --> 00:19:59,760 would inject tracers, and slice up brains, 488 00:19:59,760 --> 00:20:00,840 and look in other places. 489 00:20:00,840 --> 00:20:03,480 And it was just a vast amount of literature 490 00:20:03,480 --> 00:20:05,315 that did that for many decades. 491 00:20:05,315 --> 00:20:06,690 It's sort of fallen out of favor, 492 00:20:06,690 --> 00:20:08,940 even though these things are, at least-- 493 00:20:08,940 --> 00:20:11,850 I don't know-- these are really crucial questions. 494 00:20:11,850 --> 00:20:13,830 Now people use other methods to do that. 495 00:20:13,830 --> 00:20:16,350 You can use all kinds of optogenetic and other methods 496 00:20:16,350 --> 00:20:18,360 to map connectivity in animals. 497 00:20:18,360 --> 00:20:19,020 Yeah. 498 00:20:19,020 --> 00:20:20,582 AUDIENCE: I have a question. 499 00:20:20,582 --> 00:20:22,680 [INAUDIBLE] 500 00:20:22,680 --> 00:20:24,660 NANCY KANWISHER: In here? 501 00:20:24,660 --> 00:20:26,760 Oh, that's a good question. 502 00:20:26,760 --> 00:20:29,220 Oh, it's probably dorsal and ventral pathways. 503 00:20:29,220 --> 00:20:30,390 Let me see here. 504 00:20:30,390 --> 00:20:31,650 Yeah. 505 00:20:31,650 --> 00:20:33,570 Yeah, the red ones-- 506 00:20:33,570 --> 00:20:36,930 this is another thing I didn't even really mention, probably 507 00:20:36,930 --> 00:20:38,430 let alone give a short [INAUDIBLE].. 508 00:20:38,430 --> 00:20:39,060 That was lame. 509 00:20:39,060 --> 00:20:42,060 But anyway, the visual-- high levels of the visual system-- 510 00:20:42,060 --> 00:20:43,800 we focused on the ventral visual pathway 511 00:20:43,800 --> 00:20:45,675 coming down the bottom of the temporal lobe. 512 00:20:45,675 --> 00:20:47,550 But there's a whole other visual pathway that 513 00:20:47,550 --> 00:20:48,870 goes up into the parietal lobe. 514 00:20:48,870 --> 00:20:51,360 Did I talk about that a little bit-- reaching and grasping? 515 00:20:51,360 --> 00:20:52,290 No, I didn't. 516 00:20:52,290 --> 00:20:54,690 Lame, lame, lame, lame. 517 00:20:54,690 --> 00:20:58,410 Anyway, a major part of the field I didn't get to. 518 00:20:58,410 --> 00:21:00,660 Anyway, it's a whole other part of the visual system 519 00:21:00,660 --> 00:21:04,770 that seems to be more involved in visually guided action. 520 00:21:04,770 --> 00:21:08,053 And they're actually very interconnected, 521 00:21:08,053 --> 00:21:10,470 but they're trying to emphasize that the dorsal pathway is 522 00:21:10,470 --> 00:21:14,230 at least somewhat separable in monkeys. 523 00:21:14,230 --> 00:21:16,240 But my point is this is monkeys where 524 00:21:16,240 --> 00:21:17,740 they have the gold-standard methods, 525 00:21:17,740 --> 00:21:21,850 and they can actually discover the real connectivity. 526 00:21:21,850 --> 00:21:25,420 Sadly, we can't do those things in humans. 527 00:21:25,420 --> 00:21:29,020 And in humans, we have only three methods, and none of them 528 00:21:29,020 --> 00:21:30,010 are very good. 529 00:21:30,010 --> 00:21:32,170 So we'll talk about them today anyway because this 530 00:21:32,170 --> 00:21:34,960 is such an important question, but the bottom line 531 00:21:34,960 --> 00:21:37,000 is-- this drives me out of my mind-- 532 00:21:37,000 --> 00:21:39,730 we basically don't know the connectivity of any 533 00:21:39,730 --> 00:21:43,450 of those regions for sure in human brains. 534 00:21:43,450 --> 00:21:45,940 And somebody's got to solve that. 535 00:21:45,940 --> 00:21:48,400 Maybe one of you will go invent a method that 536 00:21:48,400 --> 00:21:51,640 works in humans that helps us solve that problem because it's 537 00:21:51,640 --> 00:21:55,305 actually, I think, really paralyzing to our field. 538 00:21:55,305 --> 00:21:57,430 So I'll tell you what we do know, which isn't much. 539 00:21:57,430 --> 00:22:00,160 But, you know, beggars can't be choosers. 540 00:22:00,160 --> 00:22:04,480 OK, the first method has been around for a few years, 541 00:22:04,480 --> 00:22:05,830 and that's gross dissection. 542 00:22:05,830 --> 00:22:10,460 And I mean gross, like that kind of gross-- 543 00:22:10,460 --> 00:22:12,820 so only good for post-mortem brains, 544 00:22:12,820 --> 00:22:14,390 but it's really quite amazing. 545 00:22:14,390 --> 00:22:19,030 This is a bottom view of the brain back and front. 546 00:22:19,030 --> 00:22:21,500 This is actually a physically dissected brain. 547 00:22:21,500 --> 00:22:25,090 Like, it takes a real serious neuroanatomist 548 00:22:25,090 --> 00:22:26,620 and lots of fancy methods-- 549 00:22:26,620 --> 00:22:28,120 I mean, not fancy methods but lots 550 00:22:28,120 --> 00:22:32,170 of careful, precise teasing apart of bits of brain. 551 00:22:32,170 --> 00:22:36,530 And you can actually see these big fibers coming up here. 552 00:22:36,530 --> 00:22:38,200 So if this is the back of the brain 553 00:22:38,200 --> 00:22:39,790 and we're looking up like this, what 554 00:22:39,790 --> 00:22:43,510 do you think those fibers are connecting right there? 555 00:22:43,510 --> 00:22:48,280 Big fiber bundle coming from deep down in the brain 556 00:22:48,280 --> 00:22:50,980 up to right in there. 557 00:22:50,980 --> 00:22:53,440 AUDIENCE: Is that thalamus to [INAUDIBLE]?? 558 00:22:53,440 --> 00:22:54,400 NANCY KANWISHER: Bingo. 559 00:22:54,400 --> 00:22:55,600 Exactly. 560 00:22:55,600 --> 00:22:57,820 OK, so that's the LGN right there. 561 00:22:57,820 --> 00:22:59,620 And this is called the optic radiation. 562 00:22:59,620 --> 00:23:02,550 It's this huge cable of fibers that come up. 563 00:23:02,550 --> 00:23:05,290 OK, first, here's the optic tract. 564 00:23:05,290 --> 00:23:06,220 Actually, I forget. 565 00:23:06,220 --> 00:23:07,720 That's not-- I think this is the optic tract that's 566 00:23:07,720 --> 00:23:08,710 been snipped there. 567 00:23:08,710 --> 00:23:11,890 Then it comes up in here, makes a stop in the LGN, 568 00:23:11,890 --> 00:23:13,990 and then this big batch of fibers 569 00:23:13,990 --> 00:23:18,100 comes up right there to primary visual cortex, OK? 570 00:23:18,100 --> 00:23:20,150 Everybody got that? 571 00:23:20,150 --> 00:23:24,130 OK, so you can actually see it in dissecting a dead brain. 572 00:23:24,130 --> 00:23:27,380 OK, that's pretty cool. 573 00:23:27,380 --> 00:23:30,820 But what if we don't want to wait for people to die? 574 00:23:30,820 --> 00:23:33,160 Often, we want to ask questions about a person right 575 00:23:33,160 --> 00:23:34,510 now in their brain. 576 00:23:34,510 --> 00:23:35,860 Do they have this disorder? 577 00:23:35,860 --> 00:23:38,500 Are they at risk of that disorder? 578 00:23:38,500 --> 00:23:40,450 What is their connectivity? 579 00:23:40,450 --> 00:23:42,490 So for that, we have two methods, 580 00:23:42,490 --> 00:23:44,500 and I'll talk about these two methods 581 00:23:44,500 --> 00:23:45,740 in the rest of the lecture. 582 00:23:45,740 --> 00:23:48,170 The first one is diffusion imaging. 583 00:23:48,170 --> 00:23:50,440 OK, so I talked about this briefly before. 584 00:23:50,440 --> 00:23:54,260 But let me remind you of what the basic principles are. 585 00:23:54,260 --> 00:23:59,140 So here is a picture of the optic nerve with a bunch 586 00:23:59,140 --> 00:24:00,850 of axons oriented like that. 587 00:24:00,850 --> 00:24:02,920 It's a big cable with a whole bunch 588 00:24:02,920 --> 00:24:04,550 of little fibers in there. 589 00:24:04,550 --> 00:24:08,020 And the basic kind of biophysics is 590 00:24:08,020 --> 00:24:13,180 that water wants to diffuse more along this length, following 591 00:24:13,180 --> 00:24:15,130 the orientation of the fibers, than it 592 00:24:15,130 --> 00:24:18,100 wants to diffuse this way, OK? 593 00:24:18,100 --> 00:24:20,980 And diffusion imaging-- I'm not explaining any of the physics, 594 00:24:20,980 --> 00:24:24,070 but just take it for my word that what diffusion imaging 595 00:24:24,070 --> 00:24:27,760 does is give you a picture of the direction of water 596 00:24:27,760 --> 00:24:30,100 diffusion, OK? 597 00:24:30,100 --> 00:24:33,100 So you get a picture of a piece of brain, 598 00:24:33,100 --> 00:24:36,670 and it'll show you, for example, that right in there, 599 00:24:36,670 --> 00:24:37,660 all the fibers-- 600 00:24:37,660 --> 00:24:39,970 well, the water is diffusing this way. 601 00:24:39,970 --> 00:24:43,960 And over here, the water's diffusing that way, OK? 602 00:24:43,960 --> 00:24:47,560 And that's just what you see in a diffusion image, OK? 603 00:24:47,560 --> 00:24:50,230 And so the inference people make is 604 00:24:50,230 --> 00:24:52,780 if you have all those parallel lines telling you 605 00:24:52,780 --> 00:24:54,280 there's lots of diffusion like this, 606 00:24:54,280 --> 00:24:57,670 there's probably a big fiber bundle going like that-- 607 00:24:57,670 --> 00:24:58,600 and there is. 608 00:24:58,600 --> 00:25:00,710 That would be the corpus callosum. 609 00:25:00,710 --> 00:25:01,210 OK? 610 00:25:03,900 --> 00:25:04,470 All right. 611 00:25:04,470 --> 00:25:09,760 So this method works great for finding the big fiber bundles. 612 00:25:09,760 --> 00:25:10,260 OK? 613 00:25:10,260 --> 00:25:12,810 I'm going to dis diffusion imaging in a bunch of ways, 614 00:25:12,810 --> 00:25:15,420 but it is great for finding the big fiber 615 00:25:15,420 --> 00:25:17,700 bundles because in those big fiber bundles, 616 00:25:17,700 --> 00:25:19,440 axons are very parallel. 617 00:25:19,440 --> 00:25:22,230 There's a whole bunch of them, and you can really see it. 618 00:25:22,230 --> 00:25:23,010 OK. 619 00:25:23,010 --> 00:25:24,540 And so people have been using this 620 00:25:24,540 --> 00:25:28,380 for over a decade to find some of the major fiber 621 00:25:28,380 --> 00:25:29,830 bundles in the brain. 622 00:25:29,830 --> 00:25:33,720 So you may have heard of the arcuate fasciculus that 623 00:25:33,720 --> 00:25:36,960 basically connects language regions in the temporal lobe 624 00:25:36,960 --> 00:25:41,040 up to Broca's area in the frontal lobe, OK? 625 00:25:41,040 --> 00:25:42,750 It's a big bunch of fibers that go-- 626 00:25:42,750 --> 00:25:45,690 I guess in me, they go like this, boom, right? 627 00:25:45,690 --> 00:25:48,840 [INAUDIBLE] And you can see those guys with-- this 628 00:25:48,840 --> 00:25:51,330 is a distant reconstruction, but you can see 629 00:25:51,330 --> 00:25:54,400 those with diffusion imaging. 630 00:25:54,400 --> 00:25:57,637 Another one, the goes from the front of the temporal lobe up 631 00:25:57,637 --> 00:25:58,470 to the frontal lobe. 632 00:25:58,470 --> 00:25:59,680 You don't need to memorize these. 633 00:25:59,680 --> 00:26:00,400 I don't care about that. 634 00:26:00,400 --> 00:26:02,580 I just want you to get the idea of what you can see. 635 00:26:02,580 --> 00:26:03,270 Yeah, question? 636 00:26:03,270 --> 00:26:05,250 AUDIENCE: So these are discoverable without the person 637 00:26:05,250 --> 00:26:06,730 having to do anything [INAUDIBLE]?? 638 00:26:06,730 --> 00:26:07,260 NANCY KANWISHER: Yes. 639 00:26:07,260 --> 00:26:07,860 Yes. 640 00:26:07,860 --> 00:26:09,730 These are anatomical images. 641 00:26:09,730 --> 00:26:11,820 So in diffusion imaging, you don't do anything. 642 00:26:11,820 --> 00:26:13,740 You can sleep, actually. 643 00:26:13,740 --> 00:26:16,740 That's ideal because it's long and boring. 644 00:26:16,740 --> 00:26:18,540 Actually, it's not boring, but the scanner 645 00:26:18,540 --> 00:26:21,570 shakes like hell in a diffusion-imaging scan. 646 00:26:21,570 --> 00:26:23,730 It's pretty wild. 647 00:26:23,730 --> 00:26:25,380 We could charge admission for it. 648 00:26:30,078 --> 00:26:30,620 I don't know. 649 00:26:30,620 --> 00:26:33,910 I find it quite wild. 650 00:26:33,910 --> 00:26:38,040 Anyway, this is the inferior longitudinal fasciculus. 651 00:26:38,040 --> 00:26:40,240 As it goes down the temporal lobe. 652 00:26:40,240 --> 00:26:42,390 So when we talk about the ventral-visual pathway-- 653 00:26:42,390 --> 00:26:44,710 face areas, place areas, all that stuff-- 654 00:26:44,710 --> 00:26:47,550 this is the big fiber highway that sits right 655 00:26:47,550 --> 00:26:50,220 on top of that whole chunk of gray matter 656 00:26:50,220 --> 00:26:51,630 that does all the processing. 657 00:26:51,630 --> 00:26:54,030 And it's a big pile of fibers that go straight 658 00:26:54,030 --> 00:26:56,950 down the temporal lobe, OK? 659 00:26:56,950 --> 00:26:58,810 OK. 660 00:26:58,810 --> 00:27:02,690 And so here's more recent data. 661 00:27:02,690 --> 00:27:07,340 This is from Anastasia Yendiki over at MHG Charlestown 662 00:27:07,340 --> 00:27:07,840 over there. 663 00:27:07,840 --> 00:27:10,450 And she's developed this lovely piece of software 664 00:27:10,450 --> 00:27:13,660 that enables you to take diffusion images and identify, 665 00:27:13,660 --> 00:27:15,640 based on an atlas she's put together, 666 00:27:15,640 --> 00:27:19,780 18 of the major fiber tracts in the human brain-- 667 00:27:19,780 --> 00:27:21,297 so nine per hemisphere. 668 00:27:21,297 --> 00:27:23,380 And this is just showing you some of the big ones. 669 00:27:23,380 --> 00:27:25,600 This is the inferior longitudinal fasciculus 670 00:27:25,600 --> 00:27:27,250 I just showed you and so forth. 671 00:27:27,250 --> 00:27:28,540 Yeah. 672 00:27:28,540 --> 00:27:30,490 AUDIENCE: So how does this compare to just 673 00:27:30,490 --> 00:27:33,160 a postmortem dissection? 674 00:27:33,160 --> 00:27:34,300 Could you see the-- 675 00:27:34,300 --> 00:27:35,450 NANCY KANWISHER: It's a good question. 676 00:27:35,450 --> 00:27:36,110 It's a good question. 677 00:27:36,110 --> 00:27:37,090 I don't exactly know. 678 00:27:37,090 --> 00:27:38,065 It wouldn't be easy. 679 00:27:40,542 --> 00:27:42,250 That thing I showed you works because you 680 00:27:42,250 --> 00:27:43,540 take the stuff off the top, and it's just 681 00:27:43,540 --> 00:27:44,710 kind of sitting there. 682 00:27:44,710 --> 00:27:47,380 But then you would have taken a lot of other stuff out, 683 00:27:47,380 --> 00:27:49,547 and you wouldn't be able to see that other stuff you 684 00:27:49,547 --> 00:27:50,410 had to take out. 685 00:27:50,410 --> 00:27:51,285 You know what I mean? 686 00:27:51,285 --> 00:27:54,827 So here, you can surf through and pick out any of these. 687 00:27:54,827 --> 00:27:56,410 So it's definitely going to be better. 