1 00:00:00,000 --> 00:00:02,420 [SQUEAKING] 2 00:00:02,420 --> 00:00:03,872 [RUSTLING] 3 00:00:03,872 --> 00:00:07,260 [CLICKING] 4 00:00:10,115 --> 00:00:11,740 NANCY KANWISHER: All right, it's 11:05. 5 00:00:11,740 --> 00:00:12,890 Let's get started. 6 00:00:12,890 --> 00:00:15,820 So the agenda for today, we're doing this whole thing 7 00:00:15,820 --> 00:00:18,310 on the methods in human cognitive neuroscience. 8 00:00:18,310 --> 00:00:19,810 And I'm illustrating those methods 9 00:00:19,810 --> 00:00:21,580 with the case of face perception. 10 00:00:21,580 --> 00:00:23,710 Not just because I'm into face perception, 11 00:00:23,710 --> 00:00:26,710 but it's a particularly rich domain of research 12 00:00:26,710 --> 00:00:28,750 where there's lots to say about it from all 13 00:00:28,750 --> 00:00:30,460 these different methods. 14 00:00:30,460 --> 00:00:31,930 And so last time, we talked a bit 15 00:00:31,930 --> 00:00:34,240 about applying Marr's computational theory 16 00:00:34,240 --> 00:00:35,590 level to face perception. 17 00:00:35,590 --> 00:00:38,110 We talked a teeny bit about some behavioral data 18 00:00:38,110 --> 00:00:40,060 and a little bit about functional MRI. 19 00:00:40,060 --> 00:00:41,890 What I'm going to do today is quickly 20 00:00:41,890 --> 00:00:44,562 zoom through a speeded-up review of those things, 21 00:00:44,562 --> 00:00:47,020 and then we're going to get to some of these other methods. 22 00:00:47,020 --> 00:00:48,640 And there's a quiz at the end. 23 00:00:48,640 --> 00:00:50,020 All right? 24 00:00:50,020 --> 00:00:56,440 OK, so methods in any field of science 25 00:00:56,440 --> 00:00:59,650 are just there to enable us to answer scientific questions. 26 00:00:59,650 --> 00:01:01,870 They're not to impress our friends 27 00:01:01,870 --> 00:01:03,700 with all the fancy things we know how to do 28 00:01:03,700 --> 00:01:05,470 or our colleagues. 29 00:01:05,470 --> 00:01:06,925 They're just to answer questions. 30 00:01:06,925 --> 00:01:09,050 And so you always have to start with the questions. 31 00:01:09,050 --> 00:01:11,800 And so last time, I listed a bunch of questions. 32 00:01:11,800 --> 00:01:13,540 Not all of them, but a bunch of questions 33 00:01:13,540 --> 00:01:16,000 one would really want to know about face perception 34 00:01:16,000 --> 00:01:19,690 if we were to understand how it works in the brain. 35 00:01:19,690 --> 00:01:22,300 And last time, we focused on these first three. 36 00:01:22,300 --> 00:01:24,580 So let me just do a super quick review. 37 00:01:24,580 --> 00:01:27,100 The questions at the level of Marr's computational theory, 38 00:01:27,100 --> 00:01:29,830 we ask, what is the problem that's being solved 39 00:01:29,830 --> 00:01:32,020 and why is that important to the organism? 40 00:01:32,020 --> 00:01:34,030 What is the input, what is the output? 41 00:01:34,030 --> 00:01:37,120 How much you get from that input to that output, right? 42 00:01:37,120 --> 00:01:39,070 So for the case of face perception, 43 00:01:39,070 --> 00:01:40,973 here's a very simple version of it. 44 00:01:40,973 --> 00:01:42,265 Here's an example of the input. 45 00:01:42,265 --> 00:01:44,290 It goes in, hits the retina. 46 00:01:44,290 --> 00:01:46,600 The stuff that we want to understand happens in here, 47 00:01:46,600 --> 00:01:47,830 and you have an output. 48 00:01:47,830 --> 00:01:50,770 OK, so just even thinking about it that way, 49 00:01:50,770 --> 00:01:53,140 we can already just see, with common sense, 50 00:01:53,140 --> 00:01:56,080 that one of the big challenges in solving this problem 51 00:01:56,080 --> 00:01:59,350 is that faces look different every time you see them. 52 00:01:59,350 --> 00:02:01,300 The lighting changes, the orientation 53 00:02:01,300 --> 00:02:04,240 of the face changes, the hair changes, the mood changes, 54 00:02:04,240 --> 00:02:05,350 all this stuff happens. 55 00:02:05,350 --> 00:02:08,539 People put on makeup, they shave off their facial hair, 56 00:02:08,539 --> 00:02:11,470 they do all these things to make it a big challenge 57 00:02:11,470 --> 00:02:12,790 to recognize faces. 58 00:02:12,790 --> 00:02:14,870 And yet, we manage really well. 59 00:02:14,870 --> 00:02:16,910 So how do we do that? 60 00:02:16,910 --> 00:02:20,680 Well, our field has many methods to address this question. 61 00:02:20,680 --> 00:02:23,320 Last time, I talked about one little example 62 00:02:23,320 --> 00:02:26,740 of a behavioral study-- simple, cognitive psychology 63 00:02:26,740 --> 00:02:31,660 study measuring behavior-- where we showed that the way 64 00:02:31,660 --> 00:02:33,670 people solve this problem is fundamentally 65 00:02:33,670 --> 00:02:35,620 different with people they know well 66 00:02:35,620 --> 00:02:37,300 and people they don't know well. 67 00:02:37,300 --> 00:02:40,390 So I showed an example that all of you 68 00:02:40,390 --> 00:02:42,910 presumably would have no trouble determining 69 00:02:42,910 --> 00:02:45,130 that those are all pictures of the same person, 70 00:02:45,130 --> 00:02:48,670 even though at the pixel level they're wildly different. 71 00:02:48,670 --> 00:02:51,130 And yet, you have a hell of a time saying which of those 72 00:02:51,130 --> 00:02:53,810 images are of the same people and which aren't. 73 00:02:53,810 --> 00:02:56,050 And so the point is that our ability 74 00:02:56,050 --> 00:02:58,750 to extract this invariant representation, that 75 00:02:58,750 --> 00:03:02,800 is to figure out abstractly who is that, 76 00:03:02,800 --> 00:03:06,190 is really-- well, to figure out that any of these images 77 00:03:06,190 --> 00:03:08,800 are the same as each other is much better for familiar 78 00:03:08,800 --> 00:03:10,120 than unfamiliar faces. 79 00:03:10,120 --> 00:03:13,060 And that means we don't have a perfectly general ability 80 00:03:13,060 --> 00:03:18,310 to take any face and abstract out this completely 81 00:03:18,310 --> 00:03:20,150 image-independent version of it. 82 00:03:20,150 --> 00:03:21,940 That's what invariant representation is. 83 00:03:21,940 --> 00:03:23,040 Yeah? 84 00:03:23,040 --> 00:03:25,390 AUDIENCE: For the case of the Dutch politicians, 85 00:03:25,390 --> 00:03:27,610 did they ever do the study on people 86 00:03:27,610 --> 00:03:30,132 who were super recognizers? 87 00:03:30,132 --> 00:03:31,840 NANCY KANWISHER: I don't know about that, 88 00:03:31,840 --> 00:03:35,770 but they did do it on people who are 89 00:03:35,770 --> 00:03:40,120 professional TSA-type people. 90 00:03:40,120 --> 00:03:40,840 AUDIENCE: OK. 91 00:03:40,840 --> 00:03:41,080 NANCY KANWISHER: Right? 92 00:03:41,080 --> 00:03:42,747 And I'll tell you guys about that later. 93 00:03:42,747 --> 00:03:45,040 But you could think about whether you 94 00:03:45,040 --> 00:03:49,360 think it might work better with those people or not. 95 00:03:49,360 --> 00:03:51,460 OK, everybody get this general point here? 96 00:03:51,460 --> 00:03:53,200 All right. 97 00:03:53,200 --> 00:03:56,350 So I skipped over another simple behavioral finding 98 00:03:56,350 --> 00:03:58,720 last time that I want to mention now. 99 00:03:58,720 --> 00:04:01,540 And that is an extremely low tech-- 100 00:04:01,540 --> 00:04:05,170 charmingly low tech, and yet, I think very powerful-- discovery 101 00:04:05,170 --> 00:04:06,350 about face perception. 102 00:04:06,350 --> 00:04:09,310 One of the most important original bits 103 00:04:09,310 --> 00:04:11,110 of evidence that face perception might 104 00:04:11,110 --> 00:04:14,530 be a different thing in the brain came from a PhD thesis 105 00:04:14,530 --> 00:04:17,050 in this department by a guy named Robert Yin. 106 00:04:17,050 --> 00:04:20,620 And he used the extremely high tech equipment 107 00:04:20,620 --> 00:04:24,340 of a stopwatch and paper. 108 00:04:24,340 --> 00:04:26,060 OK, so what did he do? 109 00:04:26,060 --> 00:04:28,450 He presented faces to people upright. 110 00:04:28,450 --> 00:04:30,820 And he said, study these 20 faces. 111 00:04:30,820 --> 00:04:32,183 And then he tested them later. 112 00:04:32,183 --> 00:04:33,100 Did you see this face? 113 00:04:33,100 --> 00:04:34,017 Did you see this face? 114 00:04:34,017 --> 00:04:35,680 Did you see this face? 115 00:04:35,680 --> 00:04:37,750 And then he did the exact same experiment 116 00:04:37,750 --> 00:04:40,750 on a different set of faces, but they were all upside down. 117 00:04:40,750 --> 00:04:44,350 Studied upside down and tested upside down. 118 00:04:44,350 --> 00:04:45,950 And what did he find? 119 00:04:45,950 --> 00:04:49,000 He found what's known as the face inversion effect. 120 00:04:49,000 --> 00:04:52,030 Namely, people do much worse at this task 121 00:04:52,030 --> 00:04:54,040 when the faces are upside down. 122 00:04:54,040 --> 00:04:56,380 Here's errors for inverted upside down, 123 00:04:56,380 --> 00:04:58,600 errors for upright at this task. 124 00:04:58,600 --> 00:05:01,660 Even though, importantly, they were studied and tested 125 00:05:01,660 --> 00:05:05,830 upside down or studied and tested inverted-- 126 00:05:05,830 --> 00:05:07,390 upright. 127 00:05:07,390 --> 00:05:09,630 OK, everybody got what this shows? 128 00:05:09,630 --> 00:05:14,013 OK, so that's cool that this face inversion-- 129 00:05:14,013 --> 00:05:16,680 but the further cool thing is he showed that this face inversion 130 00:05:16,680 --> 00:05:21,190 effect is greater for faces than for other kinds of stimuli. 131 00:05:21,190 --> 00:05:22,900 So he tested lots of other things, 132 00:05:22,900 --> 00:05:25,800 including houses and stick figures. 133 00:05:25,800 --> 00:05:27,990 And he showed that that cost, when 134 00:05:27,990 --> 00:05:29,670 you turn the stimuli upside down, 135 00:05:29,670 --> 00:05:31,470 is greater, that difference is greater 136 00:05:31,470 --> 00:05:35,340 for faces than for other classes of stimuli. 137 00:05:35,340 --> 00:05:40,380 So what that suggests is that face recognition may just 138 00:05:40,380 --> 00:05:43,980 work differently in some deep way from recognition 139 00:05:43,980 --> 00:05:46,410 of other classes of stimuli. 140 00:05:46,410 --> 00:05:50,070 And Robert Yin actually inferred in his PhD thesis-- 141 00:05:50,070 --> 00:05:52,740 way, way back before any imaging method-- 142 00:05:52,740 --> 00:05:54,360 that maybe there are special parts 143 00:05:54,360 --> 00:05:56,340 of the brain for face recognition. 144 00:05:56,340 --> 00:05:59,040 And maybe face recognition is just a totally different thing, 145 00:05:59,040 --> 00:06:01,500 that's why it is more affected by inversion 146 00:06:01,500 --> 00:06:03,550 than recognition of other kinds of things. 147 00:06:03,550 --> 00:06:05,040 Was there a question back there? 148 00:06:05,040 --> 00:06:05,745 Yeah. 149 00:06:05,745 --> 00:06:07,920 AUDIENCE: I was going to ask, could that just be 150 00:06:07,920 --> 00:06:11,670 because faces are much more complex than houses or stick 151 00:06:11,670 --> 00:06:12,638 figures and that-- 152 00:06:12,638 --> 00:06:13,930 NANCY KANWISHER: Good question. 153 00:06:13,930 --> 00:06:14,390 Hang on to-- 154 00:06:14,390 --> 00:06:14,790 AUDIENCE: --backwards. 155 00:06:14,790 --> 00:06:16,082 NANCY KANWISHER: Good question. 156 00:06:16,082 --> 00:06:17,460 That's a very good question. 157 00:06:17,460 --> 00:06:20,730 And many people have tried to grapple with that. 158 00:06:20,730 --> 00:06:23,040 And actually, about 10 years ago, 159 00:06:23,040 --> 00:06:25,440 the idea that this disproportionate effect 160 00:06:25,440 --> 00:06:30,570 for faces was standard textbook, completely accepted. 161 00:06:30,570 --> 00:06:33,480 And now there's another round of people doubting it 162 00:06:33,480 --> 00:06:34,720 with other kinds of stimuli. 163 00:06:34,720 --> 00:06:35,800 So it's kind of ongoing. 164 00:06:35,800 --> 00:06:39,420 It's a very robust difference, but to say exactly what it 165 00:06:39,420 --> 00:06:42,822 is about face stimuli versus other kinds of things that 166 00:06:42,822 --> 00:06:44,280 is responsible for that difference, 167 00:06:44,280 --> 00:06:46,890 you can imagine it's subtle. 168 00:06:46,890 --> 00:06:49,560 For the purposes of this course, I'm 169 00:06:49,560 --> 00:06:51,120 trying to not quite lie to you guys, 170 00:06:51,120 --> 00:06:54,210 but give you the most standard view without freighting you 171 00:06:54,210 --> 00:06:57,210 with every possible objection to every little thing. 