688 00:27:56,410 --> 00:28:00,760 But which of these you can see with postmortem dissection 689 00:28:00,760 --> 00:28:03,160 I'm not sure-- some of them, some of the bigger ones 690 00:28:03,160 --> 00:28:06,805 but probably not all of them. 691 00:28:06,805 --> 00:28:09,820 OK, so here are some of the major tracts. 692 00:28:09,820 --> 00:28:11,500 OK. 693 00:28:11,500 --> 00:28:13,720 But you can do a little bit more than just find them 694 00:28:13,720 --> 00:28:14,890 with diffusion imaging. 695 00:28:14,890 --> 00:28:17,050 You can also characterize them a little bit. 696 00:28:17,050 --> 00:28:18,640 And this is a whole universe. 697 00:28:18,640 --> 00:28:20,140 There's people who spend their lives 698 00:28:20,140 --> 00:28:21,610 with all kinds of fancy measures, 699 00:28:21,610 --> 00:28:24,730 and I'm just going to tell you about the most common one. 700 00:28:24,730 --> 00:28:28,180 So recall that the whole deal with diffusion imaging 701 00:28:28,180 --> 00:28:32,750 is it's looking for orientations of maximum water diffusion. 702 00:28:32,750 --> 00:28:35,830 So some parts of the brain have a systematic set 703 00:28:35,830 --> 00:28:37,570 of directions of water diffusion, 704 00:28:37,570 --> 00:28:39,760 and other ones don't, right? 705 00:28:39,760 --> 00:28:42,130 Inside a ventricle, the water can go any which way. 706 00:28:42,130 --> 00:28:44,690 There's nothing determining which way it goes. 707 00:28:44,690 --> 00:28:47,950 So this is called isotropic because you 708 00:28:47,950 --> 00:28:50,350 have diffusion going equally-- 709 00:28:50,350 --> 00:28:53,120 in equal amounts in all directions and anisotropic, 710 00:28:53,120 --> 00:28:53,620 right? 711 00:28:53,620 --> 00:28:56,740 So it goes systematically more in one direction, 712 00:28:56,740 --> 00:28:58,750 one axis than others, OK? 713 00:28:58,750 --> 00:29:01,810 Diffusion can't see the axial direction, like left or right 714 00:29:01,810 --> 00:29:02,950 versus to left. 715 00:29:02,950 --> 00:29:04,750 It can just see that this axis is 716 00:29:04,750 --> 00:29:08,640 more prominent than this one, or this one, or this one, OK? 717 00:29:08,640 --> 00:29:13,530 So you're not actually seeing it move in a systematic direction. 718 00:29:13,530 --> 00:29:16,930 OK, so that's the basic signal. 719 00:29:16,930 --> 00:29:19,800 So what you can do is in a little patch of brain, 720 00:29:19,800 --> 00:29:23,130 you can ask not just, what direction 721 00:29:23,130 --> 00:29:25,380 has maximum diffusion, which is what 722 00:29:25,380 --> 00:29:27,120 I've been talking about so far. 723 00:29:27,120 --> 00:29:31,140 You can say, how much more does it go in that maximum direction 724 00:29:31,140 --> 00:29:32,430 than any of the others? 725 00:29:32,430 --> 00:29:34,530 Is it more like this or more like that? 726 00:29:34,530 --> 00:29:37,410 And you can imagine a whole spectrum in-between. 727 00:29:37,410 --> 00:29:39,660 OK, so you're just asking, how oriented is it? 728 00:29:39,660 --> 00:29:43,560 Is it totally oriented or just partially? 729 00:29:43,560 --> 00:29:49,097 And that measure is called Fractional Anisotropy, or FA. 730 00:29:49,097 --> 00:29:51,180 It's prominent enough in the field you should just 731 00:29:51,180 --> 00:29:52,110 learn this phrase. 732 00:29:52,110 --> 00:29:53,913 And you read any articles, especially 733 00:29:53,913 --> 00:29:55,830 any clinical articles, this is the first thing 734 00:29:55,830 --> 00:29:59,730 you'll see in any clinical papers that 735 00:29:59,730 --> 00:30:01,740 use diffusion imaging. 736 00:30:01,740 --> 00:30:04,320 OK, and so fractional anisotropy-- there's 737 00:30:04,320 --> 00:30:05,730 a bunch of fancy definitions. 738 00:30:05,730 --> 00:30:06,520 And we don't care. 739 00:30:06,520 --> 00:30:08,250 We just want the idea. 740 00:30:08,250 --> 00:30:11,840 It's just, to what degree is that little patch of brain 741 00:30:11,840 --> 00:30:13,590 in this little part of a tract that you've 742 00:30:13,590 --> 00:30:18,570 identified more like this or more like that, OK? 743 00:30:18,570 --> 00:30:21,400 Is it anisotropic or isotropic? 744 00:30:21,400 --> 00:30:23,080 OK? 745 00:30:23,080 --> 00:30:25,600 And so this has been used a lot to try 746 00:30:25,600 --> 00:30:30,220 to ask about the nature of fiber tracts in different groups-- 747 00:30:30,220 --> 00:30:34,480 young versus old, different clinical groups, 748 00:30:34,480 --> 00:30:37,900 autism versus typical, schizophrenia, you name it. 749 00:30:37,900 --> 00:30:40,270 Experience-- you train people up on a task-- 750 00:30:40,270 --> 00:30:43,540 do you change the FA, the Fractional Anisotropy, 751 00:30:43,540 --> 00:30:45,710 of some particular tract. 752 00:30:45,710 --> 00:30:46,210 OK? 753 00:30:46,210 --> 00:30:48,550 So it's just a characteristic of tract 754 00:30:48,550 --> 00:30:51,580 is, how oriented is it along the way? 755 00:30:51,580 --> 00:30:54,550 Is it super clean, totally oriented? 756 00:30:54,550 --> 00:31:00,730 Or does it have some isotropy mixed in? 757 00:31:00,730 --> 00:31:01,243 OK? 758 00:31:01,243 --> 00:31:03,160 All right, so this is all over the literature. 759 00:31:03,160 --> 00:31:06,090 Let me give you one cool example from Gabrielli lab 760 00:31:06,090 --> 00:31:07,840 that came out recently. 761 00:31:07,840 --> 00:31:12,360 So they identified the arcuate fasciculus that I showed you 762 00:31:12,360 --> 00:31:14,880 before, going from the-- basically, Wernicke's 763 00:31:14,880 --> 00:31:16,890 area curving around up to Broca's area 764 00:31:16,890 --> 00:31:20,580 up there so that you can identify it anatomically 765 00:31:20,580 --> 00:31:22,770 in each subject individually, OK? 766 00:31:22,770 --> 00:31:23,610 Now you've got it. 767 00:31:23,610 --> 00:31:25,650 You've identified which voxels are 768 00:31:25,650 --> 00:31:27,930 part of the arcuate fasciculus. 769 00:31:27,930 --> 00:31:29,490 And then what they want to ask-- 770 00:31:29,490 --> 00:31:32,880 what they wanted to ask is, is the integrity, 771 00:31:32,880 --> 00:31:35,340 or the characteristics of the arcuate fasciculus-- 772 00:31:35,340 --> 00:31:40,560 is that important for language-- for dyslexia? 773 00:31:40,560 --> 00:31:43,920 So they measured the FA along this tract. 774 00:31:43,920 --> 00:31:46,170 How oriented is it all the way along here? 775 00:31:46,170 --> 00:31:47,940 And then you get some kind of average. 776 00:31:47,940 --> 00:31:49,770 And they measured that in a bunch of kids 777 00:31:49,770 --> 00:31:53,760 with dyslexia and in a bunch of kids with no reading disability 778 00:31:53,760 --> 00:31:55,470 who are matched in other dimensions 779 00:31:55,470 --> 00:31:59,420 of non-verbal cognitive ability, OK? 780 00:31:59,420 --> 00:32:02,000 And what they found is the fractional anisotropy 781 00:32:02,000 --> 00:32:04,940 was higher in the typical kids than the kids 782 00:32:04,940 --> 00:32:08,090 with reading disability, with dyslexia, OK? 783 00:32:08,090 --> 00:32:11,990 And from that, they implicated that this connection 784 00:32:11,990 --> 00:32:15,690 may play some role in dyslexia. 785 00:32:15,690 --> 00:32:16,190 OK? 786 00:32:16,190 --> 00:32:18,602 It's not totally obvious because one 787 00:32:18,602 --> 00:32:20,810 would think this region is connected to that region-- 788 00:32:20,810 --> 00:32:23,030 those are languagey regions, right? 789 00:32:23,030 --> 00:32:24,830 It's not visual regions. 790 00:32:24,830 --> 00:32:27,530 You think of dyslexia as a problem seeing the letters 791 00:32:27,530 --> 00:32:29,210 and which ones are oriented which way. 792 00:32:29,210 --> 00:32:30,680 And this suggests that, at least, 793 00:32:30,680 --> 00:32:33,710 higher-level connectivity between language regions 794 00:32:33,710 --> 00:32:35,570 may be implicated, OK? 795 00:32:35,570 --> 00:32:36,685 Yeah. 796 00:32:36,685 --> 00:32:38,810 AUDIENCE: So this just talks about the architecture 797 00:32:38,810 --> 00:32:41,760 and not the per unit information being [INAUDIBLE]?? 798 00:32:41,760 --> 00:32:43,010 NANCY KANWISHER: That's right. 799 00:32:43,010 --> 00:32:44,300 AUDIENCE: There's no information [INAUDIBLE].. 800 00:32:44,300 --> 00:32:45,550 NANCY KANWISHER: That's right. 801 00:32:45,550 --> 00:32:47,810 It's just saying, where are the wires, 802 00:32:47,810 --> 00:32:49,820 and how organized are the wires? 803 00:32:49,820 --> 00:32:50,930 Period. 804 00:32:50,930 --> 00:32:51,750 Yeah. 805 00:32:51,750 --> 00:32:54,595 AUDIENCE: So they won't flow through some technique 806 00:32:54,595 --> 00:32:57,500 as-- if I can mess with it, like I'm 807 00:32:57,500 --> 00:32:59,330 measuring it, if there's some means 808 00:32:59,330 --> 00:33:01,310 to miss the [INAUDIBLE] diffusion, 809 00:33:01,310 --> 00:33:03,002 will that have any effect? 810 00:33:03,002 --> 00:33:04,710 NANCY KANWISHER: I didn't quite get that. 811 00:33:04,710 --> 00:33:05,600 Say it again. 812 00:33:05,600 --> 00:33:07,318 AUDIENCE: So there's this water diffusion 813 00:33:07,318 --> 00:33:09,110 that's happening that I'm going to measure. 814 00:33:09,110 --> 00:33:13,370 But what if I have some way to intervene and change 815 00:33:13,370 --> 00:33:14,703 the rate of diffusion some such? 816 00:33:14,703 --> 00:33:16,703 NANCY KANWISHER: I don't know how you'd do that. 817 00:33:16,703 --> 00:33:18,740 I mean, it's a pretty basic physical property-- 818 00:33:18,740 --> 00:33:22,130 diffusion of water and how it's constrained by lipids, right? 819 00:33:22,130 --> 00:33:25,180 AUDIENCE: But that shouldn't affect, 820 00:33:25,180 --> 00:33:28,220 like the function of information flow [INAUDIBLE]?? 821 00:33:28,220 --> 00:33:30,070 NANCY KANWISHER: Ooh. 822 00:33:30,070 --> 00:33:31,720 I have no idea. 823 00:33:31,720 --> 00:33:34,480 I mean, that's a biophysical question I don't know about. 824 00:33:34,480 --> 00:33:37,690 But notice, this is a pretty distant proxy. 825 00:33:37,690 --> 00:33:40,660 You're mostly looking at water between axons, not even 826 00:33:40,660 --> 00:33:41,900 within them. 827 00:33:41,900 --> 00:33:45,010 And so it's just a proxy for, how well can we 828 00:33:45,010 --> 00:33:48,670 see those fibers and how they're oriented, OK? 829 00:33:48,670 --> 00:33:51,850 Yeah, it's pretty removed from actual signals 830 00:33:51,850 --> 00:33:54,460 going along the wires. 831 00:33:54,460 --> 00:33:56,585 Anyway, so everybody get the sense that-- you know, 832 00:33:56,585 --> 00:33:57,740 it's one little finding. 833 00:33:57,740 --> 00:34:03,680 But it implicates something about that tract in dyslexia. 834 00:34:03,680 --> 00:34:06,440 OK, so that's interesting. 835 00:34:06,440 --> 00:34:08,457 But first, it's just correlational. 836 00:34:08,457 --> 00:34:10,040 Lots of things are just correlational. 837 00:34:10,040 --> 00:34:12,290 Most of the stuff in this class is just correlational. 838 00:34:12,290 --> 00:34:14,480 Same is true here. 839 00:34:14,480 --> 00:34:16,010 But a little more seriously, it's 840 00:34:16,010 --> 00:34:19,350 not totally clear what fractional anisotropy means, 841 00:34:19,350 --> 00:34:19,850 OK? 842 00:34:19,850 --> 00:34:23,120 So there's a real tradition of treating 843 00:34:23,120 --> 00:34:26,540 high-fractional anisotropy as if that's good. 844 00:34:26,540 --> 00:34:28,429 After all, we're in a fiber bundle. 845 00:34:28,429 --> 00:34:30,949 Shouldn't all the axons be oriented nicely 846 00:34:30,949 --> 00:34:32,389 in there and not all scrambled? 847 00:34:32,389 --> 00:34:36,260 Surely, oriented is good, and scrambled is bad. 848 00:34:36,260 --> 00:34:40,443 Well, maybe, but sometimes, fibers cross a fiber bundle. 849 00:34:40,443 --> 00:34:42,860 So you can have a fiber bundle like this with other fibers 850 00:34:42,860 --> 00:34:43,920 crossing it. 851 00:34:43,920 --> 00:34:46,250 And so when that happens, maybe that's good. 852 00:34:46,250 --> 00:34:50,420 And so people use FA as a proxy for good fiber. 853 00:34:50,420 --> 00:34:53,000 People say "fiber integrity" even. 854 00:34:53,000 --> 00:34:56,048 But there's a whole question about what exactly it means. 855 00:34:56,048 --> 00:34:57,590 People will spend their lives looking 856 00:34:57,590 --> 00:35:00,290 at the biophysics of fibers and all the different things 857 00:35:00,290 --> 00:35:03,770 that fractional anisotropy and the other measures might mean. 858 00:35:03,770 --> 00:35:08,030 And it's actually pretty complicated and unresolved. 859 00:35:08,030 --> 00:35:10,850 Another challenge with fractional anisotropy 860 00:35:10,850 --> 00:35:14,540 is it's extremely vulnerable to artifacts. 861 00:35:14,540 --> 00:35:17,150 All of diffusion imaging is extremely 862 00:35:17,150 --> 00:35:19,370 vulnerable to artifacts, OK? 863 00:35:19,370 --> 00:35:21,860 And I'm going to give you an example of a study we 864 00:35:21,860 --> 00:35:23,240 did a few years ago. 865 00:35:23,240 --> 00:35:25,760 So I was, for a while, trying to work on autism. 866 00:35:25,760 --> 00:35:28,010 I've, more or less, given up because it's, as 867 00:35:28,010 --> 00:35:29,810 far as I can tell, impossible. 868 00:35:29,810 --> 00:35:32,480 But back while I was still trying, 869 00:35:32,480 --> 00:35:35,660 we scanned a whole bunch of kids with and without autism 870 00:35:35,660 --> 00:35:38,150 with diffusion imaging, OK? 871 00:35:38,150 --> 00:35:41,480 And at the time, there were about 50 published papers, 872 00:35:41,480 --> 00:35:45,650 almost all of which said one of the things you find with autism 873 00:35:45,650 --> 00:35:47,570 is that there's an underdevelopment 874 00:35:47,570 --> 00:35:50,360 of long-range connectivity and an overdevelopment 875 00:35:50,360 --> 00:35:52,040 of short-range connectivity. 876 00:35:52,040 --> 00:35:54,980 And so then people would free associate 877 00:35:54,980 --> 00:35:56,630 with all kinds of speculations about, 878 00:35:56,630 --> 00:35:59,750 OK, this explains aspects of the autism phenotype. 