172 00:06:57,210 --> 00:06:59,010 Because pretty much every finding, 173 00:06:59,010 --> 00:07:01,385 there's somebody who has a beef with it. 174 00:07:01,385 --> 00:07:02,760 We'll tell you, that's not really 175 00:07:02,760 --> 00:07:03,870 true because blah-di-blah. 176 00:07:03,870 --> 00:07:04,380 OK? 177 00:07:04,380 --> 00:07:07,825 So yes, there's a little bit of debate going on 178 00:07:07,825 --> 00:07:08,700 about this right now. 179 00:07:08,700 --> 00:07:10,330 But for the purposes of this course, 180 00:07:10,330 --> 00:07:15,270 it's pretty damn rock solid, at least as an empirical result. 181 00:07:15,270 --> 00:07:16,590 All right. 182 00:07:16,590 --> 00:07:20,020 So there's in fact lots of versions of the face inversion 183 00:07:20,020 --> 00:07:20,520 effect. 184 00:07:20,520 --> 00:07:23,400 One you may have seen before but which is very amusing. 185 00:07:23,400 --> 00:07:26,010 If you look at faces like this that are upside down, 186 00:07:26,010 --> 00:07:28,890 they look sort of normal. 187 00:07:28,890 --> 00:07:31,350 But then if you rotate them, you realize 188 00:07:31,350 --> 00:07:34,235 there's something deeply weird going on. 189 00:07:34,235 --> 00:07:35,610 So the point is, you're much more 190 00:07:35,610 --> 00:07:39,090 sensitive to those grotesquely distorted faces 191 00:07:39,090 --> 00:07:41,340 when you see them right side up than when you see them 192 00:07:41,340 --> 00:07:42,040 upside down. 193 00:07:42,040 --> 00:07:44,650 So that's another version of the face inversion effect, 194 00:07:44,650 --> 00:07:47,130 and there are many, many incarnations of this effect. 195 00:07:47,130 --> 00:07:50,910 You'll see another one later in the lecture. 196 00:07:50,910 --> 00:07:53,580 So where did we get last time with these questions? 197 00:07:53,580 --> 00:07:57,600 We got that one of the major, if not the major central challenge 198 00:07:57,600 --> 00:08:00,000 in face recognition at a computational level, 199 00:08:00,000 --> 00:08:03,390 is the fact that we deal with huge image variation 200 00:08:03,390 --> 00:08:04,530 each time we see a face. 201 00:08:04,530 --> 00:08:06,880 And yet, somehow we're able to grapple with it. 202 00:08:06,880 --> 00:08:08,920 So to understand how face recognition works will 203 00:08:08,920 --> 00:08:11,340 be to understand, what is the code, 204 00:08:11,340 --> 00:08:13,017 ultimately-- nobody knows right now-- 205 00:08:13,017 --> 00:08:14,850 but what is a code running in our heads that 206 00:08:14,850 --> 00:08:16,050 enables us to do that? 207 00:08:16,050 --> 00:08:17,610 What is our mental representation 208 00:08:17,610 --> 00:08:21,960 of a face that enables us to deal with this problem? 209 00:08:21,960 --> 00:08:25,170 By looking at behavioral data, we got some evidence 210 00:08:25,170 --> 00:08:28,140 from the Dutch politician study that whatever 211 00:08:28,140 --> 00:08:30,940 that representation is that we extract from faces, 212 00:08:30,940 --> 00:08:33,929 it's not independent of the particular image. 213 00:08:33,929 --> 00:08:37,350 It's not that we have some platonic ideal of the face 214 00:08:37,350 --> 00:08:39,000 that we can extract from any face 215 00:08:39,000 --> 00:08:42,960 that lands on our retina, platonic ideal of that person's 216 00:08:42,960 --> 00:08:43,770 face, right? 217 00:08:43,770 --> 00:08:47,430 So whatever we're doing, it's not completely invariant, 218 00:08:47,430 --> 00:08:54,000 because we can't do that so well with unfamiliar faces. 219 00:08:54,000 --> 00:08:56,280 Also, as I just showed you-- related, 220 00:08:56,280 --> 00:08:58,210 but not exactly the same point-- 221 00:08:58,210 --> 00:08:59,910 our mental representations of faces 222 00:08:59,910 --> 00:09:02,700 are very sensitive to the orientation of the face 223 00:09:02,700 --> 00:09:04,410 more than our mental representations 224 00:09:04,410 --> 00:09:07,500 of other classes of stimuli. 225 00:09:07,500 --> 00:09:12,300 So those are just very simple insights about whatever 226 00:09:12,300 --> 00:09:14,550 our representations of faces are in our heads, 227 00:09:14,550 --> 00:09:18,150 just from simple behavioral data. 228 00:09:18,150 --> 00:09:21,960 OK, so let me just review some of the strengths and weaknesses 229 00:09:21,960 --> 00:09:23,730 of simple behavioral methods. 230 00:09:23,730 --> 00:09:25,680 Strengths are, they're good for characterizing 231 00:09:25,680 --> 00:09:27,510 the internal representation, right? 232 00:09:27,510 --> 00:09:30,360 Not with huge computational precision, 233 00:09:30,360 --> 00:09:33,295 they're more like with gisty kind of ideas. 234 00:09:33,295 --> 00:09:34,920 They're not very invariant, they depend 235 00:09:34,920 --> 00:09:36,030 on the orientation, right? 236 00:09:36,030 --> 00:09:38,340 That's not very precise, but it's a whole lot better 237 00:09:38,340 --> 00:09:41,280 than nothing. 238 00:09:41,280 --> 00:09:44,280 That's what I mean by at least qualitatively. 239 00:09:44,280 --> 00:09:46,890 They're good for disassociating mental phenomena. 240 00:09:46,890 --> 00:09:50,040 So you've already seen that when the inversion effect, 241 00:09:50,040 --> 00:09:52,210 it happens more for faces than other things. 242 00:09:52,210 --> 00:09:54,853 So that already starts to tell us, OK, maybe 243 00:09:54,853 --> 00:09:56,520 whatever the code in our head is that we 244 00:09:56,520 --> 00:09:58,260 use for face recognition, maybe it's 245 00:09:58,260 --> 00:10:00,990 pretty different than the code that we use in our head 246 00:10:00,990 --> 00:10:01,950 to recognize objects. 247 00:10:05,610 --> 00:10:07,560 OK, it's also cheap. 248 00:10:07,560 --> 00:10:08,640 It's really cheap. 249 00:10:08,640 --> 00:10:11,650 Much cheaper than all the other methods. 250 00:10:11,650 --> 00:10:15,310 OK, weaknesses-- behavioral methods 251 00:10:15,310 --> 00:10:17,860 alone don't have any relationship to the brain, 252 00:10:17,860 --> 00:10:20,690 at least without doing extra work. 253 00:10:20,690 --> 00:10:24,370 And it's not that they're useless until you link them 254 00:10:24,370 --> 00:10:26,080 to the, brain it's just that the brain is 255 00:10:26,080 --> 00:10:27,370 a whole source of other data. 256 00:10:27,370 --> 00:10:29,230 And it's nice to link them, because then you 257 00:10:29,230 --> 00:10:33,040 can connect with all those other data. 258 00:10:33,040 --> 00:10:35,530 Also, behavioral data are pretty sparse. 259 00:10:35,530 --> 00:10:38,830 For the most part, you have accuracy and reaction time, 260 00:10:38,830 --> 00:10:40,010 and that's it. 261 00:10:40,010 --> 00:10:42,670 And that's just not a whole lot of data to work with. 262 00:10:42,670 --> 00:10:44,440 You have to actually be much smarter 263 00:10:44,440 --> 00:10:47,300 to be a behavioral cognitive psychologist, 264 00:10:47,300 --> 00:10:51,070 than you have to be a cognitive neuroscientist, 265 00:10:51,070 --> 00:10:53,410 where you have much richer data to reason from. 266 00:10:53,410 --> 00:10:57,130 Cognitive psychologists really have very, very clever designs 267 00:10:57,130 --> 00:11:00,040 because they're taking this extremely limited data 268 00:11:00,040 --> 00:11:02,890 and trying to pull out interesting insights 269 00:11:02,890 --> 00:11:04,820 about mental function. 270 00:11:04,820 --> 00:11:07,240 Another way of looking at that is, 271 00:11:07,240 --> 00:11:10,840 here's an eyeball and a bunch of processing going over stages 272 00:11:10,840 --> 00:11:12,850 and a response, right? 273 00:11:12,850 --> 00:11:16,360 With behavioral data, all you have is that response. 274 00:11:16,360 --> 00:11:19,425 But presumably, for most of the mental processes that 275 00:11:19,425 --> 00:11:21,550 go on in our heads, there are many different stages 276 00:11:21,550 --> 00:11:23,860 of processing where different things are going on. 277 00:11:23,860 --> 00:11:26,110 Computations tend to have multiple stages 278 00:11:26,110 --> 00:11:27,520 and unfold over time. 279 00:11:27,520 --> 00:11:30,650 And all we have is the output. 280 00:11:30,650 --> 00:11:33,040 So really, what we want to be able to do 281 00:11:33,040 --> 00:11:36,010 is characterize the whole sequence of processes. 282 00:11:36,010 --> 00:11:37,995 And it's not that you can't get insights 283 00:11:37,995 --> 00:11:39,370 about some of those intermediates 284 00:11:39,370 --> 00:11:42,590 from behavioral data, it's just much more challenging. 285 00:11:42,590 --> 00:11:45,610 So if we had a way to look at those things independently, 286 00:11:45,610 --> 00:11:49,200 wouldn't that be awesome? 287 00:11:49,200 --> 00:11:51,570 OK, so there's lots of ways to do that. 288 00:11:51,570 --> 00:11:54,160 And a particularly good one is functional MRI. 289 00:11:54,160 --> 00:11:55,980 So as I mentioned before-- 290 00:11:55,980 --> 00:11:59,400 I mentioned this very briefly-- 291 00:11:59,400 --> 00:12:03,510 this very early experiment that I did way back asking 292 00:12:03,510 --> 00:12:05,970 whether there is a region of the brain that's selectively 293 00:12:05,970 --> 00:12:08,183 involved in processing faces. 294 00:12:08,183 --> 00:12:10,350 And I'm going to put a slightly different spin on it 295 00:12:10,350 --> 00:12:11,308 from what I put before. 296 00:12:11,308 --> 00:12:12,820 It's the same experiment, same data, 297 00:12:12,820 --> 00:12:15,930 but I want to emphasize more the logic 298 00:12:15,930 --> 00:12:18,030 of the experimental design because you guys will 299 00:12:18,030 --> 00:12:23,970 be designing an experiment on a different topic due Monday 300 00:12:23,970 --> 00:12:24,900 night. 301 00:12:24,900 --> 00:12:27,780 That we're going to discuss Sunday night-- that we're going 302 00:12:27,780 --> 00:12:30,240 to discuss in class on Monday. 303 00:12:30,240 --> 00:12:34,140 So we start with a hypothesis that there's 304 00:12:34,140 --> 00:12:36,600 a region of the brain that's selectively 305 00:12:36,600 --> 00:12:37,560 responsive to faces. 306 00:12:37,560 --> 00:12:38,940 That's the hypothesis. 307 00:12:38,940 --> 00:12:42,270 The way we test it is to pop people in a scanner 308 00:12:42,270 --> 00:12:44,140 and show them faces and objects. 309 00:12:44,140 --> 00:12:45,660 The data that I showed you before 310 00:12:45,660 --> 00:12:47,550 is that this little patch of the brain-- 311 00:12:47,550 --> 00:12:49,980 remember, this is a horizontal slice, back of the head, 312 00:12:49,980 --> 00:12:51,370 left and right are flipped. 313 00:12:51,370 --> 00:12:54,240 So that little region in me is right about in there. 314 00:12:54,240 --> 00:12:55,590 Everybody oriented? 315 00:12:55,590 --> 00:12:56,730 OK. 316 00:12:56,730 --> 00:13:00,720 That region responds much more to faces than objects. 317 00:13:00,720 --> 00:13:02,460 Is that clear to everybody what that is? 318 00:13:02,460 --> 00:13:03,510 OK. 319 00:13:03,510 --> 00:13:07,470 So yes, you see that in most subjects. 320 00:13:07,470 --> 00:13:09,780 So yes, there's a bit that responds more to faces 321 00:13:09,780 --> 00:13:11,520 than objects. 322 00:13:11,520 --> 00:13:14,040 But now, let's consider the hypothesis 323 00:13:14,040 --> 00:13:17,400 that that region is really selective to faces per se. 324 00:13:17,400 --> 00:13:20,550 And the way you evaluate whether these data, 325 00:13:20,550 --> 00:13:23,308 how strongly these data support that hypothesis-- 326 00:13:23,308 --> 00:13:24,850 they're certainly consistent with it, 327 00:13:24,850 --> 00:13:27,660 but do they nail that hypothesis fully-- 328 00:13:27,660 --> 00:13:30,870 is to consider, are there any other alternative 329 00:13:30,870 --> 00:13:34,200 accounts we can think of that are consistent with these data 330 00:13:34,200 --> 00:13:36,600 and different from that hypothesis? 331 00:13:36,600 --> 00:13:37,650 Is that clear? 332 00:13:37,650 --> 00:13:38,580 It's really important. 333 00:13:38,580 --> 00:13:42,600 That's just the whole kernel of scientific thinking 334 00:13:42,600 --> 00:13:45,420 and evaluating evidence is asking yourself that question. 