879 00:35:59,750 --> 00:36:01,533 They can't put different ideas together 880 00:36:01,533 --> 00:36:03,950 because their connections across the brain aren't as good. 881 00:36:03,950 --> 00:36:06,020 And they're obsessed with little details 882 00:36:06,020 --> 00:36:09,680 because they have too many local connections and all kinds of-- 883 00:36:09,680 --> 00:36:12,870 suggestive, but very, very fuzzy ideas like that. 884 00:36:12,870 --> 00:36:17,720 So 50 papers pretty much all found underdeveloped lower 885 00:36:17,720 --> 00:36:20,810 fractional anisotropy and long-range connections, 886 00:36:20,810 --> 00:36:25,480 long-range tracts in autism, a very established finding 887 00:36:25,480 --> 00:36:28,150 So we went in not to raise hell but just 888 00:36:28,150 --> 00:36:29,800 to kind of replicate some of those 889 00:36:29,800 --> 00:36:32,560 basic findings while studying some other things. 890 00:36:32,560 --> 00:36:36,040 And, in fact, when we did what everyone else does-- 891 00:36:36,040 --> 00:36:38,530 that is standard analysis-- 892 00:36:38,530 --> 00:36:41,170 you collect your diffusion-imaging data, 893 00:36:41,170 --> 00:36:43,660 and you eyeball it loosely, and if it really 894 00:36:43,660 --> 00:36:45,640 looks terribly tainted with artifact, 895 00:36:45,640 --> 00:36:46,790 you throw that subject out. 896 00:36:46,790 --> 00:36:49,870 And, otherwise, you keep it, and you analyze your data. 897 00:36:49,870 --> 00:36:52,750 And you look at, here are the 18 fiber tracts 898 00:36:52,750 --> 00:36:54,730 that I showed you before. 899 00:36:54,730 --> 00:36:58,450 And you ask, which of those have higher or lower 900 00:36:58,450 --> 00:37:01,300 fractional anisotropy in kids with autism 901 00:37:01,300 --> 00:37:03,190 compared to typical kids? 902 00:37:03,190 --> 00:37:04,840 And the basic finding-- 903 00:37:04,840 --> 00:37:06,610 we replicated the usual finding. 904 00:37:06,610 --> 00:37:07,780 And that is, overall-- 905 00:37:07,780 --> 00:37:10,510 this is column A-- most of those tracts 906 00:37:10,510 --> 00:37:12,910 showed lower fractional anisotropy 907 00:37:12,910 --> 00:37:17,230 in the kids with autism than the typical kids, OK? 908 00:37:17,230 --> 00:37:19,180 Many of those differences were individually 909 00:37:19,180 --> 00:37:21,430 significant in individual tracts. 910 00:37:21,430 --> 00:37:23,680 Those are the ones with the asterisks. 911 00:37:23,680 --> 00:37:24,190 OK. 912 00:37:24,190 --> 00:37:26,232 So that's the standard finding in the literature, 913 00:37:26,232 --> 00:37:27,730 and we replicated it. 914 00:37:27,730 --> 00:37:33,070 However, we noticed that a lot of the data 915 00:37:33,070 --> 00:37:34,300 really seemed suspect. 916 00:37:34,300 --> 00:37:36,190 And we started measuring the amount 917 00:37:36,190 --> 00:37:39,430 of head motion between the kids with autism and the kids 918 00:37:39,430 --> 00:37:40,600 without. 919 00:37:40,600 --> 00:37:41,740 And guess what. 920 00:37:41,740 --> 00:37:45,640 Kids with autism move in the scanner more than kids without. 921 00:37:45,640 --> 00:37:47,380 And guess what. 922 00:37:47,380 --> 00:37:51,040 Diffusion imaging and fractional anisotropy in particular 923 00:37:51,040 --> 00:37:53,690 are highly influenced by head motion. 924 00:37:53,690 --> 00:37:57,700 So then we said, OK, let's get a little more careful. 925 00:37:57,700 --> 00:37:59,980 And so we did a more stringent analysis. 926 00:37:59,980 --> 00:38:01,660 And we looked at the kids. 927 00:38:01,660 --> 00:38:04,270 We had quite a few of them in each group. 928 00:38:04,270 --> 00:38:06,370 And we took the subset of kids who we could 929 00:38:06,370 --> 00:38:08,680 match for head motion, OK? 930 00:38:08,680 --> 00:38:11,650 So now we've got the kids with autism and the typical kids, 931 00:38:11,650 --> 00:38:14,860 but we've now got the subset we have to choose 932 00:38:14,860 --> 00:38:16,700 to match for head motion, OK? 933 00:38:16,700 --> 00:38:19,300 It usually means the typical kids who move a little more 934 00:38:19,300 --> 00:38:21,910 and the autistic kids who had slightly less head motion. 935 00:38:21,910 --> 00:38:24,560 That's what you need to do to match them. 936 00:38:24,560 --> 00:38:28,010 And when we do that, the usual pattern disappears. 937 00:38:28,010 --> 00:38:30,400 Now there's only a single tract that 938 00:38:30,400 --> 00:38:32,470 shows lower fractional anisotropy 939 00:38:32,470 --> 00:38:36,520 in the kids with autism than the typical kids, OK? 940 00:38:36,520 --> 00:38:40,000 The inferior longitudinal fasciculus. 941 00:38:40,000 --> 00:38:44,350 So that's worrying. 942 00:38:44,350 --> 00:38:48,507 But then further, we thought, OK since many of those kids 943 00:38:48,507 --> 00:38:50,590 we had, especially the typical kids-- many of them 944 00:38:50,590 --> 00:38:52,900 we had scanned twice. 945 00:38:52,900 --> 00:38:55,690 So we thought, OK, let's really make this case. 946 00:38:55,690 --> 00:38:57,620 This is clearly a problem in the field. 947 00:38:57,620 --> 00:38:59,050 In fact, it's a broader problem. 948 00:38:59,050 --> 00:39:01,660 Pretty much any comparison across age groups 949 00:39:01,660 --> 00:39:02,950 or clinical groups-- 950 00:39:02,950 --> 00:39:05,095 one group moves more than the other group. 951 00:39:05,095 --> 00:39:06,310 Uh-oh. 952 00:39:06,310 --> 00:39:08,680 What about the entire literature? 953 00:39:08,680 --> 00:39:11,680 Hundreds, probably thousands of published papers, 954 00:39:11,680 --> 00:39:14,180 essentially, none of which pay attention to this-- 955 00:39:14,180 --> 00:39:15,370 this is 2014. 956 00:39:15,370 --> 00:39:18,460 People have cleaned up their act since, but up until 2014, 957 00:39:18,460 --> 00:39:20,140 almost none of them paid any attention 958 00:39:20,140 --> 00:39:22,160 to this whopping problem. 959 00:39:22,160 --> 00:39:24,365 So we figured we better make this point salient 960 00:39:24,365 --> 00:39:25,990 because there's a lot of money and time 961 00:39:25,990 --> 00:39:29,200 being wasted publishing garbage. 962 00:39:29,200 --> 00:39:31,030 And we want to make the point saliently. 963 00:39:31,030 --> 00:39:36,070 So we took the typical kids who we had scanned twice, OK? 964 00:39:36,070 --> 00:39:38,170 And sometimes a kid will-- the same kid 965 00:39:38,170 --> 00:39:41,140 will move more in one session than another session. 966 00:39:41,140 --> 00:39:44,950 So we said, OK, let's compare the very same kids 967 00:39:44,950 --> 00:39:47,440 on the session where they move more than the session 968 00:39:47,440 --> 00:39:48,640 where they moved less. 969 00:39:48,640 --> 00:39:49,990 And you know what? 970 00:39:49,990 --> 00:39:53,160 We replicated the autism phenotype. 971 00:39:53,160 --> 00:39:55,980 Those were typical kids, not autistic kids. 972 00:39:55,980 --> 00:39:58,860 The point is head motion alone will 973 00:39:58,860 --> 00:40:01,740 reduce fractional anisotropy and will look a whole lot 974 00:40:01,740 --> 00:40:02,980 like a clinical disorder. 975 00:40:02,980 --> 00:40:05,820 And so every time you see that a clinical disorder is marked 976 00:40:05,820 --> 00:40:08,850 by some anatomical difference, your first thought should be, 977 00:40:08,850 --> 00:40:11,340 how carefully did they deal with head 978 00:40:11,340 --> 00:40:13,020 motion and other artifacts that are 979 00:40:13,020 --> 00:40:15,220 going to differ between groups? 980 00:40:15,220 --> 00:40:16,830 OK? 981 00:40:16,830 --> 00:40:19,140 So I say that not to dis the entire literature 982 00:40:19,140 --> 00:40:22,770 but just to alert you that these things can really matter. 983 00:40:22,770 --> 00:40:25,710 The paper from Gabrielli Lab that I just described-- 984 00:40:25,710 --> 00:40:27,927 I looked, of course, before I presented it in here, 985 00:40:27,927 --> 00:40:29,760 and they cited us, and they used our methods 986 00:40:29,760 --> 00:40:31,020 for matching head motion-- 987 00:40:31,020 --> 00:40:32,080 good for them. 988 00:40:32,080 --> 00:40:34,860 So these things are changing, and I 989 00:40:34,860 --> 00:40:37,380 think the field will start cleaning up its act. 990 00:40:37,380 --> 00:40:42,590 But it's amazing it took this long. 991 00:40:42,590 --> 00:40:44,800 OK. 992 00:40:44,800 --> 00:40:47,440 All right, so finding fiber tracts 993 00:40:47,440 --> 00:40:50,770 and characterizing them with fractional anisotropy 994 00:40:50,770 --> 00:40:53,260 are nice, but, really, what we want to know 995 00:40:53,260 --> 00:40:56,140 is what's connected to what, OK? 996 00:40:56,140 --> 00:40:58,370 Which of these things are connected to each other? 997 00:40:58,370 --> 00:41:01,210 Which other brain regions are they connected to? 998 00:41:01,210 --> 00:41:06,070 And so to find that out, we need to not just study 999 00:41:06,070 --> 00:41:08,050 white matter itself and the tracts, 1000 00:41:08,050 --> 00:41:10,150 but we have to get out of the tracts 1001 00:41:10,150 --> 00:41:12,100 and into the gray matter. 1002 00:41:12,100 --> 00:41:14,620 So we need to start in a patch of gray matter 1003 00:41:14,620 --> 00:41:18,910 and figure out where we can go by following those axons, OK? 1004 00:41:18,910 --> 00:41:22,060 And so the method for doing that is called tractography. 1005 00:41:22,060 --> 00:41:23,560 So there are many versions of this. 1006 00:41:23,560 --> 00:41:25,810 I showed briefly these pictures before. 1007 00:41:25,810 --> 00:41:28,246 I'm sure you've seen these. 1008 00:41:28,246 --> 00:41:30,460 The simplest idea of what you do, 1009 00:41:30,460 --> 00:41:34,660 leaving out all the details, is you 1010 00:41:34,660 --> 00:41:37,000 start in some gray-matter region, 1011 00:41:37,000 --> 00:41:38,680 some voxel in the gray matter. 1012 00:41:38,680 --> 00:41:40,900 You want to know what it's connected to. 1013 00:41:40,900 --> 00:41:44,365 You just follow those little orientations, and you can-- 1014 00:41:44,365 --> 00:41:46,780 you see where you can go. 1015 00:41:46,780 --> 00:41:48,250 OK? 1016 00:41:48,250 --> 00:41:50,850 And so that's basically how you make this diagram. 1017 00:41:50,850 --> 00:41:51,350 OK? 1018 00:41:51,350 --> 00:41:52,660 You're just following those. 1019 00:41:52,660 --> 00:41:55,118 You start over here you follow the orientation-- go do, do, 1020 00:41:55,118 --> 00:41:56,260 do, do, do, do, do, right? 1021 00:41:56,260 --> 00:41:57,490 OK? 1022 00:41:57,490 --> 00:41:59,590 I mean, you do that in a computer, right? 1023 00:41:59,590 --> 00:42:02,560 An algorithm does that, follows along-- ch, ch, OK? 1024 00:42:02,560 --> 00:42:04,720 OK, so that's called tractography. 1025 00:42:04,720 --> 00:42:06,580 And the idea's awesome-- how great 1026 00:42:06,580 --> 00:42:09,710 to be able to see what's connected to what. 1027 00:42:09,710 --> 00:42:11,530 And there are many, many thousands 1028 00:42:11,530 --> 00:42:14,410 of papers that do this for good reason. 1029 00:42:14,410 --> 00:42:16,540 We need to know what's connected to what. 1030 00:42:16,540 --> 00:42:18,640 This is our currently best method 1031 00:42:18,640 --> 00:42:20,800 for looking at the structural connectivity 1032 00:42:20,800 --> 00:42:24,950 of different gray-matter regions to each other. 1033 00:42:24,950 --> 00:42:27,160 And so you can ask, for example, OK, 1034 00:42:27,160 --> 00:42:29,620 let's put a seed in the fusiform face area 1035 00:42:29,620 --> 00:42:31,090 and see where it goes. 1036 00:42:31,090 --> 00:42:34,060 Wouldn't that be cool? 1037 00:42:34,060 --> 00:42:35,380 Right. 1038 00:42:35,380 --> 00:42:37,030 Wouldn't it be cool? 1039 00:42:37,030 --> 00:42:40,820 Unfortunately, it doesn't work. 1040 00:42:40,820 --> 00:42:42,640 So I have to tell you that I don't 1041 00:42:42,640 --> 00:42:45,560 know if I'm the best person to report to this because I'm 1042 00:42:45,560 --> 00:42:46,060 not-- 1043 00:42:46,060 --> 00:42:48,308 I've only been trying to do this for a few years. 1044 00:42:48,308 --> 00:42:50,350 But, I've been collaborating with the best people 1045 00:42:50,350 --> 00:42:52,960 in the world over there at MGH Charlestown who 1046 00:42:52,960 --> 00:42:54,230 are working closely with us. 1047 00:42:54,230 --> 00:42:57,310 And we can't get this thing to work worth a damn. 1048 00:42:57,310 --> 00:43:00,610 And so now I'm actually confused whether the entire literature 1049 00:43:00,610 --> 00:43:01,330 is garbage. 1050 00:43:01,330 --> 00:43:02,920 I don't think it's entirely garbage. 1051 00:43:02,920 --> 00:43:06,700 But I think it's full of overoptimistic evaluations 1052 00:43:06,700 --> 00:43:10,210 of what you can tell from tractography 1053 00:43:10,210 --> 00:43:13,990 because in our hands, we started with reality checks, 1054 00:43:13,990 --> 00:43:16,270 put a seed in the lateral geniculate nucleus. 1055 00:43:16,270 --> 00:43:19,450 Let's make damn sure we can get up to V1. 1056 00:43:19,450 --> 00:43:21,700 Well, you can get up to V1, but you 1057 00:43:21,700 --> 00:43:24,250 can get up to V2, and V3, and V4, 1058 00:43:24,250 --> 00:43:26,320 as well, which are all wrong, right? 1059 00:43:26,320 --> 00:43:28,540 LGN only goes to V1. 1060 00:43:28,540 --> 00:43:32,428 Worse, you stick a seed next door in the medial geniculate 1061 00:43:32,428 --> 00:43:34,720 nucleus, which is the part of the thalamus that goes up 1062 00:43:34,720 --> 00:43:37,990 to auditory cortex, you also end up in V1. 1063 00:43:37,990 --> 00:43:38,500 Wrong. 1064 00:43:38,500 --> 00:43:38,890 Wrong. 1065 00:43:38,890 --> 00:43:39,390 Wrong. 1066 00:43:39,390 --> 00:43:40,570 Wrong. 1067 00:43:40,570 --> 00:43:44,155 There's not very many anatomical connections in the human brain 1068 00:43:44,155 --> 00:43:46,030 where we actually know the right answer where 1069 00:43:46,030 --> 00:43:48,640 we can do these reality checks, but of the ones 1070 00:43:48,640 --> 00:43:51,687 we know that we've tried, it doesn't work. 1071 00:43:51,687 --> 00:43:53,770 And we're using the best diffusion-imaging scanner 1072 00:43:53,770 --> 00:43:54,340 in the world. 1073 00:43:54,340 --> 00:43:55,880 It's right over there. 1074 00:43:55,880 --> 00:43:57,610 So maybe I'm doing everything wrong. 1075 00:43:57,610 --> 00:43:59,960 But at the very least, I think there 1076 00:43:59,960 --> 00:44:01,585 are a lot of problems with this method. 1077 00:44:05,550 --> 00:44:07,590 This is not just me worrying about this. 1078 00:44:07,590 --> 00:44:09,007 And many people have been worrying 1079 00:44:09,007 --> 00:44:12,090 about this for the whole 15, 20-year life of diffusion 1080 00:44:12,090 --> 00:44:12,810 tractography. 1081 00:44:12,810 --> 00:44:15,480 And some of the challenges are, like, famous. 1082 00:44:15,480 --> 00:44:17,335 So to follow those little orientations, 1083 00:44:17,335 --> 00:44:19,710 you need to-- you can see, like there'd be lots of places 1084 00:44:19,710 --> 00:44:23,220 where, OK, there's a bunch of different ways you could go. 1085 00:44:23,220 --> 00:44:24,180 It's ill posed, right? 1086 00:44:24,180 --> 00:44:26,970 So people use heuristics to constrain those solutions. 1087 00:44:26,970 --> 00:44:29,100 And those heuristics are based on assumptions 1088 00:44:29,100 --> 00:44:31,320 about how fibers bend in the brain, namely 1089 00:44:31,320 --> 00:44:34,290 that they don't make really sharp angles, right? 1090 00:44:34,290 --> 00:44:35,110 That's reasonable. 1091 00:44:35,110 --> 00:44:38,260 Most of the time they don't, but sometimes they do. 1092 00:44:38,260 --> 00:44:41,190 And in particular, when you're going from white matter 1093 00:44:41,190 --> 00:44:44,370 to cortex, often, you make a very sharp turn. 1094 00:44:44,370 --> 00:44:46,650 And so it's very, very difficult to figure out 1095 00:44:46,650 --> 00:44:48,930 how to get from a given gray-matter patch 1096 00:44:48,930 --> 00:44:51,510 into the underlying white matter exactly what 1097 00:44:51,510 --> 00:44:53,130 the connectivity is. 1098 00:44:53,130 --> 00:44:55,330 So that's one problem. 1099 00:44:55,330 --> 00:44:59,100 Another famous problem with tractography 1100 00:44:59,100 --> 00:45:01,510 is called the crossing-fiber problem. 1101 00:45:01,510 --> 00:45:04,380 So imagine a bunch of axons somewhere in the brain 1102 00:45:04,380 --> 00:45:08,820 that cross like this versus imagine a bunch of fibers 1103 00:45:08,820 --> 00:45:12,660 in the brain that come up to each other and then go apart, 1104 00:45:12,660 --> 00:45:13,290 OK? 1105 00:45:13,290 --> 00:45:14,082 Everybody get this? 1106 00:45:14,082 --> 00:45:16,935 The connectivity's totally different here-- 1107 00:45:16,935 --> 00:45:18,810 no way you're ever going to distinguish those 1108 00:45:18,810 --> 00:45:20,908 with diffusion tractography. 1109 00:45:20,908 --> 00:45:22,950 So people try to get higher and higher resolution 1110 00:45:22,950 --> 00:45:27,210 to see down those individual things, but they're not there. 1111 00:45:27,210 --> 00:45:28,430 Yeah. 1112 00:45:28,430 --> 00:45:31,233 AUDIENCE: Why would something like that happen? 1113 00:45:31,233 --> 00:45:33,150 NANCY KANWISHER: Yeah, why would they do that? 1114 00:45:33,150 --> 00:45:36,720 Weird stuff happens in the brain. 1115 00:45:36,720 --> 00:45:40,860 So it's not incredibly common, but it's not unheard of. 1116 00:45:40,860 --> 00:45:45,270 Yeah, remember, the brain wasn't dissolved-- 1117 00:45:45,270 --> 00:45:49,680 designed now optimally to solve all the problems 1118 00:45:49,680 --> 00:45:54,180 it needs to solve with the optimal solution from scratch. 1119 00:45:54,180 --> 00:45:56,030 It evolved gradually over time. 1120 00:45:56,030 --> 00:45:57,780 And so there are all kinds of weird things 1121 00:45:57,780 --> 00:46:00,840 that are workarounds for pre-existing decisions 1122 00:46:00,840 --> 00:46:02,790 that evolution made earlier. 1123 00:46:02,790 --> 00:46:07,800 And so both brain and body have lots of bizarre attributes 1124 00:46:07,800 --> 00:46:10,180 that aren't how you would design it from scratch. 1125 00:46:10,180 --> 00:46:15,550 They're just the fix that evolution made at that point 1126 00:46:15,550 --> 00:46:18,930 given what had already been fixed. 1127 00:46:18,930 --> 00:46:21,550 And so there's weird stuff like that. 1128 00:46:21,550 --> 00:46:23,490 Anyway, I mention this to say I'm 1129 00:46:23,490 --> 00:46:26,130 more negative about diffusion tractography than probably 1130 00:46:26,130 --> 00:46:28,547 anyone else because I've spent a lot of the last two years 1131 00:46:28,547 --> 00:46:31,050 trying to do it, and it's big bust, and I'm cranky. 1132 00:46:31,050 --> 00:46:33,300 So it's probably not as bad as I'm laying out. 1133 00:46:33,300 --> 00:46:34,260 Plenty of people do it. 1134 00:46:34,260 --> 00:46:36,030 They get some kind of answers out of it, 1135 00:46:36,030 --> 00:46:39,450 but it's problematic at least. 1136 00:46:39,450 --> 00:46:43,770 My best guess is that it is OK for fingerprints. 1137 00:46:43,770 --> 00:46:46,590 If you're asking, OK, here's some patch of brain. 1138 00:46:46,590 --> 00:46:50,880 How much does it connect to, say, these 85 other regions? 1139 00:46:50,880 --> 00:46:53,700 And is that different than the fingerprint for this region? 1140 00:46:53,700 --> 00:46:57,780 That's probably OK because a lot of those individual solutions 1141 00:46:57,780 --> 00:46:59,970 might be wrong, and there's still enough left over 1142 00:46:59,970 --> 00:47:01,840 to see a kind of difference. 1143 00:47:01,840 --> 00:47:04,050 So I feel like you can-- conductivity fingerprints 1144 00:47:04,050 --> 00:47:06,130 are probably worth doing. 1145 00:47:06,130 --> 00:47:08,680 But, actually, just answering the question of, 1146 00:47:08,680 --> 00:47:11,400 is there a structural connection from A to B-- 1147 00:47:11,400 --> 00:47:12,270 I don't know. 1148 00:47:12,270 --> 00:47:14,140 I can't get it to work. 1149 00:47:14,140 --> 00:47:15,150 OK? 1150 00:47:15,150 --> 00:47:18,358 OK, so I think I did all this before just to-- 1151 00:47:18,358 --> 00:47:20,400 conductivity fingerprints-- do you remember this? 1152 00:47:20,400 --> 00:47:23,250 You start with one place, and you 1153 00:47:23,250 --> 00:47:25,410 measure how well you can get from each location 1154 00:47:25,410 --> 00:47:26,850 to each of these other ones. 1155 00:47:26,850 --> 00:47:30,240 And I showed you before that the work 1156 00:47:30,240 --> 00:47:33,090 of Zeynep Saygin and a bunch of other people 1157 00:47:33,090 --> 00:47:35,790 has shown that you can-- actually, in an adult, 1158 00:47:35,790 --> 00:47:40,290 you can predict where that adult's fusiform face 1159 00:47:40,290 --> 00:47:43,560 area is just from their diffusion tractography data 1160 00:47:43,560 --> 00:47:46,470 alone because it has a distinctive connectivity 1161 00:47:46,470 --> 00:47:48,510 fingerprint, OK? 1162 00:47:48,510 --> 00:47:50,260 I don't want to go through all that again. 1163 00:47:50,260 --> 00:47:52,343 Do you guys remember that, more or less, the gist? 1164 00:47:52,343 --> 00:47:53,980 OK, so that just tells you that there 1165 00:47:53,980 --> 00:47:58,240 is this systematic mapping between the connectivity 1166 00:47:58,240 --> 00:48:00,880 of a region and its function. 1167 00:48:00,880 --> 00:48:04,532 And connectivity fingerprints, despite all these problems 1168 00:48:04,532 --> 00:48:06,490 I've been carrying on about, have enough signal 1169 00:48:06,490 --> 00:48:11,280 left in there to predict the function of a region 1170 00:48:11,280 --> 00:48:17,600 and maybe to say something about homologies across species. 1171 00:48:17,600 --> 00:48:20,030 OK, blah, blah, blah. 1172 00:48:20,030 --> 00:48:20,690 Right. 1173 00:48:20,690 --> 00:48:21,200 OK. 1174 00:48:21,200 --> 00:48:22,760 So where do we get? 1175 00:48:22,760 --> 00:48:25,790 You can find the major fiber bundles with diffusion imaging. 1176 00:48:25,790 --> 00:48:26,900 That's worthwhile. 1177 00:48:26,900 --> 00:48:29,730 You can characterize fractional anisotropy. 1178 00:48:29,730 --> 00:48:33,020 I don't really know what it means, but it means something. 1179 00:48:33,020 --> 00:48:35,330 And you can find very approximate connectivity 1180 00:48:35,330 --> 00:48:38,900 fingerprints good enough the predict function. 1181 00:48:38,900 --> 00:48:41,570 OK, so that's worthwhile. 1182 00:48:41,570 --> 00:48:44,150 But actual structural connections of one 1183 00:48:44,150 --> 00:48:46,760 particular cortical area-- 1184 00:48:46,760 --> 00:48:48,110 not very good. 1185 00:48:48,110 --> 00:48:51,210 At best, it's a weak signal. 1186 00:48:51,210 --> 00:48:53,040 So that's a drag. 1187 00:48:53,040 --> 00:48:55,410 So let's consider the other method 1188 00:48:55,410 --> 00:48:59,160 people have used to try to work these things out. 1189 00:48:59,160 --> 00:49:02,700 And that's resting functional correlations. 1190 00:49:02,700 --> 00:49:06,120 So let me describe where this story starts. 1191 00:49:06,120 --> 00:49:10,440 This story starts with a paper in 1995 by Biswal, 1192 00:49:10,440 --> 00:49:12,370 and this is the figure from his paper. 1193 00:49:12,370 --> 00:49:14,730 So, first he had people move-- 1194 00:49:14,730 --> 00:49:17,280 they had-- he had people in the scanner doing finger-tapping. 1195 00:49:17,280 --> 00:49:18,613 So they're lying in the scanner. 1196 00:49:18,613 --> 00:49:22,140 He's scanning their brains while they tap both fingers or not, 1197 00:49:22,140 --> 00:49:24,310 or tap both fingers or not. 1198 00:49:24,310 --> 00:49:26,310 And you get these-- it's hard to see-- these two 1199 00:49:26,310 --> 00:49:28,320 little bits of motor cortex corresponding 1200 00:49:28,320 --> 00:49:30,210 to the finger-motor region. 1201 00:49:30,210 --> 00:49:31,480 OK, no surprise there. 1202 00:49:31,480 --> 00:49:34,320 We're just mapping a little bit of motor cortex. 1203 00:49:34,320 --> 00:49:35,730 But then he does something cool. 1204 00:49:38,520 --> 00:49:42,480 He looks at the time course over that experiment 1205 00:49:42,480 --> 00:49:44,010 in one of those motor regions-- 1206 00:49:44,010 --> 00:49:49,110 in one hemisphere, and he looks at the time course 1207 00:49:49,110 --> 00:49:54,150 in the other hemisphere when the subject is at rest, not 1208 00:49:54,150 --> 00:49:55,652 doing anything, OK? 1209 00:49:55,652 --> 00:49:56,610 Sorry, I left this out. 1210 00:49:56,610 --> 00:49:58,027 You scan them doing this, and then 1211 00:49:58,027 --> 00:50:01,200 you scan them just lying there going, dum-de dum-de dum, 1212 00:50:01,200 --> 00:50:05,170 or whatever you do when you don't have a task, OK? 1213 00:50:05,170 --> 00:50:10,180 And he finds that these very far-apart regions at rest, 1214 00:50:10,180 --> 00:50:12,730 when the subject is not tapping their fingers, 1215 00:50:12,730 --> 00:50:17,010 are extremely correlated. 1216 00:50:17,010 --> 00:50:18,900 So that's very not obvious. 1217 00:50:18,900 --> 00:50:21,660 These things are centimeters apart. 1218 00:50:21,660 --> 00:50:23,842 The subject isn't doing anything in particular. 1219 00:50:23,842 --> 00:50:25,300 You're not telling them what to do. 1220 00:50:25,300 --> 00:50:27,640 And they certainly aren't tapping their fingers. 1221 00:50:27,640 --> 00:50:31,080 So why are these two bits of finger-motor cortex going up 1222 00:50:31,080 --> 00:50:34,940 and down in lockstep like that? 1223 00:50:34,940 --> 00:50:37,340 Well, nobody knows, actually. 1224 00:50:37,340 --> 00:50:39,800 I mean, this was however-- 1225 00:50:39,800 --> 00:50:42,658 20-some years ago, right? 1226 00:50:42,658 --> 00:50:45,200 Still, nobody really knows why those damn things are going up 1227 00:50:45,200 --> 00:50:46,610 in lockstep like that. 1228 00:50:46,610 --> 00:50:49,460 But it's systematic, it's tantalizing, 1229 00:50:49,460 --> 00:50:52,525 it makes you want to play more, and many people have. 1230 00:50:52,525 --> 00:50:53,900 OK, so I would say we still don't 1231 00:50:53,900 --> 00:50:56,960 know exactly why those things are going up and down together. 1232 00:50:56,960 --> 00:50:59,630 But the pattern of brain regions that go up and down together 1233 00:50:59,630 --> 00:51:01,970 has proven to be a whole fascinating window 1234 00:51:01,970 --> 00:51:03,320 into the brain. 1235 00:51:03,320 --> 00:51:05,030 OK, so that's our next topic here. 1236 00:51:08,720 --> 00:51:14,330 OK, so here's another depiction of more exactly what you do. 1237 00:51:14,330 --> 00:51:17,180 OK, so step 1-- 1238 00:51:17,180 --> 00:51:21,560 you find a seed region in here, in left somatosensory motor 1239 00:51:21,560 --> 00:51:22,820 cortex, OK? 1240 00:51:22,820 --> 00:51:24,020 So that's that region there. 1241 00:51:24,020 --> 00:51:27,410 You get its time course, OK? 1242 00:51:27,410 --> 00:51:28,567 Sorry, at rest. 1243 00:51:28,567 --> 00:51:30,650 You find that region, and then you scan the person 1244 00:51:30,650 --> 00:51:33,800 while they're just told to do nothing in particular. 1245 00:51:33,800 --> 00:51:37,310 You get the time course averaged over all those voxels at rest. 1246 00:51:37,310 --> 00:51:39,500 There it is, OK? 1247 00:51:39,500 --> 00:51:42,290 Now you take that time course. 1248 00:51:42,290 --> 00:51:44,450 And you correlate it with the time course 1249 00:51:44,450 --> 00:51:46,520 of every other voxel in the brain. 1250 00:51:46,520 --> 00:51:48,500 And you say, show me all the voxels 1251 00:51:48,500 --> 00:51:52,790 that are correlated with this region at rest. 1252 00:51:52,790 --> 00:51:54,640 And you get this-- 1253 00:51:54,640 --> 00:51:56,950 lots of systematic brain regions that 1254 00:51:56,950 --> 00:52:00,100 are highly correlated at rest with that region 1255 00:52:00,100 --> 00:52:02,700 you started with. 1256 00:52:02,700 --> 00:52:06,100 Everybody get what we just did? 1257 00:52:06,100 --> 00:52:08,080 OK, totally non obvious. 1258 00:52:08,080 --> 00:52:09,640 Well, you might say, OK, fine. 1259 00:52:09,640 --> 00:52:10,807 This is finger-motor cortex. 1260 00:52:10,807 --> 00:52:11,723 This is the other one. 1261 00:52:11,723 --> 00:52:13,600 That's what I showed you from Biswal before. 1262 00:52:13,600 --> 00:52:14,950 But why this thing? 1263 00:52:14,950 --> 00:52:17,030 Why this thing way down deep in the brain? 1264 00:52:17,030 --> 00:52:19,330 Why that thing down in the cerebellum miles 1265 00:52:19,330 --> 00:52:20,900 away in the brain? 1266 00:52:20,900 --> 00:52:24,220 Why are they all in cahoots with each other. 1267 00:52:24,220 --> 00:52:26,620 I like using "in cahoots" when talking about correlations 1268 00:52:26,620 --> 00:52:28,620 because nobody knows what the correlations mean. 1269 00:52:28,620 --> 00:52:33,890 So "in cahoots" is as technical as I think we should get. 1270 00:52:33,890 --> 00:52:34,390 Yeah? 1271 00:52:34,390 --> 00:52:36,307 AUDIENCE: --correlations without a time shift. 1272 00:52:36,307 --> 00:52:37,270 Just [INAUDIBLE]. 1273 00:52:37,270 --> 00:52:37,990 NANCY KANWISHER: Good question. 1274 00:52:37,990 --> 00:52:38,740 Good question. 1275 00:52:38,740 --> 00:52:40,960 Wouldn't we love to know about the time shift? 