335 00:13:45,420 --> 00:13:48,090 Is there any other way we could get those data where 336 00:13:48,090 --> 00:13:50,130 that hypothesis wasn't true? 337 00:13:50,130 --> 00:13:52,540 And if so, you've got to grapple with it. 338 00:13:52,540 --> 00:13:57,000 So what you do next is you think up alternative hypotheses 339 00:13:57,000 --> 00:13:59,220 to the one you started with, that 340 00:13:59,220 --> 00:14:01,690 is different accounts of the same data. 341 00:14:01,690 --> 00:14:04,560 And so in our case, you guys suggested a whole bunch, 342 00:14:04,560 --> 00:14:06,120 I suggested a bunch. 343 00:14:06,120 --> 00:14:08,130 And then the next thing I showed you 344 00:14:08,130 --> 00:14:10,740 is that we can test those alternative hypotheses, 345 00:14:10,740 --> 00:14:15,022 at least these ones here, by first-- 346 00:14:15,022 --> 00:14:16,980 what we did was, I didn't really emphasize this 347 00:14:16,980 --> 00:14:19,380 before-- but we reran that experiment 348 00:14:19,380 --> 00:14:22,380 in a new bunch of subjects, each subject individually. 349 00:14:22,380 --> 00:14:25,900 We found in each subject the little bit that does this. 350 00:14:25,900 --> 00:14:30,090 We write down exactly where that is in that person's brain. 351 00:14:30,090 --> 00:14:32,730 Now that we found that region-- that's called a localizer 352 00:14:32,730 --> 00:14:35,070 run, because we're finding that region in each subject 353 00:14:35,070 --> 00:14:39,060 individually-- now we can ask it new questions. 354 00:14:39,060 --> 00:14:41,940 And so the new questions we asked it last time 355 00:14:41,940 --> 00:14:44,280 was to present faces and hands. 356 00:14:44,280 --> 00:14:48,970 And we found, oh, that region right there responds like this. 357 00:14:48,970 --> 00:14:52,810 So the key ideas here is that we can identify 358 00:14:52,810 --> 00:14:54,670 that region in each subject individually 359 00:14:54,670 --> 00:14:56,210 with a functional scan. 360 00:14:56,210 --> 00:14:58,840 The reason that's important-- which I'll carry on about 361 00:14:58,840 --> 00:15:00,220 in more detail later-- 362 00:15:00,220 --> 00:15:03,670 is that the exact location of that region varies from one 363 00:15:03,670 --> 00:15:05,470 subject to the next. 364 00:15:05,470 --> 00:15:08,350 So if we just grab the whole fusiform 365 00:15:08,350 --> 00:15:11,410 gyrus or the whole lateral side of the fusiform 366 00:15:11,410 --> 00:15:14,560 gyrus in each subject, we'll get lots of stuff 367 00:15:14,560 --> 00:15:16,930 that is that region and lots of cortical neighbors 368 00:15:16,930 --> 00:15:19,120 that's something else. 369 00:15:19,120 --> 00:15:22,240 And if we took the exact location 370 00:15:22,240 --> 00:15:24,280 of that region in my brain and registered it 371 00:15:24,280 --> 00:15:25,780 to any of your brains and said, OK, 372 00:15:25,780 --> 00:15:27,155 let's take the part of your brain 373 00:15:27,155 --> 00:15:29,890 that registers spatially as well as we can with mine, 374 00:15:29,890 --> 00:15:32,720 we're not going to exactly get the right bit. 375 00:15:32,720 --> 00:15:34,540 So to study that thing, we've got 376 00:15:34,540 --> 00:15:36,040 to first find it functionally. 377 00:15:36,040 --> 00:15:38,390 And then we can ask it new questions. 378 00:15:38,390 --> 00:15:40,250 Does that make sense? 379 00:15:40,250 --> 00:15:41,860 OK, if anybody's unclear about that, 380 00:15:41,860 --> 00:15:43,720 I have actually online talks that 381 00:15:43,720 --> 00:15:45,970 go through the whole logic of this in painful detail. 382 00:15:45,970 --> 00:15:49,210 And I'm happy to answer other questions about it later. 383 00:15:49,210 --> 00:15:52,600 OK, so I put the word conditions in red 384 00:15:52,600 --> 00:15:55,900 because somebody asked one of the TAs what a condition was. 385 00:15:55,900 --> 00:15:59,060 And that's not stupid, I should have made that clear. 386 00:15:59,060 --> 00:16:04,630 This is just experimental design gobbledygook that means any-- 387 00:16:07,330 --> 00:16:09,100 OK, what is the definition of condition? 388 00:16:09,100 --> 00:16:11,560 In an experimental design, you have 389 00:16:11,560 --> 00:16:14,950 things that you are manipulating and measuring. 390 00:16:14,950 --> 00:16:18,040 So in this case, we're manipulating the stimulus. 391 00:16:18,040 --> 00:16:20,350 And we're measuring the magnitude of response 392 00:16:20,350 --> 00:16:24,860 in the fusiform face area or in the brain. 393 00:16:24,860 --> 00:16:26,830 So what we're manipulating, in this case, 394 00:16:26,830 --> 00:16:28,990 is the stimulus condition. 395 00:16:28,990 --> 00:16:30,610 So that would be one condition, that's 396 00:16:30,610 --> 00:16:33,070 another condition, that's another condition. 397 00:16:33,070 --> 00:16:34,580 Does that make sense? 398 00:16:34,580 --> 00:16:39,370 OK, so for your experimental design assignment for Monday 399 00:16:39,370 --> 00:16:43,300 night, you will be designing one or more experiments. 400 00:16:43,300 --> 00:16:46,000 And you will be describing exactly what conditions you are 401 00:16:46,000 --> 00:16:47,260 going to test. 402 00:16:47,260 --> 00:16:48,670 Everybody clear on that? 403 00:16:48,670 --> 00:16:50,110 OK. 404 00:16:50,110 --> 00:16:53,020 All right, so these data enable us 405 00:16:53,020 --> 00:16:55,720 to rule out those hypotheses. 406 00:16:55,720 --> 00:16:58,780 And now what you could ask, OK, once you 407 00:16:58,780 --> 00:17:00,790 get more data like this, have you completely 408 00:17:00,790 --> 00:17:02,080 nailed that hypothesis? 409 00:17:02,080 --> 00:17:04,720 Is there just no way that hypothesis could be wrong now 410 00:17:04,720 --> 00:17:06,640 given these data and those data? 411 00:17:06,640 --> 00:17:08,349 And I'll let you percolate on that. 412 00:17:08,349 --> 00:17:10,329 There are ways it could be wrong, 413 00:17:10,329 --> 00:17:12,790 but you have to work harder to come up with them. 414 00:17:15,430 --> 00:17:19,900 OK, so skipping ahead, just to give you the gist. 415 00:17:19,900 --> 00:17:22,810 This field has been going on for a long time. 416 00:17:22,810 --> 00:17:24,700 And there are now many, many studies-- 417 00:17:24,700 --> 00:17:28,240 100 maybe even, I don't know, God, maybe even thousands, 418 00:17:28,240 --> 00:17:32,230 I don't know-- studies of this region in which-- 419 00:17:32,230 --> 00:17:35,140 and so this is sort of a summary statement from a long time ago. 420 00:17:35,140 --> 00:17:37,630 In my lab, we've tested the response of this region 421 00:17:37,630 --> 00:17:39,670 to lots of different kinds of stimuli. 422 00:17:39,670 --> 00:17:41,770 With that same method, localize it 423 00:17:41,770 --> 00:17:44,350 in each subject, measure its response when people 424 00:17:44,350 --> 00:17:47,500 look at that kind of stimulus, 425 00:17:47,500 --> 00:17:49,780 And so what we know now was that this region 426 00:17:49,780 --> 00:17:53,560 is found in roughly the same location in pretty 427 00:17:53,560 --> 00:17:55,360 much every normal subject. 428 00:17:55,360 --> 00:17:58,510 It responds more to faces than to any other kind of stimuli 429 00:17:58,510 --> 00:18:00,520 anyone has ever tested. 430 00:18:00,520 --> 00:18:03,020 Let me just give you one example here. 431 00:18:03,020 --> 00:18:05,208 If you haven't seen this stimulus before, 432 00:18:05,208 --> 00:18:07,000 raise your hand if you can tell what it is. 433 00:18:10,525 --> 00:18:12,400 Raise your hand if you can tell what that is. 434 00:18:14,970 --> 00:18:16,860 OK, some of you didn't quite get it yet. 435 00:18:16,860 --> 00:18:17,890 If you don't see it, don't worry. 436 00:18:17,890 --> 00:18:18,660 There's nothing wrong with you. 437 00:18:18,660 --> 00:18:19,830 It's a little subtle. 438 00:18:19,830 --> 00:18:22,590 It's a face in profile, eyes, nose, mouth. 439 00:18:22,590 --> 00:18:24,240 Everyone got it? 440 00:18:24,240 --> 00:18:25,350 OK, so here's the thing. 441 00:18:25,350 --> 00:18:28,200 That's the same stimulus, it's just upside down. 442 00:18:28,200 --> 00:18:30,420 Another version of the face inversion effect. 443 00:18:30,420 --> 00:18:33,630 In this case, you can't even make yourself see the face 444 00:18:33,630 --> 00:18:34,890 when it's upside down. 445 00:18:34,890 --> 00:18:37,505 If you think you see the upside down version of the face, 446 00:18:37,505 --> 00:18:38,880 you probably have the wrong bits. 447 00:18:38,880 --> 00:18:41,340 The thing you think is a nose probably isn't, et cetera. 448 00:18:41,340 --> 00:18:44,070 OK, so this is an extreme version of the face inversion 449 00:18:44,070 --> 00:18:44,700 effect. 450 00:18:44,700 --> 00:18:47,940 And it's a gift to an experimental psychologist. 451 00:18:47,940 --> 00:18:49,680 Why is that such a gift? 452 00:18:49,680 --> 00:18:51,780 Because it's the same damn stimulus. 453 00:18:51,780 --> 00:18:54,690 But in one case you see a face, in another case you don't. 454 00:18:54,690 --> 00:18:56,670 All we did was tip it upside down. 455 00:18:56,670 --> 00:19:00,090 And the response of the fusiform face area is much stronger 456 00:19:00,090 --> 00:19:02,610 to the upright version when you see the face than 457 00:19:02,610 --> 00:19:05,170 to the inverted version when you don't. 458 00:19:05,170 --> 00:19:09,480 So that enables us to stifle a whole line of attack 459 00:19:09,480 --> 00:19:12,330 from all of these hard core vision people who 460 00:19:12,330 --> 00:19:14,790 early on said, Kanwisher, your face area 461 00:19:14,790 --> 00:19:16,350 isn't really selective for faces. 462 00:19:16,350 --> 00:19:20,790 It's selective for these spatial frequencies or those, 463 00:19:20,790 --> 00:19:23,370 that kind of contrast, or this kind of shading information. 464 00:19:23,370 --> 00:19:25,170 It's like, no, same stimulus. 465 00:19:25,170 --> 00:19:26,087 It's just upside down. 466 00:19:26,087 --> 00:19:27,253 It makes all the difference. 467 00:19:27,253 --> 00:19:29,100 It's really whether you see a face or not. 468 00:19:29,100 --> 00:19:29,370 Yeah? 469 00:19:29,370 --> 00:19:30,745 AUDIENCE: When you were measuring 470 00:19:30,745 --> 00:19:34,380 the response of that example, did 471 00:19:34,380 --> 00:19:37,260 you have it so that at first, when people that the first time 472 00:19:37,260 --> 00:19:39,702 didn't recognize it and then you told them? 473 00:19:39,702 --> 00:19:41,160 NANCY KANWISHER: We did that later. 474 00:19:41,160 --> 00:19:43,350 Not in this experiment, but we did that later. 475 00:19:43,350 --> 00:19:46,830 AUDIENCE: Are they looked see like what changed [INAUDIBLE]?? 476 00:19:46,830 --> 00:19:49,110 NANCY KANWISHER: OK, so it's a great question. 477 00:19:49,110 --> 00:19:50,400 And there's a lot you could do with that. 478 00:19:50,400 --> 00:19:52,170 And actually, I think other people have published 479 00:19:52,170 --> 00:19:53,310 studies like that since. 480 00:19:53,310 --> 00:19:56,890 I can't quite remember who all has done it. 481 00:19:56,890 --> 00:19:59,040 But what we did was most of our subjects, 482 00:19:59,040 --> 00:20:01,110 especially in the context of a whole experiment, 483 00:20:01,110 --> 00:20:04,230 we chose stimuli so that most people could see the face 484 00:20:04,230 --> 00:20:07,380 in most of the upright stimuli and most people could not 485 00:20:07,380 --> 00:20:09,780 see the face in most of the inverted stimuli. 486 00:20:09,780 --> 00:20:11,370 It wasn't perfect at all. 487 00:20:11,370 --> 00:20:13,650 They didn't see faces in all of the upright ones 488 00:20:13,650 --> 00:20:16,530 and they didn't fail to see them in all of the inverted ones. 489 00:20:16,530 --> 00:20:18,960 And that's probably why this difference in response 490 00:20:18,960 --> 00:20:22,122 is not 2 to 1, but it's close. 491 00:20:22,122 --> 00:20:24,330 But you could do lots of other experiments like that, 492 00:20:24,330 --> 00:20:25,950 and you should think about what kinds of designs 493 00:20:25,950 --> 00:20:27,990 would be good ones to do and what it would 494 00:20:27,990 --> 00:20:29,235 enable you to test exactly. 495 00:20:31,740 --> 00:20:33,690 All right. 496 00:20:33,690 --> 00:20:36,762 So OK, I'm, as usual taking too long to do things 497 00:20:36,762 --> 00:20:38,970 so I'm just going to throw out some questions for you 498 00:20:38,970 --> 00:20:41,095 to percolate on and we will come back to them later 499 00:20:41,095 --> 00:20:42,570 in the course. 500 00:20:42,570 --> 00:20:45,840 Do these data-- the fact that you can see this so robustly 501 00:20:45,840 --> 00:20:48,120 in all subjects and that all this evidence suggests 502 00:20:48,120 --> 00:20:50,550 it's really very selective for faces-- 503 00:20:50,550 --> 00:20:55,040 does that tell us that this region is innate? 