1276 00:52:40,960 --> 00:52:42,652 But here's the problem. 1277 00:52:42,652 --> 00:52:44,860 There should be a time shift because it takes a while 1278 00:52:44,860 --> 00:52:48,820 to conduct down axons from here to here, probably 1279 00:52:48,820 --> 00:52:50,532 a few milliseconds. 1280 00:52:50,532 --> 00:52:52,240 But a few milliseconds we are never going 1281 00:52:52,240 --> 00:52:54,600 to see with functional MRI. 1282 00:52:54,600 --> 00:52:56,660 So, surely, there is a time shift, 1283 00:52:56,660 --> 00:53:01,770 but this method can't exploit it, OK? 1284 00:53:01,770 --> 00:53:03,400 Yeah, I'll just leave it at that. 1285 00:53:03,400 --> 00:53:05,520 OK, but does everybody get what this map is? 1286 00:53:05,520 --> 00:53:07,980 We've just chosen a seed region, a starting 1287 00:53:07,980 --> 00:53:09,570 point just for the hell of it. 1288 00:53:09,570 --> 00:53:11,790 And we've asked, what other bits of the brain 1289 00:53:11,790 --> 00:53:13,920 are correlated with that region at rest? 1290 00:53:13,920 --> 00:53:15,840 It's a pretty weird thing to do. 1291 00:53:15,840 --> 00:53:19,080 And you wouldn't do it if you didn't find systematically 1292 00:53:19,080 --> 00:53:23,250 replicable answers that are repeatable across subjects. 1293 00:53:23,250 --> 00:53:24,990 And when that happens, you go, OK, I 1294 00:53:24,990 --> 00:53:27,570 don't know what this means, but it's pretty systematic. 1295 00:53:27,570 --> 00:53:29,630 Let's keep following the thread. 1296 00:53:29,630 --> 00:53:30,630 OK? 1297 00:53:30,630 --> 00:53:31,740 Question? 1298 00:53:31,740 --> 00:53:35,357 AUDIENCE: What do you mean by correlated [INAUDIBLE]?? 1299 00:53:35,357 --> 00:53:37,190 NANCY KANWISHER: OK, so let's do this again. 1300 00:53:37,190 --> 00:53:39,390 You scan people moving their fingers. 1301 00:53:39,390 --> 00:53:41,560 You find little-finger region here. 1302 00:53:41,560 --> 00:53:44,200 Now you scan the same person just in the scanner. 1303 00:53:44,200 --> 00:53:46,200 You say, I'm going to scan you for five minutes. 1304 00:53:46,200 --> 00:53:48,930 Just close your eyes and don't do anything in particular. 1305 00:53:48,930 --> 00:53:49,530 You lie there. 1306 00:53:49,530 --> 00:53:51,420 I scan your brain. 1307 00:53:51,420 --> 00:53:54,210 Now I take that region, which I found before. 1308 00:53:54,210 --> 00:53:56,580 And I take the time course of that region 1309 00:53:56,580 --> 00:53:59,260 while you were just lying in the scanner doing nothing, 1310 00:53:59,260 --> 00:54:02,550 and I get some randomish-looking thing like this. 1311 00:54:02,550 --> 00:54:05,490 Now I take that time course, and I say, 1312 00:54:05,490 --> 00:54:08,040 let's see if there are any other voxels in your brain 1313 00:54:08,040 --> 00:54:11,220 that were correlated when you were lying there with that time 1314 00:54:11,220 --> 00:54:13,290 course. 1315 00:54:13,290 --> 00:54:16,170 And I color them in, and there are lots of them, 1316 00:54:16,170 --> 00:54:19,790 even regions that are far away. 1317 00:54:19,790 --> 00:54:20,540 OK? 1318 00:54:20,540 --> 00:54:22,160 It's really not obvious. 1319 00:54:22,160 --> 00:54:25,280 You wouldn't have predicted this would happen, yeah? 1320 00:54:25,280 --> 00:54:26,390 Yeah? 1321 00:54:26,390 --> 00:54:28,460 AUDIENCE: So if you look at the brain just as 1322 00:54:28,460 --> 00:54:33,580 some underlying resting rhythm and like just all regions 1323 00:54:33,580 --> 00:54:35,778 of the brain just have some resting rhythm, 1324 00:54:35,778 --> 00:54:37,570 wouldn't it be just always be [INAUDIBLE]?? 1325 00:54:37,570 --> 00:54:38,487 NANCY KANWISHER: Yeah. 1326 00:54:38,487 --> 00:54:41,200 OK, so there's been a whole suite of speculations 1327 00:54:41,200 --> 00:54:42,640 of exactly that kind. 1328 00:54:42,640 --> 00:54:45,640 Are there endogenous rhythms that 1329 00:54:45,640 --> 00:54:47,920 are characteristic of particular brain regions 1330 00:54:47,920 --> 00:54:49,510 and so those things go together? 1331 00:54:49,510 --> 00:54:53,020 Maybe, but so far, that doesn't seem to be the main answer. 1332 00:54:53,020 --> 00:54:54,520 For a long time, people thought, OK, 1333 00:54:54,520 --> 00:54:56,080 is it just blood-flow supply? 1334 00:54:56,080 --> 00:54:58,540 Maybe the blood-flow supply to the brain branches 1335 00:54:58,540 --> 00:55:00,550 and feeds those regions, and that somehow 1336 00:55:00,550 --> 00:55:02,753 regulates the bold response in those regions. 1337 00:55:02,753 --> 00:55:05,170 There have been many accounts like this, and none of those 1338 00:55:05,170 --> 00:55:06,400 seem to really capture it. 1339 00:55:06,400 --> 00:55:10,120 It really seems like, probably, those neurons are firing 1340 00:55:10,120 --> 00:55:12,850 in sync with each other, right? 1341 00:55:12,850 --> 00:55:15,050 Yeah, question, Nava? 1342 00:55:15,050 --> 00:55:17,740 AUDIENCE: Yes, before a question in direction-- 1343 00:55:17,740 --> 00:55:20,230 if you have a time delay, I guess 1344 00:55:20,230 --> 00:55:22,960 the question was because if you would have a time delay, 1345 00:55:22,960 --> 00:55:24,730 you could see what's further right. 1346 00:55:24,730 --> 00:55:26,380 Can you, instead of the time delay, 1347 00:55:26,380 --> 00:55:30,550 see if you measure it in one of the regions that seems to be-- 1348 00:55:30,550 --> 00:55:33,580 seems to be correlated-- if you measure from one of those, 1349 00:55:33,580 --> 00:55:36,310 if you could estimate the distance based on how strongly 1350 00:55:36,310 --> 00:55:37,543 they correlate [INAUDIBLE]? 1351 00:55:37,543 --> 00:55:39,460 NANCY KANWISHER: Oh, with the thought of maybe 1352 00:55:39,460 --> 00:55:41,085 there's not a time delay, but maybe you 1353 00:55:41,085 --> 00:55:44,220 lose some of your correlation with distance. 1354 00:55:44,220 --> 00:55:44,850 You could. 1355 00:55:44,850 --> 00:55:48,510 But just looking at this, these guys are pretty far apart. 1356 00:55:48,510 --> 00:55:52,005 So it's certainly not that there's just things are-- 1357 00:55:52,005 --> 00:55:55,260 that nearby things have a similar. 1358 00:55:55,260 --> 00:55:58,290 AUDIENCE: No, but I mean, if you would 1359 00:55:58,290 --> 00:56:00,660 say those are five different regions, 1360 00:56:00,660 --> 00:56:01,910 did you measure from region 1? 1361 00:56:01,910 --> 00:56:03,535 NANCY KANWISHER: Yeah, yeah, I got you. 1362 00:56:03,535 --> 00:56:04,110 Yeah. 1363 00:56:04,110 --> 00:56:04,610 Yeah. 1364 00:56:04,610 --> 00:56:05,490 Yeah, you could. 1365 00:56:05,490 --> 00:56:07,410 I think that's not going to work because there 1366 00:56:07,410 --> 00:56:09,240 are big, big correlations that people find 1367 00:56:09,240 --> 00:56:12,210 between very distant regions. 1368 00:56:12,210 --> 00:56:13,110 I'll show you more. 1369 00:56:13,110 --> 00:56:13,440 Yeah. 1370 00:56:13,440 --> 00:56:16,065 I mean, you could try that, and I'm sure people have done that. 1371 00:56:16,065 --> 00:56:17,430 I can't tell you exactly where. 1372 00:56:17,430 --> 00:56:20,370 Actually, they do it as part of their-- one of the common ways 1373 00:56:20,370 --> 00:56:23,640 you normalize your data is to take this and normalize it 1374 00:56:23,640 --> 00:56:27,240 for distance from the seed region, which would be a way 1375 00:56:27,240 --> 00:56:28,277 to build that factor in. 1376 00:56:28,277 --> 00:56:30,360 And once you do that, you still get lots of stuff. 1377 00:56:30,360 --> 00:56:32,190 AUDIENCE: They take the distance and the image space 1378 00:56:32,190 --> 00:56:32,760 for that, right? 1379 00:56:32,760 --> 00:56:34,677 NANCY KANWISHER: You can do it different ways. 1380 00:56:34,677 --> 00:56:36,780 There's an algorithm that somebody at MGH 1381 00:56:36,780 --> 00:56:41,340 wrote that is distanced by, most likely, a white-matter path 1382 00:56:41,340 --> 00:56:44,048 or as the crow flies, not that the crow can fly straight 1383 00:56:44,048 --> 00:56:45,840 through the brain, but you see what I mean. 1384 00:56:48,810 --> 00:56:49,350 Yeah. 1385 00:56:49,350 --> 00:56:50,160 Sorry, go ahead. 1386 00:56:50,160 --> 00:56:51,535 AUDIENCE: Is the result different 1387 00:56:51,535 --> 00:56:53,640 in if you measure the correlation when 1388 00:56:53,640 --> 00:56:56,850 they're doing the finger-tapping action versus [INAUDIBLE]?? 1389 00:56:56,850 --> 00:56:57,810 NANCY KANWISHER: Yeah. 1390 00:56:57,810 --> 00:57:00,582 OK, so this is a really important question, 1391 00:57:00,582 --> 00:57:02,040 and it's a whole part of this field 1392 00:57:02,040 --> 00:57:03,270 that I'm leaving out of this lecture 1393 00:57:03,270 --> 00:57:04,980 because I'm sort of suspicious of it. 1394 00:57:04,980 --> 00:57:07,320 But your question is a good one. 1395 00:57:07,320 --> 00:57:09,270 So you're saying, would they be correlated 1396 00:57:09,270 --> 00:57:10,437 while you're finger-tapping? 1397 00:57:10,437 --> 00:57:12,780 Well, certainly, if we did the paradigm 1398 00:57:12,780 --> 00:57:14,272 while they're tapping both fingers, 1399 00:57:14,272 --> 00:57:16,230 they're going to be correlated because we built 1400 00:57:16,230 --> 00:57:17,700 the correlation into the task. 1401 00:57:17,700 --> 00:57:19,510 We said, while you're doing this, do that. 1402 00:57:19,510 --> 00:57:22,325 And so they will surely be correlated. 1403 00:57:22,325 --> 00:57:23,700 And so there's a whole enterprise 1404 00:57:23,700 --> 00:57:27,120 where people try to factor out those things and ask, 1405 00:57:27,120 --> 00:57:30,000 even after you account for the activation of the task, 1406 00:57:30,000 --> 00:57:32,910 are there changes in these patterns of correlation 1407 00:57:32,910 --> 00:57:35,700 with the task you're doing? 1408 00:57:35,700 --> 00:57:39,900 And that's called PPI for physiological interactions. 1409 00:57:39,900 --> 00:57:41,910 And lots and lots of people do it-- 1410 00:57:41,910 --> 00:57:43,200 hundreds, thousands of papers. 1411 00:57:43,200 --> 00:57:44,575 It's probably pretty respectable, 1412 00:57:44,575 --> 00:57:46,890 but it drives me nuts because I don't feel like there's 1413 00:57:46,890 --> 00:57:48,690 any way you could know that you're fully 1414 00:57:48,690 --> 00:57:50,080 accounting for the task. 1415 00:57:50,080 --> 00:57:51,870 And so I think those correlations may 1416 00:57:51,870 --> 00:57:55,780 be largely reflecting regions commonly activated by the task, 1417 00:57:55,780 --> 00:57:59,200 and that's why I didn't put that in this lecture. 1418 00:57:59,200 --> 00:58:02,040 But surely, task will also produce correlations, right? 1419 00:58:02,040 --> 00:58:04,770 Let me just put it another way. 1420 00:58:04,770 --> 00:58:08,160 If I flash up a bunch of faces versus-- 1421 00:58:08,160 --> 00:58:10,560 it's like faces versus nothing-- 1422 00:58:10,560 --> 00:58:12,900 and then we look at the correlations 1423 00:58:12,900 --> 00:58:15,180 during that period, well, you'll find correlations 1424 00:58:15,180 --> 00:58:18,510 between V1 and the FFA because when their face is on, 1425 00:58:18,510 --> 00:58:20,550 both V1 and the FFA turn on. 1426 00:58:20,550 --> 00:58:23,170 And when there aren't, they both turn off. 1427 00:58:23,170 --> 00:58:25,960 That's just a task response, right? 1428 00:58:25,960 --> 00:58:30,790 So to be able to look at how these endogenous correlations 1429 00:58:30,790 --> 00:58:32,440 are affected by task, we would have 1430 00:58:32,440 --> 00:58:35,380 to be certain we could siphon off the entire task effect 1431 00:58:35,380 --> 00:58:37,320 so that we could look at just the residual. 1432 00:58:37,320 --> 00:58:38,943 And I don't think any of our analysis 1433 00:58:38,943 --> 00:58:41,110 are good enough to siphon off an entire task effect. 1434 00:58:41,110 --> 00:58:43,060 And that's why I just don't go there with PPI, 1435 00:58:43,060 --> 00:58:44,635 even though everyone else does. 1436 00:58:44,635 --> 00:58:46,510 If you didn't follow that, it doesn't matter. 1437 00:58:46,510 --> 00:58:48,128 I'm just trying to give you an answer. 1438 00:58:48,128 --> 00:58:50,170 I'm going to take just questions of clarification 1439 00:58:50,170 --> 00:58:52,753 now because there's a couple of things I really want to get to 1440 00:58:52,753 --> 00:58:55,750 and I'm running out of time. 1441 00:58:55,750 --> 00:58:58,600 OK. 1442 00:58:58,600 --> 00:59:00,700 But everybody should understand this-- 1443 00:59:00,700 --> 00:59:04,810 an activation map that's made by asking, which brain regions are 1444 00:59:04,810 --> 00:59:09,670 correlated at rest with a given region I choose a seed region 1445 00:59:09,670 --> 00:59:10,450 I choose? 1446 00:59:10,450 --> 00:59:13,340 OK. 1447 00:59:13,340 --> 00:59:16,820 OK, important caveat-- even though people 1448 00:59:16,820 --> 00:59:20,340 call this "resting functional connectivity," 1449 00:59:20,340 --> 00:59:22,580 we will not be using that phrase in this class 1450 00:59:22,580 --> 00:59:24,950 because we do not know that it's connectivity 1451 00:59:24,950 --> 00:59:26,450 in the structural sense. 1452 00:59:26,450 --> 00:59:29,450 It's just a correlation, OK? 1453 00:59:29,450 --> 00:59:31,140 And I'll say more about that later. 1454 00:59:31,140 --> 00:59:33,620 But if you read about resting functional connectivity, 1455 00:59:33,620 --> 00:59:34,602 it's the same thing. 1456 00:59:34,602 --> 00:59:36,560 It's just, I think, people are making a mistake 1457 00:59:36,560 --> 00:59:38,390 using that word. 1458 00:59:38,390 --> 00:59:42,480 OK, so let me get this idea across here. 1459 00:59:42,480 --> 00:59:44,660 You may have heard of the default-mode network. 1460 00:59:44,660 --> 00:59:46,490 There's heaps of papers on this. 1461 00:59:46,490 --> 00:59:47,240 It's a thing. 1462 00:59:47,240 --> 00:59:48,800 There's a lot of discussion of it. 1463 00:59:48,800 --> 00:59:49,940 And it's bizarre. 1464 00:59:49,940 --> 00:59:52,920 It has arisen from two independent findings, 1465 00:59:52,920 --> 00:59:53,420 OK? 1466 00:59:53,420 --> 00:59:55,410 So let's do these findings one at a time. 1467 00:59:55,410 --> 00:59:58,100 The first one is people started noticing around 15 years 1468 00:59:58,100 --> 01:00:00,770 ago that across lots of different kinds of tasks, 1469 01:00:00,770 --> 01:00:05,240 if you looked at not the intended direction, like, say, 1470 01:00:05,240 --> 01:00:07,714 reading sentences versus staring at a dot, 1471 01:00:07,714 --> 01:00:12,350 or doing a demanding working-memory task 1472 01:00:12,350 --> 01:00:16,430 versus a really passive-viewing task, anything where there's 1473 01:00:16,430 --> 01:00:19,550 a really engaging task versus an easy task, 1474 01:00:19,550 --> 01:00:21,500 you would find a bunch of regions 1475 01:00:21,500 --> 01:00:24,650 that were activated in the reverse contrast, regions that 1476 01:00:24,650 --> 01:00:27,535 are more activated when you're doing less mental activity, 1477 01:00:27,535 --> 01:00:29,660 typically, regions that are active when you're just 1478 01:00:29,660 --> 01:00:34,010 lying there at rest compared to doing something difficult. 1479 01:00:34,010 --> 01:00:37,240 And so originally, people were like, what's up with that? 1480 01:00:37,240 --> 01:00:37,990 How can that be? 1481 01:00:37,990 --> 01:00:41,600 It seemed paradoxical, impossible. 1482 01:00:41,600 --> 01:00:44,410 But, in fact, it's not impossible, right? 1483 01:00:44,410 --> 01:00:47,740 Suppose I had you do a bunch of mental-arithmetic tasks, 1484 01:00:47,740 --> 01:00:48,970 and they're pretty demanding. 1485 01:00:48,970 --> 01:00:51,400 And I compared that to just having you lie there 1486 01:00:51,400 --> 01:00:52,660 in the scanner doing nothing. 1487 01:00:52,660 --> 01:00:56,980 It's like, OK, do mental arithmetic for 20 seconds, rest 1488 01:00:56,980 --> 01:00:59,260 20 seconds, mental arithmetic for 20 seconds, 1489 01:00:59,260 --> 01:01:01,150 rest for 20 seconds. 1490 01:01:01,150 --> 01:01:04,000 Now imagine we find parts of your brain, systematic ones, 1491 01:01:04,000 --> 01:01:06,840 that are more engaged at rest. 1492 01:01:06,840 --> 01:01:07,890 What might that mean? 1493 01:01:12,340 --> 01:01:12,945 David. 1494 01:01:12,945 --> 01:01:15,070 AUDIENCE: That part of the brain could be, I think, 1495 01:01:15,070 --> 01:01:16,480 like daydreaming. 1496 01:01:16,480 --> 01:01:18,440 NANCY KANWISHER: Daydreaming, exactly. 1497 01:01:18,440 --> 01:01:18,940 Yeah. 1498 01:01:18,940 --> 01:01:20,260 You can't turn your brain off. 1499 01:01:20,260 --> 01:01:22,450 You don't turn your brain off at rest. 1500 01:01:22,450 --> 01:01:23,710 You daydream. 1501 01:01:23,710 --> 01:01:25,120 Absolutely. 1502 01:01:25,120 --> 01:01:25,630 What else? 1503 01:01:25,630 --> 01:01:26,830 What are the things-- some of the things 1504 01:01:26,830 --> 01:01:27,872 you do when you daydream? 1505 01:01:30,275 --> 01:01:32,150 What are the typical contents of daydreaming? 1506 01:01:32,150 --> 01:01:34,775 I guess it depends who it is and what you're daydreaming about, 1507 01:01:34,775 --> 01:01:36,860 but there are very systematic things people do. 1508 01:01:36,860 --> 01:01:39,320 They recall episodic memories. 1509 01:01:39,320 --> 01:01:42,230 It's like, oh, yeah, before I got in here-- you replay 1510 01:01:42,230 --> 01:01:44,110 things that were happening. 1511 01:01:44,110 --> 01:01:45,110 And what else do you do? 1512 01:01:45,110 --> 01:01:46,590 You think about people? 1513 01:01:46,590 --> 01:01:47,090 Why? 1514 01:01:47,090 --> 01:01:48,530 Because we're social primates, and that's 1515 01:01:48,530 --> 01:01:49,790 what we care a lot about. 1516 01:01:49,790 --> 01:01:51,927 You don't think only about people. 1517 01:01:51,927 --> 01:01:54,260 Some of you guys might be trying to solve a math problem 1518 01:01:54,260 --> 01:01:56,390 that you couldn't solve before. 1519 01:01:56,390 --> 01:01:58,880 But most people in the scan are asked to do nothing, 1520 01:01:58,880 --> 01:02:02,300 are recalling events, which usually involve people, 1521 01:02:02,300 --> 01:02:05,360 or thinking about people, OK? 1522 01:02:05,360 --> 01:02:07,930 So this whole suite of brain regions that was called 1523 01:02:07,930 --> 01:02:10,430 the "default-mode network" is just the regions that are more 1524 01:02:10,430 --> 01:02:13,863 engaged when nobody tells you what to do than for a whole 1525 01:02:13,863 --> 01:02:15,780 bunch of things when they tell you what to do. 1526 01:02:15,780 --> 01:02:19,273 And so it's some weird mix of daydreaming and other stuff. 1527 01:02:19,273 --> 01:02:20,690 And the interesting thing about it 1528 01:02:20,690 --> 01:02:22,550 is they're reasonably systematic. 1529 01:02:22,550 --> 01:02:25,362 So those are the-- 1530 01:02:25,362 --> 01:02:26,570 I keep getting confused here. 1531 01:02:26,570 --> 01:02:26,840 Hang on. 1532 01:02:26,840 --> 01:02:27,757 Let me get this right. 1533 01:02:30,390 --> 01:02:33,390 Did I label this backwards? 1534 01:02:33,390 --> 01:02:35,310 All right, they're the green guys. 1535 01:02:35,310 --> 01:02:38,580 Yeah, deactivated during demanding tasks. 1536 01:02:38,580 --> 01:02:39,145 Yeah. 1537 01:02:39,145 --> 01:02:40,770 There's too many negatives for me here. 1538 01:02:40,770 --> 01:02:45,880 The green guys here-- does that look familiar, that patch? 1539 01:02:45,880 --> 01:02:51,020 What does that look like kind of, not exactly but kind of? 1540 01:02:51,020 --> 01:02:51,520 Sorry? 1541 01:02:51,520 --> 01:02:53,832 AUDIENCE: Visual, the visual system. 1542 01:02:53,832 --> 01:02:56,290 NANCY KANWISHER: It's sort of near the visual system, yeah. 1543 01:02:56,290 --> 01:02:58,207 It is, but it's also like something else we've 1544 01:02:58,207 --> 01:02:59,860 been talking about recently. 1545 01:02:59,860 --> 01:03:01,690 Our TPT. 1546 01:03:01,690 --> 01:03:04,480 It's a little further back, but it's right in the same region, 1547 01:03:04,480 --> 01:03:07,240 OK? 1548 01:03:07,240 --> 01:03:09,145 And here are these medial regions. 1549 01:03:09,145 --> 01:03:11,020 There's a medial view of the left hemisphere, 1550 01:03:11,020 --> 01:03:13,120 like, "take my right hemisphere out and look at the inside" 1551 01:03:13,120 --> 01:03:13,620 view. 1552 01:03:13,620 --> 01:03:15,490 That's this-- [INAUDIBLE] and sulcus, right? 1553 01:03:15,490 --> 01:03:17,750 All these medial regions. 1554 01:03:17,750 --> 01:03:20,420 It looks a whole lot like the social-cognition network that I 1555 01:03:20,420 --> 01:03:23,840 talked about last time, that you identify with the contrast 1556 01:03:23,840 --> 01:03:26,840 of belief task versus-- 1557 01:03:26,840 --> 01:03:29,920 the false-belief test versus a false-photo test. 1558 01:03:29,920 --> 01:03:32,140 So that's weird finding number one, 1559 01:03:32,140 --> 01:03:34,030 that there's a systematic set of regions 1560 01:03:34,030 --> 01:03:35,950 that are engaged at rest. 1561 01:03:35,950 --> 01:03:38,110 They're called the default-mode network 1562 01:03:38,110 --> 01:03:41,260 because they're what you do by default when nobody's 1563 01:03:41,260 --> 01:03:42,708 controlling you externally. 1564 01:03:42,708 --> 01:03:45,250 So that's finding number one, and there's finding number two. 1565 01:03:45,250 --> 01:03:46,640 But first, Jack, did you have a question? 1566 01:03:46,640 --> 01:03:47,265 AUDIENCE: Yeah. 1567 01:03:47,265 --> 01:03:48,910 I was just wondering, does deactivated 1568 01:03:48,910 --> 01:03:50,920 during demanding tasks necessarily 1569 01:03:50,920 --> 01:03:53,297 imply that it is activated during not-demanding tasks? 1570 01:03:53,297 --> 01:03:55,630 NANCY KANWISHER: We're not distinguishing between those. 1571 01:03:55,630 --> 01:03:57,340 We're just taking those two conditions. 1572 01:03:57,340 --> 01:03:59,260 You'd have to have some third baseline 1573 01:03:59,260 --> 01:04:01,310 to figure out whether those two were different. 1574 01:04:01,310 --> 01:04:03,310 And that's very problematic because we're are we 1575 01:04:03,310 --> 01:04:05,727 having a problem saying, what counts as a baseline, right? 1576 01:04:05,727 --> 01:04:09,140 So we'll just compare those two, OK? 1577 01:04:09,140 --> 01:04:10,838 So that's weird finding number one. 1578 01:04:10,838 --> 01:04:13,130 And it's not that weird when you think about it further 1579 01:04:13,130 --> 01:04:14,713 because, of course, you're doing stuff 1580 01:04:14,713 --> 01:04:16,790 when you're lying there, right? 1581 01:04:16,790 --> 01:04:19,550 But the further finding that really put the default mode 1582 01:04:19,550 --> 01:04:21,500 network on the map was when people 1583 01:04:21,500 --> 01:04:26,390 started putting seeds in parts of that default-mode network 1584 01:04:26,390 --> 01:04:32,170 up here and finding that they got the whole rest 1585 01:04:32,170 --> 01:04:35,480 of the network at rest, OK? 1586 01:04:35,480 --> 01:04:38,120 So all of those things are correlated at rest. 1587 01:04:38,120 --> 01:04:40,580 It's not just that they're all activated at rest. 1588 01:04:40,580 --> 01:04:43,400 Their time courses are correlated at rest. 1589 01:04:43,400 --> 01:04:46,640 So, actually, what people mean by default mode network now 1590 01:04:46,640 --> 01:04:50,420 is not, I took the reverse contrast and all the stuff that 1591 01:04:50,420 --> 01:04:53,540 activated more for rest than task, I call that default mode. 1592 01:04:53,540 --> 01:04:56,843 Actually, what they mean is I stuck a seed in there 1593 01:04:56,843 --> 01:04:58,760 during my rest scans, and I took all the stuff 1594 01:04:58,760 --> 01:05:01,250 that was correlated with that position 1595 01:05:01,250 --> 01:05:04,490 because those pick out, more or less, the same thing. 1596 01:05:04,490 --> 01:05:05,990 OK? 1597 01:05:05,990 --> 01:05:08,300 So that's led to a whole lot of discussion 1598 01:05:08,300 --> 01:05:11,060 about what the default mode network is, and what it means, 1599 01:05:11,060 --> 01:05:13,530 and what we can learn from it. 1600 01:05:13,530 --> 01:05:16,615 That's what all this says. 1601 01:05:16,615 --> 01:05:17,990 I'm just trying to figure out how 1602 01:05:17,990 --> 01:05:19,580 I'm going to do this because I'm going to run out of time. 1603 01:05:19,580 --> 01:05:20,830 Maybe I won't run out of time. 1604 01:05:20,830 --> 01:05:22,940 We'll just go for it. 1605 01:05:22,940 --> 01:05:23,600 OK. 1606 01:05:23,600 --> 01:05:28,040 So people started messing around with these correlations 1607 01:05:28,040 --> 01:05:28,850 at rest. 1608 01:05:28,850 --> 01:05:30,350 And they found that you could find 1609 01:05:30,350 --> 01:05:34,280 other systematic sets of regions if you stuck seeds elsewhere. 1610 01:05:34,280 --> 01:05:37,460 And so another systematic region, a set of regions, 1611 01:05:37,460 --> 01:05:41,850 is all the hot-color ones, the yellow and red ones here. 1612 01:05:41,850 --> 01:05:43,880 And so if you look in there, you see 1613 01:05:43,880 --> 01:05:46,130 various things-- the interparietal sulcus, 1614 01:05:46,130 --> 01:05:50,150 a bunch of frontal regions, Visual-Motion area, MT, 1615 01:05:50,150 --> 01:05:52,970 or other visual regions down there. 1616 01:05:52,970 --> 01:05:55,010 And they found that set of regions 1617 01:05:55,010 --> 01:05:57,560 was strongly correlated with each other at rest. 1618 01:05:57,560 --> 01:06:01,070 You stick a seed in here, and you get all that yellow stuff, 1619 01:06:01,070 --> 01:06:03,160 OK? 1620 01:06:03,160 --> 01:06:06,520 And then they looked at it, and they said, yeah, right, we've 1621 01:06:06,520 --> 01:06:07,840 seen those regions engage. 1622 01:06:07,840 --> 01:06:12,160 Whenever people do demanding-- potentially demanding tasks-- 1623 01:06:12,160 --> 01:06:14,720 they've seen that before in other task contrasts. 1624 01:06:14,720 --> 01:06:17,142 So all those things that turn on when you really 1625 01:06:17,142 --> 01:06:18,850 have to pay a lot of attention and you're 1626 01:06:18,850 --> 01:06:20,200 doing a really hard task-- 1627 01:06:20,200 --> 01:06:23,830 all those regions do that, and they're also correlated 1628 01:06:23,830 --> 01:06:25,840 with each other at rest. 1629 01:06:25,840 --> 01:06:26,830 OK? 1630 01:06:26,830 --> 01:06:29,330 And so there's this convergence of these two different lines 1631 01:06:29,330 --> 01:06:30,040 of work-- 1632 01:06:30,040 --> 01:06:32,350 task contrasts that just say, what 1633 01:06:32,350 --> 01:06:35,440 makes a given set of regions turn on or off, 1634 01:06:35,440 --> 01:06:39,220 and correlations-- which things are correlated at rest? 1635 01:06:39,220 --> 01:06:40,090 OK? 1636 01:06:40,090 --> 01:06:42,310 And so they're both converging here with these two 1637 01:06:42,310 --> 01:06:43,810 different networks. 1638 01:06:43,810 --> 01:06:45,640 OK, so I need to do a little sidebar 1639 01:06:45,640 --> 01:06:50,200 on this other hot-color network, not the default-mode network 1640 01:06:50,200 --> 01:06:51,220 but this other one. 1641 01:06:51,220 --> 01:06:53,290 It was originally called "task-positive" 1642 01:06:53,290 --> 01:06:55,930 because it turns on more when you do tasks than rest. 1643 01:06:55,930 --> 01:06:58,660 I mean, that's a really vague statement, OK? 1644 01:06:58,660 --> 01:07:00,820 But it also has lots of other names, 1645 01:07:00,820 --> 01:07:03,220 and the name that we're going to refer to here 1646 01:07:03,220 --> 01:07:06,400 is the multiple-demand regions, OK? 1647 01:07:06,400 --> 01:07:09,190 Multiple demand comes out from another line of work 1648 01:07:09,190 --> 01:07:10,480 that just converged with us. 1649 01:07:10,480 --> 01:07:13,063 They're picking out pretty much the same set of brain regions. 1650 01:07:13,063 --> 01:07:16,360 But "multiple demand" means lots of different kinds 1651 01:07:16,360 --> 01:07:20,210 of cognitive demand activate those same regions. 1652 01:07:20,210 --> 01:07:20,710 OK? 1653 01:07:20,710 --> 01:07:24,040 So I can give you a difficult spatial working-memory task. 1654 01:07:24,040 --> 01:07:26,500 I can give you a difficult perceptual 1655 01:07:26,500 --> 01:07:30,610 orientation-judgment task, a difficult arithmetic task. 1656 01:07:30,610 --> 01:07:32,410 In each of those cases, I can compare them 1657 01:07:32,410 --> 01:07:35,060 to an easy version of the same task, 1658 01:07:35,060 --> 01:07:38,500 and I'll get, more or less, those regions there, 1659 01:07:38,500 --> 01:07:39,740 which are-- 1660 01:07:39,740 --> 01:07:41,380 it's getting a little vague here, 1661 01:07:41,380 --> 01:07:45,880 but they're pretty similar to the task-positive ones, OK? 1662 01:07:45,880 --> 01:07:51,070 So this is both interesting and scandalous. 