504 00:20:55,040 --> 00:20:58,700 It's in the same place, more or less, in pretty much everyone. 505 00:20:58,700 --> 00:21:00,380 Does that mean it's innate? 506 00:21:00,380 --> 00:21:01,880 Think about it, OK? 507 00:21:01,880 --> 00:21:04,590 It's not immediately obvious. 508 00:21:04,590 --> 00:21:08,720 Another question, does the fact that this thing responds so 509 00:21:08,720 --> 00:21:12,530 selectively to faces in pretty much everyone 510 00:21:12,530 --> 00:21:14,900 mean that it's necessary for face recognition? 511 00:21:17,503 --> 00:21:18,920 What do you guys think about that? 512 00:21:23,800 --> 00:21:26,440 In the sense of, does that necessarily 513 00:21:26,440 --> 00:21:28,060 mean that if you lost that thing, 514 00:21:28,060 --> 00:21:30,410 you wouldn't be able to recognize faces? 515 00:21:30,410 --> 00:21:30,910 Isabelle. 516 00:21:30,910 --> 00:21:31,920 Is that Isabelle? 517 00:21:31,920 --> 00:21:32,560 AUDIENCE: Yes. 518 00:21:32,560 --> 00:21:35,750 Well, I would think to really test that hypothesis, 519 00:21:35,750 --> 00:21:38,214 you'd have to find someone that [INAUDIBLE] 520 00:21:38,214 --> 00:21:40,110 in that specific area. 521 00:21:40,110 --> 00:21:41,370 NANCY KANWISHER: Exactly. 522 00:21:41,370 --> 00:21:42,560 Exactly. 523 00:21:42,560 --> 00:21:45,040 Exactly, and we'll talk more about that in a moment. 524 00:21:45,040 --> 00:21:48,638 The critical thing is that it's fabulous and powerful and cool 525 00:21:48,638 --> 00:21:50,430 to be able to find this thing in everybody, 526 00:21:50,430 --> 00:21:51,570 measure its response. 527 00:21:51,570 --> 00:21:53,040 It's taken us very far. 528 00:21:53,040 --> 00:21:55,650 But just the fact that people have that thing 529 00:21:55,650 --> 00:21:57,930 doesn't tell us that you need it for face recognition. 530 00:21:57,930 --> 00:22:01,968 It just tells you it turns on when you recognize faces. 531 00:22:01,968 --> 00:22:03,010 This is really important. 532 00:22:03,010 --> 00:22:06,900 We'll keep coming around to this. 533 00:22:06,900 --> 00:22:09,180 Does this tell us how face recognition actually 534 00:22:09,180 --> 00:22:11,670 works in the human brain? 535 00:22:11,670 --> 00:22:13,080 No. 536 00:22:13,080 --> 00:22:16,172 I mean, it's important, but it's barely step zero. 537 00:22:16,172 --> 00:22:17,880 Unfortunately, the field is kind of still 538 00:22:17,880 --> 00:22:20,820 at step zero for most things. 539 00:22:20,820 --> 00:22:23,190 Step zero's better than I guess, I don't know, 540 00:22:23,190 --> 00:22:25,080 maybe I should call it step one. 541 00:22:25,080 --> 00:22:28,470 Anyway, it's something, but doesn't tell us how it works. 542 00:22:28,470 --> 00:22:30,390 OK. 543 00:22:30,390 --> 00:22:32,730 All right, so advantages and disadvantages 544 00:22:32,730 --> 00:22:33,720 of functional MRI. 545 00:22:33,720 --> 00:22:35,820 Advantages, it is, as I mentioned 546 00:22:35,820 --> 00:22:37,620 last time, the best spatial resolution 547 00:22:37,620 --> 00:22:40,080 available for studies on normal subjects 548 00:22:40,080 --> 00:22:43,590 without opening their heads. 549 00:22:43,590 --> 00:22:45,780 That's what it means to say noninvasive. 550 00:22:45,780 --> 00:22:48,990 Disadvantages, as I just said, we 551 00:22:48,990 --> 00:22:51,660 don't know-- just because we see a response there doesn't mean 552 00:22:51,660 --> 00:22:54,210 that that region is causally involved in perception 553 00:22:54,210 --> 00:22:58,260 or cognition or experience. 554 00:22:58,260 --> 00:23:01,590 We don't know exactly what is going on at a neural level 555 00:23:01,590 --> 00:23:04,530 underlying that bold response, that blood flow change. 556 00:23:04,530 --> 00:23:07,560 It could be any metabolic change, not necessarily 557 00:23:07,560 --> 00:23:09,332 neuronal spiking. 558 00:23:09,332 --> 00:23:11,040 So it's a little bit-- it's very indirect 559 00:23:11,040 --> 00:23:13,200 and a little imprecise. 560 00:23:13,200 --> 00:23:14,610 Spatial resolution is much better 561 00:23:14,610 --> 00:23:17,640 than anything else in humans, but it's appallingly bad 562 00:23:17,640 --> 00:23:21,090 compared to anything that people who work on animals can do 563 00:23:21,090 --> 00:23:23,670 or they routinely record from individual neurons 564 00:23:23,670 --> 00:23:27,790 or even dendrites on a neuron. 565 00:23:27,790 --> 00:23:30,630 We are summing over hundreds of thousands 566 00:23:30,630 --> 00:23:33,120 of neurons in each pixel or voxel 567 00:23:33,120 --> 00:23:35,670 that we measure with functional MRI. 568 00:23:35,670 --> 00:23:37,387 It's very expensive. 569 00:23:37,387 --> 00:23:39,720 It's a little cheaper than that here, but in most places 570 00:23:39,720 --> 00:23:41,610 it's more than $600 an hour. 571 00:23:41,610 --> 00:23:44,010 That is a lot. 572 00:23:44,010 --> 00:23:46,620 There are other-- there are parts of the brain where 573 00:23:46,620 --> 00:23:48,240 it's really hard to get any signal 574 00:23:48,240 --> 00:23:49,960 for various physics-y reasons. 575 00:23:52,710 --> 00:23:55,650 And it makes a loud noise, which is not always a problem, 576 00:23:55,650 --> 00:23:58,770 but it's a problem for some things like scanning infants 577 00:23:58,770 --> 00:24:03,180 or like doing auditory experiments. 578 00:24:03,180 --> 00:24:07,230 The temporal resolution is not even close to the time 579 00:24:07,230 --> 00:24:09,300 scale on which vision happens. 580 00:24:09,300 --> 00:24:15,010 So vision is really fast and functional MRI is really slow. 581 00:24:15,010 --> 00:24:15,510 Right? 582 00:24:15,510 --> 00:24:17,100 It's slow, why is it slow? 583 00:24:20,210 --> 00:24:20,930 Yeah. 584 00:24:20,930 --> 00:24:22,460 AUDIENCE: Blood levels take time to change. 585 00:24:22,460 --> 00:24:23,377 NANCY KANWISHER: Yeah. 586 00:24:23,377 --> 00:24:25,430 Just takes a long time for blood flow 587 00:24:25,430 --> 00:24:28,760 to change after the increase in neural activity. 588 00:24:28,760 --> 00:24:30,200 All right. 589 00:24:30,200 --> 00:24:32,360 OK, so back to our questions that we're 590 00:24:32,360 --> 00:24:34,010 asking about face perception. 591 00:24:34,010 --> 00:24:36,050 Where do we get with functional MRI? 592 00:24:36,050 --> 00:24:38,450 Well, actually from both behavior and functional MRI, 593 00:24:38,450 --> 00:24:40,760 it kind of looks like we have a distinct system 594 00:24:40,760 --> 00:24:44,690 for recognizing faces than for recognizing everything else. 595 00:24:44,690 --> 00:24:46,310 I don't think we've totally nailed it. 596 00:24:46,310 --> 00:24:46,970 Yes. 597 00:24:46,970 --> 00:24:48,980 AUDIENCE: So quick question regarding the fMRI. 598 00:24:48,980 --> 00:24:52,940 So the resolution is field of a couple of seconds? 599 00:24:52,940 --> 00:24:53,630 [INAUDIBLE]? 600 00:24:53,630 --> 00:24:54,560 NANCY KANWISHER: Yeah, some people 601 00:24:54,560 --> 00:24:56,685 would say you could get it down to a couple hundred 602 00:24:56,685 --> 00:24:59,030 milliseconds but that's debated. 603 00:24:59,030 --> 00:25:00,950 You have to go to great lengths to do that. 604 00:25:00,950 --> 00:25:03,856 Normal functional MRI, a couple of seconds at best. 605 00:25:03,856 --> 00:25:04,356 Yeah. 606 00:25:08,820 --> 00:25:10,080 All right. 607 00:25:10,080 --> 00:25:12,460 So let's consider this next question. 608 00:25:12,460 --> 00:25:15,270 How fast does face recognition happen? 609 00:25:15,270 --> 00:25:18,120 Now, that may seem like a completely arbitrary question 610 00:25:18,120 --> 00:25:19,410 to ask, but it's not. 611 00:25:19,410 --> 00:25:20,910 Remember, we're trying to understand 612 00:25:20,910 --> 00:25:22,827 the computations that are running in your head 613 00:25:22,827 --> 00:25:24,600 when you recognize faces. 614 00:25:24,600 --> 00:25:26,940 And you might imagine some computations 615 00:25:26,940 --> 00:25:31,140 that are iterative-- that involve multiple repeated 616 00:25:31,140 --> 00:25:34,170 testing of hypotheses, generative models, whatever-- 617 00:25:34,170 --> 00:25:37,980 things that involve lots of iterated feedback versus things 618 00:25:37,980 --> 00:25:40,200 where you just have a feed forward sweep up 619 00:25:40,200 --> 00:25:41,140 the visual system. 620 00:25:41,140 --> 00:25:43,260 And so there might be very different time scales 621 00:25:43,260 --> 00:25:48,100 for those different kinds of mental processes. 622 00:25:48,100 --> 00:25:49,373 So we just went through this. 623 00:25:49,373 --> 00:25:51,540 Functional MRI is not going to answer this question. 624 00:25:51,540 --> 00:25:52,590 It's just not. 625 00:25:52,590 --> 00:25:53,940 It's a bummer, but that's life. 626 00:25:53,940 --> 00:25:55,523 We're adults, we're going to just move 627 00:25:55,523 --> 00:25:57,277 on and use a different method. 628 00:25:57,277 --> 00:25:59,110 OK, so there's a bunch of different methods. 629 00:25:59,110 --> 00:26:01,680 One is kind of been around forever. 630 00:26:01,680 --> 00:26:04,830 You glue electrodes on the head, right? 631 00:26:04,830 --> 00:26:06,210 Sometimes you push the hair apart 632 00:26:06,210 --> 00:26:10,320 or try to find bald people and glue electrodes right on there. 633 00:26:10,320 --> 00:26:13,500 And you can use, in the old days, about 10 electrodes, 634 00:26:13,500 --> 00:26:16,020 or you can use in more modern devices 635 00:26:16,020 --> 00:26:19,980 these nets with a few hundred electrodes 636 00:26:19,980 --> 00:26:21,630 that you settle onto the head. 637 00:26:21,630 --> 00:26:25,170 And so then you just measure directly electrical potentials 638 00:26:25,170 --> 00:26:27,360 right on the scalp. 639 00:26:27,360 --> 00:26:30,960 So what's cool about that is it's totally non-invasive. 640 00:26:30,960 --> 00:26:33,780 And it gives you a beautiful online temporal measure 641 00:26:33,780 --> 00:26:36,000 of underlying neural activity. 642 00:26:36,000 --> 00:26:39,780 What's not so cool about it is that electrical potentials blur 643 00:26:39,780 --> 00:26:41,940 all over the scalp and the spatial resolution 644 00:26:41,940 --> 00:26:44,520 is really awful. 645 00:26:44,520 --> 00:26:47,490 So the analogy has been made that it 646 00:26:47,490 --> 00:26:49,470 would be like sticking a microphone 647 00:26:49,470 --> 00:26:54,450 on the inside of the top of a football stadium 648 00:26:54,450 --> 00:26:57,750 and collecting audio there. 649 00:26:57,750 --> 00:27:00,287 You would know when a touchdown was scored. 650 00:27:00,287 --> 00:27:01,620 There's a lot of noise all over. 651 00:27:01,620 --> 00:27:04,245 It's like, OK, there's an event, we detected that event, right? 652 00:27:04,245 --> 00:27:06,703 You might be able to tell a touchdown from something else. 653 00:27:06,703 --> 00:27:09,120 I don't know about football so I can't tell you what else. 654 00:27:09,120 --> 00:27:11,578 Anyway, something else, some other event that could happen. 655 00:27:11,578 --> 00:27:18,210 OK, so that will be useful for some things, but kind of crude. 656 00:27:18,210 --> 00:27:20,500 But you'd have a hell of a time telling anything else, 657 00:27:20,500 --> 00:27:22,500 like what one person is saying to another person 658 00:27:22,500 --> 00:27:24,780 in the bleachers. 659 00:27:24,780 --> 00:27:26,040 So that's the old analogy. 660 00:27:26,040 --> 00:27:30,850 This is changing slightly, and we'll get to that later. 661 00:27:30,850 --> 00:27:32,610 But first, I want to briefly mention 662 00:27:32,610 --> 00:27:34,837 one of the assigned readings that I just 663 00:27:34,837 --> 00:27:36,670 hoped you guys could figure out on your own. 664 00:27:36,670 --> 00:27:38,880 But just in case you were confused about it, 665 00:27:38,880 --> 00:27:42,690 the point I wanted you to get from the Thorpe reading 666 00:27:42,690 --> 00:27:46,950 is he's asking how quickly can we tell if an image contains 667 00:27:46,950 --> 00:27:48,180 an animal or not? 668 00:27:48,180 --> 00:27:51,990 It's a kind of way to say, how fast is object recognition? 669 00:27:51,990 --> 00:27:52,930 So what does he do? 670 00:27:52,930 --> 00:27:56,040 He has people look at a bunch of images and they press this 671 00:27:56,040 --> 00:27:58,980 button if it has an animal in this button if it doesn't. 672 00:27:58,980 --> 00:28:00,580 Really simple task. 673 00:28:00,580 --> 00:28:03,728 So first question is, why not just use those reaction times? 674 00:28:03,728 --> 00:28:06,270 We can measure how fast it takes for people to press a button 675 00:28:06,270 --> 00:28:07,920 after the image comes on. 676 00:28:07,920 --> 00:28:10,200 Why not just use that? 677 00:28:10,200 --> 00:28:12,960 Does that tell us how fast object recognition occurs? 678 00:28:12,960 --> 00:28:14,400 Yeah, Jimmy. 679 00:28:14,400 --> 00:28:17,070 AUDIENCE: It doesn't because if you perceive that and then 680 00:28:17,070 --> 00:28:18,695 it also activates the motor neurons 681 00:28:18,695 --> 00:28:19,903 and it takes time to respond. 682 00:28:19,903 --> 00:28:22,528 NANCY KANWISHER: Yeah, you have to take all that time to figure 683 00:28:22,528 --> 00:28:24,210 out, OK, I see the animal. 684 00:28:24,210 --> 00:28:25,530 OK, which button is that? 685 00:28:25,530 --> 00:28:27,102 And then which finger do I push? 686 00:28:27,102 --> 00:28:29,310 And then you've got to send a signal all the way down 687 00:28:29,310 --> 00:28:31,727 here, conduction velocity all the way down to your finger, 688 00:28:31,727 --> 00:28:32,800 that takes a long time. 689 00:28:32,800 --> 00:28:34,740 And so it includes all that motor stuff 690 00:28:34,740 --> 00:28:36,450 in with the perceptual stuff. 691 00:28:36,450 --> 00:28:39,570 We could make some guesses about how long that motor 692 00:28:39,570 --> 00:28:42,210 stuff takes, but it's still not very precise. 693 00:28:42,210 --> 00:28:44,010 So the point of the Thorpe paper is 694 00:28:44,010 --> 00:28:46,590 they're basically trying to collect a reaction time out 695 00:28:46,590 --> 00:28:48,510 of the neurons in the head, right? 696 00:28:48,510 --> 00:28:50,190 So they're trying to actually collect-- 697 00:28:50,190 --> 00:28:51,857 it's essentially what they're collecting 698 00:28:51,857 --> 00:28:54,150 in this case is more of the motor response 699 00:28:54,150 --> 00:28:56,590 because they're collecting responses over frontal lobes, 700 00:28:56,590 --> 00:28:57,090 right? 701 00:28:57,090 --> 00:28:58,673 And we haven't talked about this much. 702 00:28:58,673 --> 00:29:01,170 But all of the visual stuff we've been talking about all 703 00:29:01,170 --> 00:29:03,960 happens in the back of the head. 704 00:29:03,960 --> 00:29:05,868 More motor planning stuff mostly happens 705 00:29:05,868 --> 00:29:06,910 in the front of the head. 706 00:29:06,910 --> 00:29:09,270 And so they're collecting responses out of here, 707 00:29:09,270 --> 00:29:12,030 averaging over a bunch of frontal responses. 708 00:29:12,030 --> 00:29:14,070 And they see the average response when there's 709 00:29:14,070 --> 00:29:16,140 an animal-- this is just potential average 710 00:29:16,140 --> 00:29:19,320 over those frontal electrodes-- is like this. 711 00:29:19,320 --> 00:29:21,760 And when there's no animal it's like that. 712 00:29:21,760 --> 00:29:26,040 And so what does that tell us about how fast people can 713 00:29:26,040 --> 00:29:28,980 distinguish whether an image has an animal or not? 714 00:29:32,270 --> 00:29:33,020 Yes? 715 00:29:33,020 --> 00:29:33,670 Yeah. 716 00:29:33,670 --> 00:29:35,212 AUDIENCE: It's less than that number. 717 00:29:35,212 --> 00:29:37,330 NANCY KANWISHER: Less than? 718 00:29:37,330 --> 00:29:39,020 AUDIENCE: 150, 160. 719 00:29:39,020 --> 00:29:41,680 NANCY KANWISHER: OK, why less than 150? 720 00:29:41,680 --> 00:29:44,230 AUDIENCE: I've read the paper so it's kind of cheating, so. 721 00:29:44,230 --> 00:29:44,930 NANCY KANWISHER: That's OK. 722 00:29:44,930 --> 00:29:45,760 That's good. 723 00:29:45,760 --> 00:29:46,360 That's fine. 724 00:29:46,360 --> 00:29:46,860 Go ahead. 725 00:29:49,530 --> 00:29:51,540 AUDIENCE: It gives you around-- 726 00:29:51,540 --> 00:29:53,400 the 150 second is giving you a [INAUDIBLE] 727 00:29:53,400 --> 00:29:56,700 saying some process has been registered 728 00:29:56,700 --> 00:29:58,950 and now you're trying to do something else 729 00:29:58,950 --> 00:30:00,272 in the case of non-animals. 730 00:30:00,272 --> 00:30:01,230 NANCY KANWISHER: Right. 731 00:30:01,230 --> 00:30:03,540 AUDIENCE: So the deviation starts 732 00:30:03,540 --> 00:30:05,700 getting you that OK, two different actions have 733 00:30:05,700 --> 00:30:06,630 started taking place. 734 00:30:06,630 --> 00:30:07,505 NANCY KANWISHER: Yep. 735 00:30:07,505 --> 00:30:09,150 AUDIENCE: So by that time, the image 736 00:30:09,150 --> 00:30:11,280 ought to have been sort of fully processed. 737 00:30:11,280 --> 00:30:13,533 So that should be something less than that number. 738 00:30:13,533 --> 00:30:14,450 NANCY KANWISHER: Yeah. 739 00:30:14,450 --> 00:30:16,990 Yeah, did everybody get that? 740 00:30:16,990 --> 00:30:18,180 It's actually quite subtle. 741 00:30:18,180 --> 00:30:22,350 So the key thing is, these curves 742 00:30:22,350 --> 00:30:24,420 diverge right there at 150. 743 00:30:24,420 --> 00:30:27,535 So that tells you that by 150 milliseconds, 744 00:30:27,535 --> 00:30:29,910 something in your brain is happening different if there's 745 00:30:29,910 --> 00:30:32,100 an animal and not an animal. 746 00:30:32,100 --> 00:30:33,360 That's the key question. 747 00:30:33,360 --> 00:30:35,040 But what is that something? 748 00:30:35,040 --> 00:30:38,340 It may be your motor preparation of the response. 749 00:30:38,340 --> 00:30:40,188 In that case, the actual visual part 750 00:30:40,188 --> 00:30:41,730 happened before, because you wouldn't 751 00:30:41,730 --> 00:30:43,730 know which button to press if you hadn't already 752 00:30:43,730 --> 00:30:45,120 recognized it. 753 00:30:45,120 --> 00:30:49,790 So it's an upper bound for when that process happened, 754 00:30:49,790 --> 00:30:51,540 because maybe it happened before and we're 755 00:30:51,540 --> 00:30:54,510 looking at a later stage, OK? 756 00:30:54,510 --> 00:30:56,850 Does that make sense? 757 00:30:56,850 --> 00:31:01,170 But also, it's an upper bound for the beginning of that 758 00:31:01,170 --> 00:31:02,130 process. 759 00:31:02,130 --> 00:31:04,950 Because the fact that those electrode responses have 760 00:31:04,950 --> 00:31:07,320 diverged doesn't mean you've finished processing 761 00:31:07,320 --> 00:31:09,810 whether it's an animal or not. 762 00:31:09,810 --> 00:31:12,270 So it's kind of a subtle business reasoning from this. 763 00:31:15,090 --> 00:31:18,840 OK, so that's all that. 764 00:31:18,840 --> 00:31:21,990 So that's a case with detecting animals. 765 00:31:21,990 --> 00:31:26,220 What about faces, to get back to our theme for today? 766 00:31:26,220 --> 00:31:29,550 Yes, you can learn about the speed of face detection 767 00:31:29,550 --> 00:31:31,710 at least with the ERPs. 768 00:31:31,710 --> 00:31:35,010 And so here's the first paper that did that back in 1996. 769 00:31:35,010 --> 00:31:37,690 They had electrodes where are these? 770 00:31:37,690 --> 00:31:39,120 Just right around here and here. 771 00:31:39,120 --> 00:31:44,130 I actually have those electrode locations tattooed on my scalp, 772 00:31:44,130 --> 00:31:45,420 color-coded anyway. 773 00:31:45,420 --> 00:31:46,085 Yes? 774 00:31:46,085 --> 00:31:48,505 AUDIENCE: Is ERP just the same as an EEG, just 775 00:31:48,505 --> 00:31:49,380 in a specific plan e? 776 00:31:49,380 --> 00:31:52,520 NANCY KANWISHER: Yes, exactly. 777 00:31:52,520 --> 00:31:55,410 It's the same as an EEG except what you do 778 00:31:55,410 --> 00:31:59,670 is you time lock the data collection to stimulus onset. 779 00:31:59,670 --> 00:32:02,640 So it actually stands for Event-Related Potential. 780 00:32:02,640 --> 00:32:04,140 And the reason it's event-related 781 00:32:04,140 --> 00:32:06,510 is you collect all those trials and you time 782 00:32:06,510 --> 00:32:10,363 lock to stimulus onset, and then you signal average. 783 00:32:10,363 --> 00:32:12,030 I had a slide on that but I took it out. 784 00:32:12,030 --> 00:32:12,863 It was too detailed. 785 00:32:12,863 --> 00:32:15,300 But that's exactly the idea, yeah. 786 00:32:15,300 --> 00:32:18,930 So here, stimulus onset is right around here. 787 00:32:18,930 --> 00:32:20,432 This is time going this way. 788 00:32:20,432 --> 00:32:22,140 And what you see-- it's hard to see here, 789 00:32:22,140 --> 00:32:23,640 but the faces are right there. 790 00:32:23,640 --> 00:32:27,000 And at 170 milliseconds after stimulus onset, 791 00:32:27,000 --> 00:32:31,470 there's a bigger bump for faces at an electrode 792 00:32:31,470 --> 00:32:33,330 approximately here. 793 00:32:33,330 --> 00:32:34,920 And even more so-- 794 00:32:34,920 --> 00:32:38,340 actually, even more so over the right hemisphere right there. 795 00:32:38,340 --> 00:32:41,880 Compared to cars and scrambled faces and stuff like that. 796 00:32:41,880 --> 00:32:42,658 Yeah? 797 00:32:42,658 --> 00:32:44,325 AUDIENCE: What is ERP exactly measuring? 798 00:32:44,325 --> 00:32:45,163 Is it just activity? 799 00:32:45,163 --> 00:32:46,080 NANCY KANWISHER: Yeah. 800 00:32:46,080 --> 00:32:50,970 So again, it's electrodes glued on your scalp 801 00:32:50,970 --> 00:32:54,600 or just stuck there with some kind of icky gel. 802 00:32:54,600 --> 00:32:56,640 And so they're just measuring potentials. 803 00:32:56,640 --> 00:32:59,400 And so the idea is that's neural activity somewhere 804 00:32:59,400 --> 00:33:04,440 underneath those electrodes, but maybe anywhere within inches. 805 00:33:04,440 --> 00:33:07,200 Like a long-- probably average is over much of the whole lobe 806 00:33:07,200 --> 00:33:07,740 underneath. 807 00:33:07,740 --> 00:33:09,870 So it's very spatially blurry, but it's 808 00:33:09,870 --> 00:33:14,370 giving you summed idea of activity under that electrode. 809 00:33:14,370 --> 00:33:15,390 Make sense? 810 00:33:15,390 --> 00:33:17,040 Electrical activity, because it's 811 00:33:17,040 --> 00:33:19,980 the direct electrical consequence of neural activity, 812 00:33:19,980 --> 00:33:24,870 it's very precisely time locked, unlike functional MRI, which 813 00:33:24,870 --> 00:33:26,130 is going by way of blood flow. 814 00:33:28,770 --> 00:33:32,490 OK, so that tells us that we have a face-specific response 815 00:33:32,490 --> 00:33:35,580 at 170 milliseconds. 816 00:33:35,580 --> 00:33:38,070 And that's sort of more evidence that there 817 00:33:38,070 --> 00:33:40,680 might be something special in the brain for face recognition. 818 00:33:40,680 --> 00:33:41,940 That's useful. 819 00:33:41,940 --> 00:33:46,530 It tells us that faces are discriminated from non-faces, 820 00:33:46,530 --> 00:33:48,960 or they've begun to be discriminated 821 00:33:48,960 --> 00:33:51,540 from non-faces by 170 milliseconds 822 00:33:51,540 --> 00:33:53,520 after the stimulus comes on. 823 00:33:53,520 --> 00:33:55,050 Make sense? 824 00:33:55,050 --> 00:33:58,230 OK, now do we know whether the signals coming 825 00:33:58,230 --> 00:34:00,420 from the fusiform face area? 826 00:34:00,420 --> 00:34:02,460 No, we have no idea. 827 00:34:02,460 --> 00:34:04,500 It's probably somewhere in the back of the head, 828 00:34:04,500 --> 00:34:05,820 because you get it better with electrodes 829 00:34:05,820 --> 00:34:07,237 back here than electrodes up here. 830 00:34:07,237 --> 00:34:08,170 But that's about it. 831 00:34:08,170 --> 00:34:11,460 That's all you can tell. 832 00:34:11,460 --> 00:34:14,190 So can we do a little bit better localizing 833 00:34:14,190 --> 00:34:15,540 the source of that signal? 834 00:34:15,540 --> 00:34:18,960 Well, maybe a hair better using a very similar method 835 00:34:18,960 --> 00:34:22,030 called magnetoencephalography. 836 00:34:22,030 --> 00:34:26,670 So this is a picture that Chris Brewer took of Leyla Isik 837 00:34:26,670 --> 00:34:29,820 postdoc in my lab, and me and the MEG system. 838 00:34:29,820 --> 00:34:32,730 This is in on the other side of the building. 839 00:34:32,730 --> 00:34:37,380 So MEG is a lot like EEG and ERPs 840 00:34:37,380 --> 00:34:41,520 except that it detects magnetic fields, not electric fields. 841 00:34:41,520 --> 00:34:47,070 And it does this by having these several hundred devices that 842 00:34:47,070 --> 00:34:48,690 are placed right next to your head 843 00:34:48,690 --> 00:34:50,130 in this big hairdryer thing. 844 00:34:50,130 --> 00:34:53,159 There's 300 devices in there that measure 845 00:34:53,159 --> 00:34:56,639 teeny tiny magnetic field changes that 846 00:34:56,639 --> 00:34:58,780 happen with neural activity. 