1663 01:07:51,070 --> 01:07:54,460 It's scandalous because-- to me only, not to anyone else-- 1664 01:07:54,460 --> 01:07:57,100 because unlike all the regions we've been talking about so far 1665 01:07:57,100 --> 01:07:58,960 that have these very specific functions-- 1666 01:07:58,960 --> 01:08:02,950 they just do face recognition or just theory of mind-- 1667 01:08:02,950 --> 01:08:07,660 these ones will do anything, almost, anything difficult, OK? 1668 01:08:07,660 --> 01:08:09,987 So whenever you engage in a difficult task-- 1669 01:08:09,987 --> 01:08:11,570 I'm skipping over a whole literature-- 1670 01:08:11,570 --> 01:08:12,862 it's a big literature on this-- 1671 01:08:12,862 --> 01:08:15,130 but lots and lots of totally different kinds 1672 01:08:15,130 --> 01:08:17,080 of tasks that have nothing in common other 1673 01:08:17,080 --> 01:08:20,640 than they're very demanding-- you engage those regions. 1674 01:08:20,640 --> 01:08:23,100 And in some ways, that's an even more fascinating puzzle. 1675 01:08:23,100 --> 01:08:25,439 Like, what the hell would those operations be? 1676 01:08:25,439 --> 01:08:28,560 What is in common between spatial working memory 1677 01:08:28,560 --> 01:08:31,793 and arithmetic and line-orientation judgment 1678 01:08:31,793 --> 01:08:33,210 and all the other things that have 1679 01:08:33,210 --> 01:08:35,970 been shown to activate these regions when they're demanding? 1680 01:08:35,970 --> 01:08:36,660 Nobody knows. 1681 01:08:36,660 --> 01:08:38,609 I think it's a big, fascinating puzzle. 1682 01:08:38,609 --> 01:08:40,350 Someday we'll have a computational story 1683 01:08:40,350 --> 01:08:42,430 about what's actually computed in those regions, 1684 01:08:42,430 --> 01:08:45,930 but we don't yet, OK? 1685 01:08:45,930 --> 01:08:48,180 There's a lot of stuff on the multiple-demand regions, 1686 01:08:48,180 --> 01:08:49,180 and they're interesting. 1687 01:08:49,180 --> 01:08:51,990 But I can't resist one little thing. 1688 01:08:51,990 --> 01:08:53,340 All right, I can't. 1689 01:08:53,340 --> 01:08:55,050 I have no self-control because I'm not-- 1690 01:08:55,050 --> 01:08:57,000 I don't have enough multiple-demand activity 1691 01:08:57,000 --> 01:08:58,950 right now, so I'm going to have to just tell you these things 1692 01:08:58,950 --> 01:08:59,825 because they're cool. 1693 01:08:59,825 --> 01:09:04,380 OK, so a guy named John Duncan has spent the last 15 1694 01:09:04,380 --> 01:09:06,600 years arguing that the multiple-demand regions first 1695 01:09:06,600 --> 01:09:08,505 are really truly multiple demand. 1696 01:09:08,505 --> 01:09:10,380 He's tested lots and lots of different tasks, 1697 01:09:10,380 --> 01:09:13,620 and they're very, very domain general. 1698 01:09:13,620 --> 01:09:15,450 But second, he thinks they're implicated 1699 01:09:15,450 --> 01:09:17,520 in fluid intelligence. 1700 01:09:17,520 --> 01:09:20,670 Fluid intelligence differs from crystallized intelligence. 1701 01:09:20,670 --> 01:09:24,279 Crystallized intelligence is stuff like your vocabulary, 1702 01:09:24,279 --> 01:09:26,609 just stuff you've learned and cached away, 1703 01:09:26,609 --> 01:09:29,189 and facts you've stored, and abilities you've-- specific 1704 01:09:29,189 --> 01:09:30,540 abilities you've stored. 1705 01:09:30,540 --> 01:09:32,520 Fluid intelligence you measure with stuff 1706 01:09:32,520 --> 01:09:35,365 like Raven's matrices, where nothing you know 1707 01:09:35,365 --> 01:09:36,490 is going to help you do it. 1708 01:09:36,490 --> 01:09:39,060 You just have to be smart and see some abstract pattern 1709 01:09:39,060 --> 01:09:41,170 or something like that. 1710 01:09:41,170 --> 01:09:44,010 And so Duncan thinks that these regions are 1711 01:09:44,010 --> 01:09:46,450 related to fluid intelligence. 1712 01:09:46,450 --> 01:09:49,200 And one of his measures of that is if you find people 1713 01:09:49,200 --> 01:09:50,170 with brain damage-- 1714 01:09:50,170 --> 01:09:51,953 he had a big set of around 80 people 1715 01:09:51,953 --> 01:09:53,370 who had brain damage who he'd been 1716 01:09:53,370 --> 01:09:56,580 studying in all different parts of the brain. 1717 01:09:56,580 --> 01:10:00,540 And what he found is if you have brain damage in those regions, 1718 01:10:00,540 --> 01:10:04,470 your IQ goes down as a result of the damage in proportion 1719 01:10:04,470 --> 01:10:10,980 to the amount of cortical volume destroyed by the damage. 1720 01:10:10,980 --> 01:10:13,140 If you have damage anywhere else in the brain, 1721 01:10:13,140 --> 01:10:15,360 your IQ is unaffected. 1722 01:10:15,360 --> 01:10:19,200 You may become paralyzed, or aphasic, or prosopagnosic, 1723 01:10:19,200 --> 01:10:21,180 or akinetopsic. 1724 01:10:21,180 --> 01:10:23,370 You may have any of these very specific deficits 1725 01:10:23,370 --> 01:10:27,090 according to where it lands, but it won't affect your IQ. 1726 01:10:27,090 --> 01:10:29,670 And so the picture here is that in addition 1727 01:10:29,670 --> 01:10:32,400 to all these special-purpose processors that this course has 1728 01:10:32,400 --> 01:10:34,590 been focusing on, we have this thing that's 1729 01:10:34,590 --> 01:10:39,030 kind of like the brain's CPU or something like that. 1730 01:10:39,030 --> 01:10:42,150 And it seems to live in approximately those regions, 1731 01:10:42,150 --> 01:10:44,070 and it seems to under-- 1732 01:10:44,070 --> 01:10:48,475 it seems to be essential for fluid intelligence, OK? 1733 01:10:48,475 --> 01:10:50,100 So I'm skipping over lots of literature 1734 01:10:50,100 --> 01:10:53,100 just to heighten that this is a particularly interesting set 1735 01:10:53,100 --> 01:10:55,320 of regions here, OK? 1736 01:10:55,320 --> 01:10:56,310 Yeah. 1737 01:10:56,310 --> 01:10:58,598 AUDIENCE: Do people study novelty? 1738 01:10:58,598 --> 01:10:59,515 NANCY KANWISHER: Yeah. 1739 01:10:59,515 --> 01:11:01,750 AUDIENCE: The regions that specialize in novelty. 1740 01:11:01,750 --> 01:11:05,800 NANCY KANWISHER: These guys will be interested in novelty. 1741 01:11:05,800 --> 01:11:08,170 These guys will be interested in novelty but not only. 1742 01:11:08,170 --> 01:11:11,320 You can do the same boring but difficult task on, and on, 1743 01:11:11,320 --> 01:11:14,880 and on, and they'll keep going. 1744 01:11:14,880 --> 01:11:16,290 OK. 1745 01:11:16,290 --> 01:11:18,270 The reason I went on that sidebar 1746 01:11:18,270 --> 01:11:20,640 is that you can identify these regions, not just 1747 01:11:20,640 --> 01:11:23,400 by scanning people while they're doing difficult tasks 1748 01:11:23,400 --> 01:11:25,620 but by sticking a seed in any of those regions 1749 01:11:25,620 --> 01:11:27,480 and getting the others, OK? 1750 01:11:27,480 --> 01:11:31,085 That's the task-positive network pretty much, OK? 1751 01:11:31,085 --> 01:11:32,460 So we're getting this convergence 1752 01:11:32,460 --> 01:11:35,750 between sets of brain regions that we 1753 01:11:35,750 --> 01:11:38,450 find with a task contrast and sets of brain regions 1754 01:11:38,450 --> 01:11:40,990 we find by finding what's correlated with what. 1755 01:11:40,990 --> 01:11:42,740 And the bigger picture of this whole thing 1756 01:11:42,740 --> 01:11:45,590 is that I've been focusing on individual regions 1757 01:11:45,590 --> 01:11:48,440 and what they do, and the gist of this whole resting 1758 01:11:48,440 --> 01:11:49,880 functional-correlation literature 1759 01:11:49,880 --> 01:11:53,300 is a very relevant level of organization 1760 01:11:53,300 --> 01:11:56,510 of the brain is not just an individual cortical region 1761 01:11:56,510 --> 01:12:01,897 but a set of cortical regions that seem to be in cahoots. 1762 01:12:01,897 --> 01:12:03,980 And, again, we don't know what that means exactly, 1763 01:12:03,980 --> 01:12:06,272 but they're correlated at rest, and they have something 1764 01:12:06,272 --> 01:12:08,030 to do with each other, OK? 1765 01:12:08,030 --> 01:12:10,340 So we're finding this higher-level organization, 1766 01:12:10,340 --> 01:12:13,040 and the multiple-demand system is part of it. 1767 01:12:13,040 --> 01:12:16,943 OK, so how are we going to look at this? 1768 01:12:16,943 --> 01:12:18,110 So there's a bunch of these. 1769 01:12:18,110 --> 01:12:21,855 I've talked only about the default mode network and-- 1770 01:12:21,855 --> 01:12:24,230 it's another name for the same thing-- executive control. 1771 01:12:24,230 --> 01:12:25,115 Don't worry about it. 1772 01:12:25,115 --> 01:12:26,657 You think of that as multiple demand. 1773 01:12:26,657 --> 01:12:28,100 It doesn't matter. 1774 01:12:28,100 --> 01:12:30,950 But there's a bunch of them that you can find by sticking seeds 1775 01:12:30,950 --> 01:12:32,580 in different places. 1776 01:12:32,580 --> 01:12:34,850 And so yes, I just said the big idea here 1777 01:12:34,850 --> 01:12:37,870 is that networks are an interesting level-- 1778 01:12:37,870 --> 01:12:39,530 an interesting kind of unit in thinking 1779 01:12:39,530 --> 01:12:41,477 about brain organization-- bigger 1780 01:12:41,477 --> 01:12:42,560 than an individual region. 1781 01:12:42,560 --> 01:12:47,460 It's a set of regions that have something in common, OK? 1782 01:12:47,460 --> 01:12:51,567 But I've sort of backed into in this awkward way of saying, 1783 01:12:51,567 --> 01:12:53,900 OK, here are things that are correlated with each other, 1784 01:12:53,900 --> 01:12:55,817 and here's what we know about the same regions 1785 01:12:55,817 --> 01:12:57,970 from previous task analysis. 1786 01:12:57,970 --> 01:13:00,340 Most of the literature on resting functional correlation 1787 01:13:00,340 --> 01:13:03,437 just looks at correlations and doesn't try to put it together 1788 01:13:03,437 --> 01:13:05,020 with what we know about those regions, 1789 01:13:05,020 --> 01:13:08,950 and that just seems deeply weird to me. 1790 01:13:08,950 --> 01:13:10,930 So for years, I ignored this whole thing 1791 01:13:10,930 --> 01:13:12,430 because it's like, I don't know what 1792 01:13:12,430 --> 01:13:13,900 these resting correlations are. 1793 01:13:13,900 --> 01:13:15,430 And if I don't know what they are, 1794 01:13:15,430 --> 01:13:17,020 I'm not going to work on them. 1795 01:13:17,020 --> 01:13:19,990 And then Idan Blank came along. 1796 01:13:19,990 --> 01:13:22,540 And when he was a first-year grad student at Fedorenko, 1797 01:13:22,540 --> 01:13:24,748 said, hey, we have all these resting-functional data. 1798 01:13:24,748 --> 01:13:27,340 Let's have Idan, for his rotation, for a month, 1799 01:13:27,340 --> 01:13:28,870 analyze the resting-functional data. 1800 01:13:28,870 --> 01:13:30,040 I said, resting functional-- we don't 1801 01:13:30,040 --> 01:13:30,640 know what the hell it means. 1802 01:13:30,640 --> 01:13:31,360 Let's not bother. 1803 01:13:31,360 --> 01:13:33,610 She's like, get over yourself. 1804 01:13:33,610 --> 01:13:35,803 Let's let him play with it. 1805 01:13:35,803 --> 01:13:37,720 Well, thank God she's not as stuck in her ways 1806 01:13:37,720 --> 01:13:40,570 as I am because Idan spent just a month playing 1807 01:13:40,570 --> 01:13:43,690 with some of our data, and what he found blew me away. 1808 01:13:43,690 --> 01:13:45,520 So here's what he did. 1809 01:13:45,520 --> 01:13:49,570 He said, OK, let's start with actually identified regions 1810 01:13:49,570 --> 01:13:52,390 of brain where we know something about what they do, 1811 01:13:52,390 --> 01:13:56,020 like the language system and the multiple-demand system. 1812 01:13:56,020 --> 01:13:57,730 OK? 1813 01:13:57,730 --> 01:14:00,070 Let's identify those regions in each subject 1814 01:14:00,070 --> 01:14:08,252 individually, and then let's scan subjects at rest, OK? 1815 01:14:08,252 --> 01:14:11,480 First, you scan subjects with sentences versus nonwords. 1816 01:14:11,480 --> 01:14:13,070 You find the language regions. 1817 01:14:13,070 --> 01:14:15,800 Then you scan them with a difficult-versus-easy spatial 1818 01:14:15,800 --> 01:14:16,700 working-memory task. 1819 01:14:16,700 --> 01:14:19,010 You find the multiple-demand regions. 1820 01:14:19,010 --> 01:14:23,870 Then you scan them at rest, and you get the average timecourse 1821 01:14:23,870 --> 01:14:26,180 from each of those regions at rest. 1822 01:14:26,180 --> 01:14:28,830 These are fake data, just to give you the gist. 1823 01:14:28,830 --> 01:14:29,330 Makes sense? 1824 01:14:29,330 --> 01:14:31,010 Are you with me now? 1825 01:14:31,010 --> 01:14:33,200 Now you can ask, OK, which of these things 1826 01:14:33,200 --> 01:14:35,283 are correlated with each other at rest? 1827 01:14:35,283 --> 01:14:36,950 And this is now a more interesting thing 1828 01:14:36,950 --> 01:14:39,590 to do because we're asking this principled question of regions 1829 01:14:39,590 --> 01:14:42,380 we know something about rather than random seeds 1830 01:14:42,380 --> 01:14:45,140 in some random location, right? 1831 01:14:45,140 --> 01:14:48,810 OK, so now what we do is we examine those correlations. 1832 01:14:48,810 --> 01:14:51,290 And we just ask, for example, how correlated 1833 01:14:51,290 --> 01:14:54,950 are those timecourses of two different language regions 1834 01:14:54,950 --> 01:14:59,910 or those two different parts of the multiple-demand system? 1835 01:14:59,910 --> 01:15:03,000 And how strong are the correlations 1836 01:15:03,000 --> 01:15:06,720 between the systems, some little piece of the language system 1837 01:15:06,720 --> 01:15:09,450 and some little piece of the multiple-demand system? 1838 01:15:09,450 --> 01:15:10,950 Makes sense? 1839 01:15:10,950 --> 01:15:15,098 So that's a cool question to ask, and that's what Idan did. 1840 01:15:15,098 --> 01:15:16,140 And here's what he found. 1841 01:15:16,140 --> 01:15:17,710 Let me first orient you. 1842 01:15:17,710 --> 01:15:20,520 So here are lots and lots of regions of interest 1843 01:15:20,520 --> 01:15:22,950 that were identified functionally-- a whole bunch 1844 01:15:22,950 --> 01:15:24,960 of language regions up here, a whole bunch 1845 01:15:24,960 --> 01:15:26,520 of multiple-demand regions down here. 1846 01:15:26,520 --> 01:15:28,950 The details don't matter, OK? 1847 01:15:28,950 --> 01:15:31,320 But what I'm going to show you is in each cell, 1848 01:15:31,320 --> 01:15:34,023 we're going to have a correlation between a given-- 1849 01:15:34,023 --> 01:15:36,690 so this would be a given part of the cell over here-- given part 1850 01:15:36,690 --> 01:15:37,940 of the multiple demand system. 