847 00:34:58,780 --> 00:35:02,460 And the crux of the idea is this is a cross-section 848 00:35:02,460 --> 00:35:03,280 through the brain. 849 00:35:03,280 --> 00:35:07,860 So remember in Graybiel's dissection, this is cortex here 850 00:35:07,860 --> 00:35:09,060 and this is underlying. 851 00:35:09,060 --> 00:35:13,040 What is this stuff underneath it? 852 00:35:13,040 --> 00:35:13,942 Sorry? 853 00:35:13,942 --> 00:35:14,900 AUDIENCE: White matter. 854 00:35:14,900 --> 00:35:15,830 NANCY KANWISHER: White matter, yeah. 855 00:35:15,830 --> 00:35:17,210 Well, those are all the fibers. 856 00:35:17,210 --> 00:35:21,340 OK, so the activity that underlies perception 857 00:35:21,340 --> 00:35:23,840 and cognition mostly happens in the gray matter 858 00:35:23,840 --> 00:35:26,010 where the cell bodies are. 859 00:35:26,010 --> 00:35:29,060 And so a lot of that activity goes in a direction 860 00:35:29,060 --> 00:35:31,730 perpendicular to the cortical orientation 861 00:35:31,730 --> 00:35:35,730 with these cells that cross the cortical surface like that. 862 00:35:35,730 --> 00:35:38,960 So if you remember 8.02-- 863 00:35:38,960 --> 00:35:40,340 if you have activity that's going 864 00:35:40,340 --> 00:35:43,790 through the cortex like this, right hand rule, 865 00:35:43,790 --> 00:35:46,940 the magnetic field here is going to be 866 00:35:46,940 --> 00:35:50,480 a consequence of that electrical activity in this direction. 867 00:35:50,480 --> 00:35:53,030 It's going to mostly stay within the cortex. 868 00:35:53,030 --> 00:35:54,500 Everybody see how that's true? 869 00:35:54,500 --> 00:35:57,290 That's not so great, because our detectors are out there, 870 00:35:57,290 --> 00:35:59,060 outside the cortex. 871 00:35:59,060 --> 00:36:02,570 However, consider the activity that's in the sulcus in here, 872 00:36:02,570 --> 00:36:04,760 in this fold of the brain. 873 00:36:04,760 --> 00:36:07,970 Electrical activity in this direction, right hand rule, 874 00:36:07,970 --> 00:36:09,830 will stick outside the brain. 875 00:36:09,830 --> 00:36:12,170 And we can detect it with our magnetic sensors. 876 00:36:12,170 --> 00:36:13,980 Does that make sense? 877 00:36:13,980 --> 00:36:17,900 So you can sort of see most cortical activity 878 00:36:17,900 --> 00:36:21,080 better if it's in a sulcus, or at least in part 879 00:36:21,080 --> 00:36:23,750 of the cortical surface that's perpendicular to the scalp 880 00:36:23,750 --> 00:36:26,600 where the detectors are just because of the orientation 881 00:36:26,600 --> 00:36:28,310 in the right hand rule. 882 00:36:28,310 --> 00:36:35,540 OK, so it primarily sees activity in the folds or sulci, 883 00:36:35,540 --> 00:36:39,440 not in the outer bumps gyri. 884 00:36:39,440 --> 00:36:42,650 Field strengths are minuscule as a consequence 885 00:36:42,650 --> 00:36:43,920 of neural activity. 886 00:36:43,920 --> 00:36:48,470 So the fields we measure are 10 to the minus 13th Tesla, 887 00:36:48,470 --> 00:36:51,600 a million times weaker than the Earth's magnetic field. 888 00:36:51,600 --> 00:36:55,190 So you can imagine that if you set up an MEG system 889 00:36:55,190 --> 00:36:57,200 you need a lot of shielding. 890 00:36:57,200 --> 00:37:00,590 We had a whole rigmarole when the MEG system was set up 891 00:37:00,590 --> 00:37:03,380 in this building because it's right near the subway 892 00:37:03,380 --> 00:37:05,180 and the train. 893 00:37:05,180 --> 00:37:07,400 And so there are many, many layers 894 00:37:07,400 --> 00:37:09,180 of copper shielding to protect it. 895 00:37:09,180 --> 00:37:12,680 So we can detect these teeny tiny magnetic fields 896 00:37:12,680 --> 00:37:15,700 from the brain's activity separated 897 00:37:15,700 --> 00:37:17,450 from the noise of the outside world, which 898 00:37:17,450 --> 00:37:20,000 is much greater in magnitude. 899 00:37:20,000 --> 00:37:22,610 OK, so-- all right. 900 00:37:22,610 --> 00:37:25,640 So actually, MEG was invented here at MIT 901 00:37:25,640 --> 00:37:27,410 by this guy, David Cohen. 902 00:37:27,410 --> 00:37:31,880 And this is the first MEG device ever built, very cool, 903 00:37:31,880 --> 00:37:36,230 way back in 1968. 904 00:37:36,230 --> 00:37:39,000 And what can it tell us about face perception? 905 00:37:39,000 --> 00:37:39,680 Well, a lot. 906 00:37:39,680 --> 00:37:43,190 I'll give you just one rudimentary example. 907 00:37:43,190 --> 00:37:46,670 That M170 that you can detect with scalp electrodes, 908 00:37:46,670 --> 00:37:49,800 you can also detect with magnetic sensors on the head. 909 00:37:49,800 --> 00:37:52,320 So here's some of our data from a long time ago. 910 00:37:52,320 --> 00:37:55,400 This is the strength of the magnetic field 911 00:37:55,400 --> 00:37:58,760 at sites right about out here. 912 00:37:58,760 --> 00:38:01,670 And you can see a face-selective response also 913 00:38:01,670 --> 00:38:09,320 at 170 milliseconds, just like you can with scalp electrodes. 914 00:38:09,320 --> 00:38:12,740 So that tells us that at least you've 915 00:38:12,740 --> 00:38:15,710 started to detect faces by 170 milliseconds. 916 00:38:15,710 --> 00:38:18,470 That's pretty fast. 917 00:38:18,470 --> 00:38:20,480 And again, it's more evidence that there's 918 00:38:20,480 --> 00:38:21,710 specialized machinery. 919 00:38:21,710 --> 00:38:26,030 These data don't yet go beyond the EEG data, the ERP data 920 00:38:26,030 --> 00:38:27,920 from electrical potentials. 921 00:38:27,920 --> 00:38:29,810 But they might, in principle, and there's 922 00:38:29,810 --> 00:38:32,630 lots of ongoing work trying to do that. 923 00:38:32,630 --> 00:38:36,290 OK, overview, advantages of these methods, 924 00:38:36,290 --> 00:38:38,690 both EEG and MEG. 925 00:38:38,690 --> 00:38:40,310 They're non-invasive-- that means you 926 00:38:40,310 --> 00:38:41,477 don't need to open the head. 927 00:38:41,477 --> 00:38:44,840 A very good thing, especially if you're the subject. 928 00:38:44,840 --> 00:38:46,920 They have very good temporal resolution. 929 00:38:46,920 --> 00:38:48,470 And if we want to see computations 930 00:38:48,470 --> 00:38:53,780 unfolding over time in the brain, this is a good way. 931 00:38:53,780 --> 00:38:57,410 I just said why we'd I care about that. 932 00:38:57,410 --> 00:39:01,658 OK, so far-- well, never mind, I'm going to skip this point. 933 00:39:01,658 --> 00:39:02,450 Not that important. 934 00:39:02,450 --> 00:39:05,360 We will get back and do more sophisticated things with EEG 935 00:39:05,360 --> 00:39:07,790 and MEG in subsequent lectures. 936 00:39:07,790 --> 00:39:11,780 Disadvantages-- spatial resolution is terrible. 937 00:39:11,780 --> 00:39:14,700 And this is another kind of ill-posed problem. 938 00:39:14,700 --> 00:39:16,490 So just as the brain is facing lots 939 00:39:16,490 --> 00:39:20,060 of ill-posed problems in perception and cognition, 940 00:39:20,060 --> 00:39:22,520 we scientists are facing ill-posed problems 941 00:39:22,520 --> 00:39:26,390 when we collect electrical or magnetic activity at the scalp 942 00:39:26,390 --> 00:39:28,730 and try to infer the exact location in the brain 943 00:39:28,730 --> 00:39:30,560 where it's coming from. 944 00:39:30,560 --> 00:39:33,200 It's a similar problem to the problem of invariant object 945 00:39:33,200 --> 00:39:34,140 recognition. 946 00:39:34,140 --> 00:39:36,380 There are many possible configurations 947 00:39:36,380 --> 00:39:38,900 of sources in the brain that could give rise 948 00:39:38,900 --> 00:39:41,570 to the same set of electrical and magnetic fields out 949 00:39:41,570 --> 00:39:42,530 of the scalp. 950 00:39:42,530 --> 00:39:44,210 And that means it's ill-posed. 951 00:39:44,210 --> 00:39:47,210 We don't have a way to get a unique solution. 952 00:39:47,210 --> 00:39:51,020 So all that to say we can't figure out the exact sources. 953 00:39:51,020 --> 00:39:54,360 We can make some guesses, but it's not very good. 954 00:39:54,360 --> 00:39:55,610 So what do we do? 955 00:39:55,610 --> 00:39:56,860 Just give up? 956 00:39:56,860 --> 00:40:00,300 No, we use another method. 957 00:40:00,300 --> 00:40:03,450 So here's an amazing method. 958 00:40:03,450 --> 00:40:06,230 This is the one method in humans that 959 00:40:06,230 --> 00:40:10,200 gives us high resolution in both space and time. 960 00:40:10,200 --> 00:40:12,740 And that's when we have the very rare opportunity 961 00:40:12,740 --> 00:40:16,410 to record directly from inside the human brain. 962 00:40:16,410 --> 00:40:20,480 This happens only in the context of neurosurgery. 963 00:40:20,480 --> 00:40:25,010 So neurosurgical patients-- like this guy here, 964 00:40:25,010 --> 00:40:27,620 who you'll meet in a little bit-- 965 00:40:27,620 --> 00:40:30,740 this guy had intractable epilepsy. 966 00:40:30,740 --> 00:40:32,570 And most people with epilepsy are 967 00:40:32,570 --> 00:40:36,170 treated well by drugs that suppress seizures. 968 00:40:36,170 --> 00:40:39,470 But some people are just not responsive to drugs. 969 00:40:39,470 --> 00:40:41,090 And if the seizures are bad enough, 970 00:40:41,090 --> 00:40:43,490 they can be totally life disrupting. 971 00:40:43,490 --> 00:40:46,020 If they happen dozens of times a day, 972 00:40:46,020 --> 00:40:47,450 you just can't live a normal life. 973 00:40:47,450 --> 00:40:50,810 And under those rather extreme circumstances, 974 00:40:50,810 --> 00:40:53,240 sometimes the best option is neurosurgery. 975 00:40:53,240 --> 00:40:57,110 That is, trying to find the source of those seizures 976 00:40:57,110 --> 00:40:59,060 and trying to remove it surgically. 977 00:40:59,060 --> 00:41:02,090 OK, so you hope you never have to go through this or anyone 978 00:41:02,090 --> 00:41:03,590 you care about has to go through it. 979 00:41:03,590 --> 00:41:04,970 It's no picnic. 980 00:41:04,970 --> 00:41:06,920 But actually, this surgical treatment 981 00:41:06,920 --> 00:41:10,460 is often very effective. 982 00:41:10,460 --> 00:41:13,550 So when neurosurgeons decide to do this, 983 00:41:13,550 --> 00:41:17,150 they have to remove a whole piece of skull bone 984 00:41:17,150 --> 00:41:19,050 to get access to the brain. 985 00:41:19,050 --> 00:41:20,660 They have to go through what structure 986 00:41:20,660 --> 00:41:26,115 that Ann Graybiel showed you in her dissection the other day. 987 00:41:26,115 --> 00:41:26,990 What do you have to-- 988 00:41:26,990 --> 00:41:30,570 after you take off the a skull patch? 989 00:41:30,570 --> 00:41:31,070 Yes. 990 00:41:31,070 --> 00:41:31,970 AUDIENCE: Dura mater. 991 00:41:31,970 --> 00:41:33,512 NANCY KANWISHER: Dura mater, exactly. 992 00:41:33,512 --> 00:41:35,870 That nice big piece of white, leathery stuff 993 00:41:35,870 --> 00:41:38,010 that was sitting over the surface of the brain. 994 00:41:38,010 --> 00:41:40,610 So you to take off a piece of skull, 995 00:41:40,610 --> 00:41:44,210 then you need to cut through and push apart the dura. 996 00:41:44,210 --> 00:41:46,850 And then what they sometimes do is stick electrodes straight 997 00:41:46,850 --> 00:41:48,890 on the surface of the brain. 998 00:41:48,890 --> 00:41:50,820 And they do that for two reasons. 999 00:41:50,820 --> 00:41:54,738 One, if they have enough of them sampled far enough apart, 1000 00:41:54,738 --> 00:41:56,780 they can kind of triangulate and figure out where 1001 00:41:56,780 --> 00:41:58,410 is the source of the seizure. 1002 00:41:58,410 --> 00:42:00,710 So the patient hangs out in the hospital for a week 1003 00:42:00,710 --> 00:42:03,200 or so with these electrodes in their head waiting 1004 00:42:03,200 --> 00:42:03,943 to have seizures. 1005 00:42:03,943 --> 00:42:05,360 And then when they have a seizure, 1006 00:42:05,360 --> 00:42:07,670 the clinicians can figure out where 1007 00:42:07,670 --> 00:42:11,510 the source is so they know what bit to cut out. 1008 00:42:11,510 --> 00:42:14,150 The other reason to do this is to map functions. 1009 00:42:14,150 --> 00:42:17,570 Because once the surgeons decide they have to go in and cut, 1010 00:42:17,570 --> 00:42:19,700 they want to try to not cut out any 1011 00:42:19,700 --> 00:42:21,290 of the most important parts. 1012 00:42:21,290 --> 00:42:23,207 I don't know what it means to have unimportant 1013 00:42:23,207 --> 00:42:26,780 parts of the brain, but they try to avoid language regions 1014 00:42:26,780 --> 00:42:29,090 and stuff like that because then patients really 1015 00:42:29,090 --> 00:42:32,823 notice if they lose those things or motor control regions. 1016 00:42:32,823 --> 00:42:34,490 OK, so they map out functions where they 1017 00:42:34,490 --> 00:42:36,350 might be planning their route. 1018 00:42:36,350 --> 00:42:38,310 OK, make sense? 1019 00:42:38,310 --> 00:42:43,640 Now, some of these patients are very kind and generous 1020 00:42:43,640 --> 00:42:47,060 to the world and say, yes, you scientists 1021 00:42:47,060 --> 00:42:48,530 can measure responses in my brain 1022 00:42:48,530 --> 00:42:52,320 while I look at your damn stimuli. 1023 00:42:52,320 --> 00:42:55,550 And so whenever we can, we ask them please, please, please, 1024 00:42:55,550 --> 00:42:57,800 can we show you some pictures or play you some tones 1025 00:42:57,800 --> 00:42:59,690 or have you read some sentences while we 1026 00:42:59,690 --> 00:43:01,310 record from your brain. 1027 00:43:01,310 --> 00:43:04,410 And some of those patients very kindly let us do that. 1028 00:43:04,410 --> 00:43:07,040 And that gives us the most amazing data 1029 00:43:07,040 --> 00:43:08,870 you can get from human brains. 1030 00:43:08,870 --> 00:43:11,270 So for example, I had a rare opportunity 1031 00:43:11,270 --> 00:43:14,090 to do this a few years ago from this lovely guy who 1032 00:43:14,090 --> 00:43:16,850 was undergoing neurosurgery in Japan. 1033 00:43:16,850 --> 00:43:19,790 And while he had electrodes in his brain, a colleague of mine 1034 00:43:19,790 --> 00:43:22,500 was there and emailed me and said, 1035 00:43:22,500 --> 00:43:23,990 look where these electrodes are-- 1036 00:43:23,990 --> 00:43:25,820 right near regions I care about-- 1037 00:43:25,820 --> 00:43:27,710 do you want to show us some stimuli 1038 00:43:27,710 --> 00:43:30,320 and we'll record responses from those electrodes? 1039 00:43:30,320 --> 00:43:34,260 And I said, damn straight I want to send you some stimuli. 1040 00:43:34,260 --> 00:43:37,130 So my students and I stayed up for a couple of days 1041 00:43:37,130 --> 00:43:39,260 and made some stimuli and shot them to Japan 1042 00:43:39,260 --> 00:43:43,650 and got some responses from those very electrodes. 1043 00:43:43,650 --> 00:43:44,780 And here they are. 1044 00:43:44,780 --> 00:43:50,690 So this is a strip of two parallel strips of electrodes 1045 00:43:50,690 --> 00:43:52,520 right along the fusiform gyrus, right 1046 00:43:52,520 --> 00:43:55,640 where the fusiform face area should be in most people. 1047 00:43:55,640 --> 00:43:58,220 And here are the responses of each of those electrodes. 1048 00:43:58,220 --> 00:44:01,680 174 is here, what's 173 and so forth. 1049 00:44:01,680 --> 00:44:04,670 And what you see is this batch of electrodes right here-- 1050 00:44:04,670 --> 00:44:07,688 this is a response when the patient was looking at faces. 1051 00:44:07,688 --> 00:44:09,230 And these are the responses when they 1052 00:44:09,230 --> 00:44:11,630 looked at a whole bunch of different kinds of stimuli. 1053 00:44:11,630 --> 00:44:13,490 Objects, and this guy is Japanese 1054 00:44:13,490 --> 00:44:16,430 so we showed him Kana and Kanji and digit strings 1055 00:44:16,430 --> 00:44:17,990 and other kinds of stuff. 1056 00:44:17,990 --> 00:44:20,150 Very low response to those other things. 1057 00:44:20,150 --> 00:44:23,300 This is a extremely selective response. 1058 00:44:23,300 --> 00:44:26,270 It's much more selective than you see with functional MRI 1059 00:44:26,270 --> 00:44:27,800 because we were recording directly 1060 00:44:27,800 --> 00:44:29,870 from the surface of the brain. 1061 00:44:29,870 --> 00:44:32,310 Further, we have time information. 1062 00:44:32,310 --> 00:44:34,550 This axis here is time, and you can 1063 00:44:34,550 --> 00:44:36,890 see that that response-- well, you can't see the axis, 1064 00:44:36,890 --> 00:44:39,830 but that response starts up at around 1:30 1065 00:44:39,830 --> 00:44:42,920 milliseconds and peaks up there at around 170. 1066 00:44:42,920 --> 00:44:44,900 Everybody clear what we're seeing here 1067 00:44:44,900 --> 00:44:48,200 and why this is so vastly better than either functional MRI 1068 00:44:48,200 --> 00:44:51,140 or MEG or ERPs or anything else? 1069 00:44:51,140 --> 00:44:53,195 Make sense? 1070 00:44:53,195 --> 00:44:55,480 OK, so these are very, very precious data. 1071 00:45:00,340 --> 00:45:04,870 OK, nonetheless, the electrodes in this case 1072 00:45:04,870 --> 00:45:08,710 are about 2 millimeters across, each electrode. 1073 00:45:08,710 --> 00:45:12,610 And that is about the size of a functional MRI 1074 00:45:12,610 --> 00:45:17,770 pixel or voxel, a little bit smaller. 1075 00:45:17,770 --> 00:45:21,400 It has less blurring because functional MRI blurs spatially 1076 00:45:21,400 --> 00:45:23,180 because it's looking at blood flow. 1077 00:45:23,180 --> 00:45:25,660 So this is a more precise spatial measurement 1078 00:45:25,660 --> 00:45:28,030 than functional MRI, but it is still 1079 00:45:28,030 --> 00:45:31,240 averaging over probably tens of thousands of neurons, 1080 00:45:31,240 --> 00:45:35,470 down from hundreds of thousands of neurons with functional MRI. 1081 00:45:35,470 --> 00:45:38,950 So can we ever get responses from individual neurons 1082 00:45:38,950 --> 00:45:40,930 in the human brain? 1083 00:45:40,930 --> 00:45:43,300 Yes, occasionally. 1084 00:45:43,300 --> 00:45:47,590 In fact, a paper came out on the bioRxiv a couple of months ago. 1085 00:45:47,590 --> 00:45:50,500 I was on this guy's PhD thesis defense. 1086 00:45:50,500 --> 00:45:54,700 And this is a guy who works with a neurosurgeon on Long Island. 1087 00:45:54,700 --> 00:45:57,130 And this neurosurgeon specializes 1088 00:45:57,130 --> 00:45:59,770 in epilepsy neurosurgery. 1089 00:45:59,770 --> 00:46:04,570 And he's very interested in not damaging people's ability 1090 00:46:04,570 --> 00:46:06,100 to recognize faces. 1091 00:46:06,100 --> 00:46:11,410 And so he sticks electrodes to map out neural activity 1092 00:46:11,410 --> 00:46:13,510 and to discover seizure foci. 1093 00:46:13,510 --> 00:46:15,610 Before the neurosurgery, he sticks 1094 00:46:15,610 --> 00:46:20,410 electrodes in parts of the brain near the fusiform face area. 1095 00:46:20,410 --> 00:46:22,810 So this is a slice like this through the brain. 1096 00:46:22,810 --> 00:46:26,080 I showed you before horizontal slices, OK, so left and right 1097 00:46:26,080 --> 00:46:29,740 are flipped, that region is right in there, everybody 1098 00:46:29,740 --> 00:46:31,820 oriented with this picture here? 1099 00:46:31,820 --> 00:46:34,950 So this is an MRI image of this person. 1100 00:46:34,950 --> 00:46:36,460 It was scanned with functional MRI 1101 00:46:36,460 --> 00:46:38,050 before the electrodes were put in. 1102 00:46:38,050 --> 00:46:42,100 And that shows you their fusiform face area right there. 1103 00:46:42,100 --> 00:46:44,680 So now, the neurosurgeons put in electrodes 1104 00:46:44,680 --> 00:46:48,340 for clinical reasons, but the electrodes this surgeon uses 1105 00:46:48,340 --> 00:46:50,080 have these little tiny micro wires 1106 00:46:50,080 --> 00:46:52,210 that come out of the tip of the electrode that 1107 00:46:52,210 --> 00:46:56,560 enable him to record from individual neurons. 1108 00:46:56,560 --> 00:46:59,170 And so these guys, for the first time, 1109 00:46:59,170 --> 00:47:01,900 have recorded from individual neurons in the fusiform face 1110 00:47:01,900 --> 00:47:03,370 area in humans. 1111 00:47:03,370 --> 00:47:06,770 And here's an example of one of these neurons. 1112 00:47:06,770 --> 00:47:09,220 So here are the different stimuli here. 1113 00:47:09,220 --> 00:47:12,130 A bunch of different face stimuli, body stimuli, houses, 1114 00:47:12,130 --> 00:47:13,630 patterns, and tools. 1115 00:47:13,630 --> 00:47:16,690 And this shows you time across here. 1116 00:47:16,690 --> 00:47:18,598 Each one of those dots is-- 1117 00:47:18,598 --> 00:47:20,890 this is all the response of a single neuron that's been 1118 00:47:20,890 --> 00:47:23,020 identified in a human brain. 1119 00:47:23,020 --> 00:47:27,230 Each dot is an action potential, is a spike out of that neuron. 1120 00:47:27,230 --> 00:47:29,410 So you can see them happening over time here 1121 00:47:29,410 --> 00:47:30,702 to all the faces. 1122 00:47:30,702 --> 00:47:32,410 And this is an average amount of activity 1123 00:47:32,410 --> 00:47:35,320 to all of the faces and average amount of activity 1124 00:47:35,320 --> 00:47:38,050 to all the other stimuli. 1125 00:47:38,050 --> 00:47:39,310 Make sense? 1126 00:47:39,310 --> 00:47:41,080 So that's pretty breathtaking to me 1127 00:47:41,080 --> 00:47:43,630 because I've been using these very indirect methods 1128 00:47:43,630 --> 00:47:46,720 for a long time, inferring that they must result 1129 00:47:46,720 --> 00:47:49,120 from the average across a lot of neurons doing that, 1130 00:47:49,120 --> 00:47:50,890 but it's pretty awesome to actually see 1131 00:47:50,890 --> 00:47:52,900 individual neurons doing that. 1132 00:47:52,900 --> 00:47:53,980 Yeah? 1133 00:47:53,980 --> 00:47:55,300 OK. 1134 00:47:55,300 --> 00:47:57,880 Here's the time course of responses just averaging 1135 00:47:57,880 --> 00:47:59,800 over this raster over time, showing you 1136 00:47:59,800 --> 00:48:04,180 a similar time course to what I've shown before. 1137 00:48:04,180 --> 00:48:07,460 And in this guy's thesis, he found three other face 1138 00:48:07,460 --> 00:48:10,720 selective neurons in the FFA, but the electrodes 1139 00:48:10,720 --> 00:48:12,400 are so rarely in the right location 1140 00:48:12,400 --> 00:48:14,428 that they only have a few in this whole thesis, 1141 00:48:14,428 --> 00:48:15,220 and there they are. 1142 00:48:15,220 --> 00:48:16,680 Yeah? 1143 00:48:16,680 --> 00:48:19,100 AUDIENCE: Even if we could measure individual neurons, 1144 00:48:19,100 --> 00:48:21,970 we don't really know which neuron it is, right? 1145 00:48:21,970 --> 00:48:24,800 If I wanted to go back and find the same neuron Again, 1146 00:48:24,800 --> 00:48:26,050 That's pretty much impossible. 1147 00:48:26,050 --> 00:48:27,175 NANCY KANWISHER: Forget it. 1148 00:48:27,175 --> 00:48:27,742 Yep. 1149 00:48:27,742 --> 00:48:29,200 Yep. 1150 00:48:29,200 --> 00:48:31,210 So people like me who almost never 1151 00:48:31,210 --> 00:48:33,550 get to see responses from individual neurons 1152 00:48:33,550 --> 00:48:35,713 in human brains have kind of neuron envy. 1153 00:48:35,713 --> 00:48:37,630 It's like everyone else in this building has-- 1154 00:48:37,630 --> 00:48:40,900 they can measure stuff from dendrites or ion 1155 00:48:40,900 --> 00:48:42,280 channels or individual neurons. 1156 00:48:42,280 --> 00:48:43,990 They can do all this amazing stuff. 1157 00:48:43,990 --> 00:48:47,200 But actually, there are a lot of limitations in those methods 1158 00:48:47,200 --> 00:48:47,837 too. 1159 00:48:47,837 --> 00:48:49,670 And you just put your finger on one of them. 1160 00:48:49,670 --> 00:48:51,370 So they're like, OK they found those neurons, 1161 00:48:51,370 --> 00:48:52,450 there are four neurons. 1162 00:48:52,450 --> 00:48:54,490 We can't go back and find those neurons again. 1163 00:48:54,490 --> 00:48:55,690 That's that, right? 1164 00:48:55,690 --> 00:48:58,730 And they're probably subtly different in different brains, 1165 00:48:58,730 --> 00:48:59,230 right? 1166 00:48:59,230 --> 00:49:04,900 So it's cool and powerful but has still has many limitations. 1167 00:49:04,900 --> 00:49:09,970 OK, does this tell us that these neurons are involved 1168 00:49:09,970 --> 00:49:11,950 in discriminating one face from another 1169 00:49:11,950 --> 00:49:13,870 or just detecting faces? 1170 00:49:13,870 --> 00:49:17,020 Can we tell from these data? 1171 00:49:17,020 --> 00:49:18,850 Are they just saying, here's a face 1172 00:49:18,850 --> 00:49:21,930 or are they saying, that's Joe? 1173 00:49:21,930 --> 00:49:23,940 AUDIENCE: Did they have different conditions 1174 00:49:23,940 --> 00:49:25,645 for different people? 1175 00:49:25,645 --> 00:49:27,645 NANCY KANWISHER: These are different faces here. 1176 00:49:31,710 --> 00:49:32,610 What do you think? 1177 00:49:32,610 --> 00:49:33,870 What are these neurons doing? 1178 00:49:37,890 --> 00:49:38,580 Yeah? 1179 00:49:38,580 --> 00:49:43,610 AUDIENCE: They're just recognizing faces [INAUDIBLE].. 