1851 01:15:40,350 --> 01:15:42,480 Sorry, a given part of the multiple-demand system 1852 01:15:42,480 --> 01:15:45,120 and some other part of the multiple-demand system, 1853 01:15:45,120 --> 01:15:49,350 or over here, a cell would be some part of the language 1854 01:15:49,350 --> 01:15:52,980 system and some part of the multiple-demand system, OK? 1855 01:15:52,980 --> 01:15:56,460 So when you do that, here's what you see. 1856 01:15:56,460 --> 01:15:59,700 Here are the correlations at rest between all 1857 01:15:59,700 --> 01:16:02,018 of these pairs of conditions. 1858 01:16:02,018 --> 01:16:03,810 And if you squint a little-- you don't even 1859 01:16:03,810 --> 01:16:05,820 have to squint much-- but the black ones 1860 01:16:05,820 --> 01:16:07,950 are the ones that are not significantly correlated 1861 01:16:07,950 --> 01:16:08,850 at all. 1862 01:16:08,850 --> 01:16:10,380 Blue means a negative correlation, 1863 01:16:10,380 --> 01:16:13,390 and hot colors mean a positive correlation. 1864 01:16:13,390 --> 01:16:17,750 And so what you see is here are all the language regions. 1865 01:16:17,750 --> 01:16:19,750 These are the right-hemisphere language regions, 1866 01:16:19,750 --> 01:16:21,340 which barely even count. 1867 01:16:21,340 --> 01:16:23,600 They're just there for the hell of it. 1868 01:16:23,600 --> 01:16:25,310 This is really the core language regions, 1869 01:16:25,310 --> 01:16:27,010 and you can see they're all correlated 1870 01:16:27,010 --> 01:16:30,880 with each other, even ones that are really far apart-- 1871 01:16:30,880 --> 01:16:32,740 Broca's area and Wernicke's area-- 1872 01:16:32,740 --> 01:16:37,150 10, 12 centimeters apart-- strongly correlated at rest, 1873 01:16:37,150 --> 01:16:38,253 OK? 1874 01:16:38,253 --> 01:16:39,670 And if you look at different parts 1875 01:16:39,670 --> 01:16:42,010 of the multiple-demand system, they're 1876 01:16:42,010 --> 01:16:44,640 all strongly correlated at rest, even 1877 01:16:44,640 --> 01:16:46,390 regions that are far apart-- something way 1878 01:16:46,390 --> 01:16:47,765 up in the frontal lobe, something 1879 01:16:47,765 --> 01:16:50,380 way back in the parietal lobe-- strongly correlated at rest. 1880 01:16:54,010 --> 01:16:54,760 And so yeah. 1881 01:16:54,760 --> 01:16:56,810 If you zoom in here-- 1882 01:16:56,810 --> 01:16:59,150 so what this is really-- 1883 01:16:59,150 --> 01:17:01,970 does everybody see that this is revealing a lot of structure? 1884 01:17:01,970 --> 01:17:02,820 Yeah, question? 1885 01:17:02,820 --> 01:17:05,320 AUDIENCE: So the diagonals would be, like, self correlation, 1886 01:17:05,320 --> 01:17:06,200 right? 1887 01:17:06,200 --> 01:17:08,075 NANCY KANWISHER: Yeah, that's why it's black. 1888 01:17:12,830 --> 01:17:13,910 Actually, what is it? 1889 01:17:13,910 --> 01:17:14,660 Oh, you know what? 1890 01:17:14,660 --> 01:17:15,452 It's split in half. 1891 01:17:15,452 --> 01:17:16,550 It's split in half. 1892 01:17:16,550 --> 01:17:18,050 It's actually a better way to do it. 1893 01:17:18,050 --> 01:17:19,820 You take your data, and you have two different halves 1894 01:17:19,820 --> 01:17:20,480 of the data. 1895 01:17:20,480 --> 01:17:23,420 And so it gives you a baseline for the-- 1896 01:17:23,420 --> 01:17:25,130 no, that doesn't make any sense, does it? 1897 01:17:25,130 --> 01:17:25,790 Never mind. 1898 01:17:25,790 --> 01:17:26,210 AUDIENCE: Yeah, [INAUDIBLE]. 1899 01:17:26,210 --> 01:17:27,225 AUDIENCE: [INAUDIBLE]. 1900 01:17:27,225 --> 01:17:29,105 AUDIENCE: [INAUDIBLE]. 1901 01:17:29,105 --> 01:17:30,355 NANCY KANWISHER: It is maroon. 1902 01:17:32,698 --> 01:17:34,240 AUDIENCE: If you look at the spectrum 1903 01:17:34,240 --> 01:17:36,925 on the bottom, [INAUDIBLE]. 1904 01:17:36,925 --> 01:17:37,550 AUDIENCE: Yeah. 1905 01:17:37,550 --> 01:17:38,120 NANCY KANWISHER: All right, I'm going 1906 01:17:38,120 --> 01:17:40,100 to have to solve this offline because I'm now confused, 1907 01:17:40,100 --> 01:17:42,433 unless Anya can figure it out right now and bail us out. 1908 01:17:42,433 --> 01:17:43,785 AUDIENCE: [INAUDIBLE] 1909 01:17:43,785 --> 01:17:45,910 NANCY KANWISHER: Yeah, but as they're pointing out, 1910 01:17:45,910 --> 01:17:47,380 it's not black. 1911 01:17:47,380 --> 01:17:47,880 OK. 1912 01:17:47,880 --> 01:17:48,020 AUDIENCE: [INAUDIBLE] 1913 01:17:48,020 --> 01:17:49,325 [INTERPOSING VOICES] 1914 01:17:49,325 --> 01:17:50,950 AUDIENCE: I'm saying it should be high. 1915 01:17:50,950 --> 01:17:52,150 I was just wondering [INAUDIBLE].. 1916 01:17:52,150 --> 01:17:53,110 NANCY KANWISHER: Oh, it's a correlation of 1. 1917 01:17:53,110 --> 01:17:53,980 Is that what it is? 1918 01:17:53,980 --> 01:17:55,090 AUDIENCE: Yeah. 1919 01:17:55,090 --> 01:17:56,960 NANCY KANWISHER: OK, because it's correlated with itself. 1920 01:17:56,960 --> 01:17:57,460 All right. 1921 01:17:57,460 --> 01:17:57,960 OK. 1922 01:17:57,960 --> 01:17:59,560 Thank you. 1923 01:17:59,560 --> 01:18:02,140 Anyway, does everybody get the gist 1924 01:18:02,140 --> 01:18:04,120 that all of these different pieces 1925 01:18:04,120 --> 01:18:06,460 of the multiple-demand system that we identified 1926 01:18:06,460 --> 01:18:09,190 individually-- they're all correlated with each other 1927 01:18:09,190 --> 01:18:09,938 at rest? 1928 01:18:09,938 --> 01:18:11,980 All these different pieces of the language system 1929 01:18:11,980 --> 01:18:14,020 are correlated with each other at rest. 1930 01:18:14,020 --> 01:18:16,120 And there's no correlation at all 1931 01:18:16,120 --> 01:18:18,100 between any part of the language system 1932 01:18:18,100 --> 01:18:20,590 and any part of the multiple-demand system at rest. 1933 01:18:20,590 --> 01:18:22,930 All of these-- OK, maybe a couple. 1934 01:18:22,930 --> 01:18:25,930 These cells are all either black for not significant 1935 01:18:25,930 --> 01:18:27,610 or inversely correlated. 1936 01:18:27,610 --> 01:18:29,500 That's the cool colors. 1937 01:18:29,500 --> 01:18:33,550 So do you see how this gives us a totally cool way aside 1938 01:18:33,550 --> 01:18:36,190 from just the functional localizers we 1939 01:18:36,190 --> 01:18:40,960 ran to identify these regions to show us that these things are 1940 01:18:40,960 --> 01:18:43,590 functioning as a system, right? 1941 01:18:43,590 --> 01:18:46,710 It's not just that Broca's area is a cool little thing that 1942 01:18:46,710 --> 01:18:49,830 does a piece of language and some bit of the temporal lobe 1943 01:18:49,830 --> 01:18:52,320 is a cool thing that does some piece of language. 1944 01:18:52,320 --> 01:18:54,900 But those guys are part of a broader system, 1945 01:18:54,900 --> 01:18:56,790 and these resting functional correlations 1946 01:18:56,790 --> 01:18:59,940 are revealing this broader system and the integrity 1947 01:18:59,940 --> 01:19:02,940 of the parts within it, as well as the distinction 1948 01:19:02,940 --> 01:19:06,150 between those parts and parts of another system, or network. 1949 01:19:06,150 --> 01:19:07,830 Does everybody get that idea? 1950 01:19:07,830 --> 01:19:08,490 Good. 1951 01:19:08,490 --> 01:19:11,130 That's a big idea for this lecture. 1952 01:19:11,130 --> 01:19:12,915 I think what I'll do-- 1953 01:19:12,915 --> 01:19:14,535 so you can-- 1954 01:19:14,535 --> 01:19:17,160 I'll skip that. 1955 01:19:17,160 --> 01:19:18,810 So this is basically the correlation 1956 01:19:18,810 --> 01:19:21,840 between all of the cells within the language system, 1957 01:19:21,840 --> 01:19:24,930 all of the cells within the multiple-demand system, 1958 01:19:24,930 --> 01:19:27,310 and any pair of cells between systems here. 1959 01:19:27,310 --> 01:19:31,680 So that's just averaging over the matrix I showed you before. 1960 01:19:31,680 --> 01:19:34,790 And so this is a cool way to ask about broader 1961 01:19:34,790 --> 01:19:36,980 systems in the brain. 1962 01:19:36,980 --> 01:19:39,920 And I was going to show you some data published just a month ago 1963 01:19:39,920 --> 01:19:42,830 that asked this question not just about the language 1964 01:19:42,830 --> 01:19:45,530 and multiple-demand systems but also, 1965 01:19:45,530 --> 01:19:51,488 about the theory-of-mind network, which is basically 1966 01:19:51,488 --> 01:19:53,280 really similar to the default mode network. 1967 01:19:53,280 --> 01:19:54,655 But the theory-of-mind network we 1968 01:19:54,655 --> 01:19:56,620 can identify that we talked about last time. 1969 01:19:56,620 --> 01:19:58,355 And you can go think offline. 1970 01:19:58,355 --> 01:19:59,730 Actually, I'll take a suggestion. 1971 01:19:59,730 --> 01:20:00,640 What do you think? 1972 01:20:00,640 --> 01:20:02,670 Should the theory-of-mind network 1973 01:20:02,670 --> 01:20:05,220 be correlated with the language system, 1974 01:20:05,220 --> 01:20:09,234 with the multiple-demand system, both, or neither? 1975 01:20:09,234 --> 01:20:10,158 AUDIENCE: Neither. 1976 01:20:10,158 --> 01:20:11,200 NANCY KANWISHER: Neither? 1977 01:20:11,200 --> 01:20:11,937 Why? 1978 01:20:11,937 --> 01:20:13,520 AUDIENCE: This is something different. 1979 01:20:13,520 --> 01:20:15,853 NANCY KANWISHER: OK, that's a totally reasonable answer, 1980 01:20:15,853 --> 01:20:18,210 and that's largely true but not 100% true. 1981 01:20:18,210 --> 01:20:19,520 What else might you think? 1982 01:20:19,520 --> 01:20:22,400 It is a different system, and it does function 1983 01:20:22,400 --> 01:20:24,350 quite independently but not perfectly. 1984 01:20:24,350 --> 01:20:26,910 Yeah. 1985 01:20:26,910 --> 01:20:40,700 AUDIENCE: Because it's [INAUDIBLE] analyzing 1986 01:20:40,700 --> 01:20:54,515 [INAUDIBLE] 1987 01:20:54,515 --> 01:20:55,640 NANCY KANWISHER: All right. 1988 01:20:55,640 --> 01:21:00,380 That's a lovely speculation and an intelligent one, 1989 01:21:00,380 --> 01:21:01,700 and it's half true. 1990 01:21:01,700 --> 01:21:03,480 I'll just skip to the data. 1991 01:21:03,480 --> 01:21:08,660 So in this very recently published paper, 1992 01:21:08,660 --> 01:21:10,803 Alex Paunov, and Idan Blank, and Ev Fedorenko 1993 01:21:10,803 --> 01:21:13,220 looked at the language system, the theory-of-mind network, 1994 01:21:13,220 --> 01:21:15,410 which is not just the TPJ but these other regions 1995 01:21:15,410 --> 01:21:17,000 that I mentioned briefly that you also 1996 01:21:17,000 --> 01:21:19,370 get in the contrast of the false-belief test 1997 01:21:19,370 --> 01:21:22,520 versus the false-photo task and the multiple-demand network. 1998 01:21:22,520 --> 01:21:25,940 Same deal-- identify each of those regions in each subject 1999 01:21:25,940 --> 01:21:28,970 individually, then scan the subject at rest 2000 01:21:28,970 --> 01:21:30,980 and see what's correlated with what. 2001 01:21:30,980 --> 01:21:33,200 And here's the answer. 2002 01:21:33,200 --> 01:21:36,410 You see, again, replicate the language system, especially 2003 01:21:36,410 --> 01:21:38,630 in the left hemisphere. 2004 01:21:38,630 --> 01:21:41,330 The theory of mind system is all a system. 2005 01:21:41,330 --> 01:21:45,390 And the multiple-demand system is a system, separate system. 2006 01:21:45,390 --> 01:21:47,780 But if you look in at the cell where 2007 01:21:47,780 --> 01:21:51,740 you have theory of mind and language, 2008 01:21:51,740 --> 01:21:55,760 it's slightly above chance, not theory of mind 2009 01:21:55,760 --> 01:21:58,760 in multiple demand but theory of mind in language, probably 2010 01:21:58,760 --> 01:22:01,862 for just the reason you said, the whole essence of language, 2011 01:22:01,862 --> 01:22:03,320 even though it's a different thing, 2012 01:22:03,320 --> 01:22:05,820 than thinking about the contents of someone else's thoughts. 2013 01:22:05,820 --> 01:22:08,130 They are so enmeshed in each other. 2014 01:22:08,130 --> 01:22:10,440 The reason we have language is to take our thoughts 2015 01:22:10,440 --> 01:22:12,440 and put them in your head and take your thoughts 2016 01:22:12,440 --> 01:22:13,760 and put them in our head. 2017 01:22:13,760 --> 01:22:16,910 And so it makes sense that those things are a little bit 2018 01:22:16,910 --> 01:22:17,530 correlated. 2019 01:22:17,530 --> 01:22:19,703 AUDIENCE: [INAUDIBLE] language? 2020 01:22:19,703 --> 01:22:20,620 NANCY KANWISHER: Yeah. 2021 01:22:20,620 --> 01:22:21,530 Yeah. 2022 01:22:21,530 --> 01:22:23,180 But neither of them is correlated 2023 01:22:23,180 --> 01:22:24,560 with multiple-demand. 2024 01:22:24,560 --> 01:22:26,750 It's 1:26, so-- or-- 2025 01:22:26,750 --> 01:22:27,768 wait. 2026 01:22:27,768 --> 01:22:29,060 Am I reading the wrong-- yeah-- 2027 01:22:29,060 --> 01:22:29,750 oh, 12:28. 2028 01:22:29,750 --> 01:22:31,053 Sorry. 2029 01:22:31,053 --> 01:22:31,970 There's a number here. 2030 01:22:31,970 --> 01:22:33,387 That's how long I've been talking. 2031 01:22:33,387 --> 01:22:35,930 So 12:28-- so if you need to go, that's fine, 2032 01:22:35,930 --> 01:22:37,590 but I'm happy to answer questions. 2033 01:22:37,590 --> 01:22:38,090 So go ahead. 2034 01:22:38,090 --> 01:22:39,650 AUDIENCE: I was just going to say, 2035 01:22:39,650 --> 01:22:42,620 there was a review paper about [INAUDIBLE] talked 2036 01:22:42,620 --> 01:22:46,250 about how there's a behavioral connection between theory 2037 01:22:46,250 --> 01:22:49,282 of mind and, like, when children started to learn [INAUDIBLE].. 2038 01:22:49,282 --> 01:22:51,740 NANCY KANWISHER: Lots of links, especially developmentally, 2039 01:22:51,740 --> 01:22:52,500 yeah. 2040 01:22:52,500 --> 01:22:53,113 Yeah. 2041 01:22:53,113 --> 01:22:53,780 Did you have a-- 2042 01:22:53,780 --> 01:22:55,197 AUDIENCE: By an extension of that, 2043 01:22:55,197 --> 01:23:00,440 [INAUDIBLE] should be correlated to the theory of mind 2044 01:23:00,440 --> 01:23:01,250 [INAUDIBLE]? 2045 01:23:01,250 --> 01:23:03,590 NANCY KANWISHER: Yes, wouldn't you think? 2046 01:23:03,590 --> 01:23:04,430 Not really. 2047 01:23:04,430 --> 01:23:05,835 AUDIENCE: [INAUDIBLE] 2048 01:23:05,835 --> 01:23:07,460 NANCY KANWISHER: The FFA is irritating. 2049 01:23:07,460 --> 01:23:09,293 It's not strongly correlated with the things 2050 01:23:09,293 --> 01:23:11,530 it ought to be correlated with.