1180 00:49:43,610 --> 00:49:45,590 NANCY KANWISHER: You mean just detecting? 1181 00:49:45,590 --> 00:49:47,150 No, just say more. 1182 00:49:47,150 --> 00:49:48,800 What do you think they're doing? 1183 00:49:48,800 --> 00:49:51,338 AUDIENCE: They're just selecting for faces. 1184 00:49:51,338 --> 00:49:52,880 There's no evidence to show that they 1185 00:49:52,880 --> 00:49:54,907 distinguished different faces. 1186 00:49:54,907 --> 00:49:56,490 NANCY KANWISHER: Well, how about this? 1187 00:49:56,490 --> 00:49:59,220 These are different faces here. 1188 00:49:59,220 --> 00:50:02,810 These are different faces here. 1189 00:50:02,810 --> 00:50:05,640 AUDIENCE: But one could ask, if it does involve them sort 1190 00:50:05,640 --> 00:50:08,730 of acknowledging what faces, did they 1191 00:50:08,730 --> 00:50:10,050 have to put a name to the face? 1192 00:50:10,050 --> 00:50:11,508 NANCY KANWISHER: Nope, they're just 1193 00:50:11,508 --> 00:50:14,040 sitting there looking at stuff. 1194 00:50:14,040 --> 00:50:16,560 So bottom line is, we don't know from this. 1195 00:50:16,560 --> 00:50:18,930 It could be just responding and saying essentially, 1196 00:50:18,930 --> 00:50:20,190 there's a face. 1197 00:50:20,190 --> 00:50:22,350 But the fact that there's different responses 1198 00:50:22,350 --> 00:50:25,290 to different faces suggests that maybe there's 1199 00:50:25,290 --> 00:50:27,210 some information in there. 1200 00:50:27,210 --> 00:50:30,420 If you ran some machine learning code on this, 1201 00:50:30,420 --> 00:50:33,120 you could tell a little bit, which face was being presented. 1202 00:50:33,120 --> 00:50:35,120 Because those neurons are responding differently 1203 00:50:35,120 --> 00:50:36,630 to different faces. 1204 00:50:36,630 --> 00:50:38,010 Yeah? 1205 00:50:38,010 --> 00:50:40,170 AUDIENCE: Is it really like if they just 1206 00:50:40,170 --> 00:50:42,280 showed the same face repeatedly, wouldn't it just 1207 00:50:42,280 --> 00:50:43,710 be like [INAUDIBLE]? 1208 00:50:43,710 --> 00:50:45,570 NANCY KANWISHER: OK, very good question. 1209 00:50:45,570 --> 00:50:46,600 Very good question. 1210 00:50:46,600 --> 00:50:48,420 That's why I said suggest, right? 1211 00:50:48,420 --> 00:50:49,560 You're absolutely right. 1212 00:50:49,560 --> 00:50:50,615 That could be just noise. 1213 00:50:50,615 --> 00:50:51,990 It could be that if you presented 1214 00:50:51,990 --> 00:50:54,407 the same face every time you'd get that same distribution. 1215 00:50:54,407 --> 00:50:55,560 You're exactly right. 1216 00:50:55,560 --> 00:50:56,820 And so we will talk-- 1217 00:50:56,820 --> 00:50:59,080 not next time, I think Wednesday next week. 1218 00:50:59,080 --> 00:51:00,720 But anyways, very soon we'll talk 1219 00:51:00,720 --> 00:51:03,660 about methods that enable us to exactly deal with that question 1220 00:51:03,660 --> 00:51:06,480 and ask, is there actually information 1221 00:51:06,480 --> 00:51:09,780 in this pattern of response across neurons or voxels 1222 00:51:09,780 --> 00:51:11,530 or whatever it is? 1223 00:51:11,530 --> 00:51:15,240 Or is that just the noise of variation? 1224 00:51:15,240 --> 00:51:15,960 Yeah? 1225 00:51:15,960 --> 00:51:16,770 OK. 1226 00:51:16,770 --> 00:51:19,530 AUDIENCE: But how many neurons are in [INAUDIBLE]?? 1227 00:51:19,530 --> 00:51:21,820 NANCY KANWISHER: Oh good question. 1228 00:51:21,820 --> 00:51:23,370 Let's see. 1229 00:51:23,370 --> 00:51:25,730 I would say, I think a few million. 1230 00:51:25,730 --> 00:51:26,730 So let's think about it. 1231 00:51:26,730 --> 00:51:31,800 Each voxel is about a half a million, 1232 00:51:31,800 --> 00:51:35,440 and they are typically maybe like 30 voxels, something 1233 00:51:35,440 --> 00:51:35,940 like that. 1234 00:51:35,940 --> 00:51:38,760 Somewhere on the order of 20 million, something like that. 1235 00:51:38,760 --> 00:51:40,260 I mean, with huge error bars. 1236 00:51:43,060 --> 00:51:47,200 OK, so this is cool and tantalizing, 1237 00:51:47,200 --> 00:51:49,818 but it doesn't even tell us what these neurons-- what exactly 1238 00:51:49,818 --> 00:51:50,860 they're participating in. 1239 00:51:50,860 --> 00:51:53,350 It doesn't tell us if those neurons are telling 1240 00:51:53,350 --> 00:51:57,520 that person which face is there or maybe what facial expression 1241 00:51:57,520 --> 00:51:59,135 the person has or how old they are 1242 00:51:59,135 --> 00:52:01,510 or whether they're male or female or God knows what else, 1243 00:52:01,510 --> 00:52:02,710 right? 1244 00:52:02,710 --> 00:52:05,110 And it certainly doesn't tell us how those neurons 1245 00:52:05,110 --> 00:52:07,270 get that information. 1246 00:52:07,270 --> 00:52:09,010 Still, it's cool. 1247 00:52:09,010 --> 00:52:12,790 OK, so intracranial recording, both with the grids 1248 00:52:12,790 --> 00:52:15,250 that I showed you and the single unit version. 1249 00:52:15,250 --> 00:52:17,800 Advantages are, this is the only method 1250 00:52:17,800 --> 00:52:20,560 in humans that has both pretty good spatial resolution 1251 00:52:20,560 --> 00:52:24,550 and temporal resolution at the same time. 1252 00:52:24,550 --> 00:52:28,210 Disadvantage-- well, you need to have a craniotomy, which 1253 00:52:28,210 --> 00:52:30,260 is no picnic, to put it mildly. 1254 00:52:30,260 --> 00:52:32,500 You need to have a huge piece of your skull removed 1255 00:52:32,500 --> 00:52:33,933 and neurosurgery. 1256 00:52:33,933 --> 00:52:35,350 And that means that the only times 1257 00:52:35,350 --> 00:52:38,500 we get to do this are when it's required clinically 1258 00:52:38,500 --> 00:52:40,930 and everything is under control of the doctors, 1259 00:52:40,930 --> 00:52:42,250 as it should be. 1260 00:52:42,250 --> 00:52:43,930 So the doctors make all the choices 1261 00:52:43,930 --> 00:52:46,792 about where the electrodes go, and we just 1262 00:52:46,792 --> 00:52:49,000 get to sit in the background and say, please, please, 1263 00:52:49,000 --> 00:52:52,570 please, look at these stimuli, but try not to hassle 1264 00:52:52,570 --> 00:52:54,190 the patients too much. 1265 00:52:54,190 --> 00:52:55,970 Right now there's a patient in Albany, 1266 00:52:55,970 --> 00:52:57,910 New York who has electrodes right 1267 00:52:57,910 --> 00:52:59,800 over a really exciting part of the brain 1268 00:52:59,800 --> 00:53:03,040 to us that I'll talk about in a few months. 1269 00:53:03,040 --> 00:53:06,880 This patient has electrodes that respond specifically to music. 1270 00:53:06,880 --> 00:53:08,270 We will talk about that later. 1271 00:53:08,270 --> 00:53:09,950 It's pretty amazing. 1272 00:53:09,950 --> 00:53:12,460 And for the last couple of days, Dana and I 1273 00:53:12,460 --> 00:53:15,010 have been-- mostly Dana has been collecting stimuli 1274 00:53:15,010 --> 00:53:17,110 because we really want to ask questions 1275 00:53:17,110 --> 00:53:19,060 about the response of those electrodes. 1276 00:53:19,060 --> 00:53:20,560 And this patient is not too thrilled 1277 00:53:20,560 --> 00:53:21,950 listening to our stimuli. 1278 00:53:21,950 --> 00:53:24,490 So we finally said, oh, OK, tell the patient 1279 00:53:24,490 --> 00:53:26,230 they can just do Instagram on their phone 1280 00:53:26,230 --> 00:53:28,105 and we'll play the stimuli in the background. 1281 00:53:28,105 --> 00:53:32,350 So hopefully, we'll have cool data from that soon. 1282 00:53:32,350 --> 00:53:36,760 OK, so to say that these data are limited and hard to control 1283 00:53:36,760 --> 00:53:37,810 is an understatement. 1284 00:53:37,810 --> 00:53:39,610 We basically can't control it at all. 1285 00:53:39,610 --> 00:53:41,950 All we can control occasionally is the stimuli. 1286 00:53:45,700 --> 00:53:47,740 And it also, like functional MRI, 1287 00:53:47,740 --> 00:53:49,750 just because we see those beautiful responses, 1288 00:53:49,750 --> 00:53:52,930 it doesn't tell us how those responses 1289 00:53:52,930 --> 00:53:55,810 are connected to behavior. 1290 00:53:55,810 --> 00:53:58,520 So that's a real challenge. 1291 00:53:58,520 --> 00:54:00,430 So that won't do. 1292 00:54:00,430 --> 00:54:02,350 We need to get beyond this problem. 1293 00:54:02,350 --> 00:54:03,940 I keep saying this method is great, 1294 00:54:03,940 --> 00:54:05,860 but it doesn't tell us the causal role 1295 00:54:05,860 --> 00:54:10,210 of that neural phenomenon in cognition and behavior. 1296 00:54:10,210 --> 00:54:12,910 As scientists, science is all about discovering 1297 00:54:12,910 --> 00:54:13,973 causal mechanisms. 1298 00:54:13,973 --> 00:54:16,390 We're not just interested in what is correlated with what, 1299 00:54:16,390 --> 00:54:17,740 we want to know what's causing what. 1300 00:54:17,740 --> 00:54:19,480 That's really of the essence, and so we 1301 00:54:19,480 --> 00:54:21,620 need to do better here. 1302 00:54:21,620 --> 00:54:22,780 So what are we going to do? 1303 00:54:22,780 --> 00:54:24,700 Somebody mentioned a while ago, maybe it 1304 00:54:24,700 --> 00:54:26,590 was Isabelle, that one of the ways 1305 00:54:26,590 --> 00:54:30,130 to do that and ask whether the face area is causally involved 1306 00:54:30,130 --> 00:54:33,370 in face perception is to look at a case 1307 00:54:33,370 --> 00:54:35,680 where the face area is altered. 1308 00:54:35,680 --> 00:54:38,810 So there's a bunch of ways to do that. 1309 00:54:38,810 --> 00:54:40,510 And one of them-- 1310 00:54:40,510 --> 00:54:41,810 OK, that's just a review. 1311 00:54:41,810 --> 00:54:43,360 We said faces are recognized fast 1312 00:54:43,360 --> 00:54:46,360 but we haven't learned much more. 1313 00:54:46,360 --> 00:54:48,610 How do we test causality? 1314 00:54:48,610 --> 00:54:52,090 OK, patients with focal brain damage. 1315 00:54:52,090 --> 00:54:53,450 Here is a patient. 1316 00:54:53,450 --> 00:54:56,290 These are vertical slices through the back 1317 00:54:56,290 --> 00:54:57,380 of this patient's head. 1318 00:54:57,380 --> 00:54:58,380 OK, let me get oriented. 1319 00:54:58,380 --> 00:54:59,995 The slice is maybe this here. 1320 00:54:59,995 --> 00:55:01,870 And as you go rightward, you're marching back 1321 00:55:01,870 --> 00:55:03,250 in the brain like that. 1322 00:55:03,250 --> 00:55:04,990 Everybody oriented? 1323 00:55:04,990 --> 00:55:06,848 What's this thing right there? 1324 00:55:06,848 --> 00:55:08,640 AUDIENCE: Cerebellum. 1325 00:55:08,640 --> 00:55:10,620 NANCY KANWISHER: Yeah, cerebellum, right. 1326 00:55:10,620 --> 00:55:13,230 That thing right there is this patient's lesion 1327 00:55:13,230 --> 00:55:16,560 that spans several slices going back like that. 1328 00:55:16,560 --> 00:55:21,600 And this patient's lesion looks a whole lot like my FFA. 1329 00:55:21,600 --> 00:55:25,830 There's my FFA, greater response to faces than objects, 1330 00:55:25,830 --> 00:55:27,630 on similar slices. 1331 00:55:27,630 --> 00:55:30,130 We don't have functional MRI from this patient 1332 00:55:30,130 --> 00:55:32,850 so we don't know exactly where this guy's FFA was. 1333 00:55:32,850 --> 00:55:36,660 But there's a good bet that it was blitzed by that lesion, 1334 00:55:36,660 --> 00:55:39,690 because it's right in the zone where it usually lands. 1335 00:55:39,690 --> 00:55:45,330 And this patient can't recognize faces at all. 1336 00:55:45,330 --> 00:55:48,210 And importantly, the patient is absolutely 1337 00:55:48,210 --> 00:55:50,550 normal at recognizing objects. 1338 00:55:50,550 --> 00:55:53,145 No problem whatsoever at recognizing objects. 1339 00:55:55,740 --> 00:55:57,660 How does this take us beyond functional MRI? 1340 00:56:00,520 --> 00:56:01,020 Yeah? 1341 00:56:01,020 --> 00:56:03,537 AUDIENCE: It implies causation. 1342 00:56:03,537 --> 00:56:04,620 NANCY KANWISHER: Speak up. 1343 00:56:04,620 --> 00:56:05,912 AUDIENCE: It implies causation. 1344 00:56:05,912 --> 00:56:07,320 NANCY KANWISHER: Yeah, say more. 1345 00:56:07,320 --> 00:56:08,220 What does it tell us? 1346 00:56:08,220 --> 00:56:11,640 AUDIENCE: So because of the fact that area's damaged 1347 00:56:11,640 --> 00:56:14,625 and then it makes it to not be able to recognize faces 1348 00:56:14,625 --> 00:56:16,500 and I can see that there's causality that oh, 1349 00:56:16,500 --> 00:56:17,430 that area's [INAUDIBLE]. 1350 00:56:17,430 --> 00:56:18,540 NANCY KANWISHER: Exactly. 1351 00:56:18,540 --> 00:56:19,040 Exactly. 1352 00:56:19,040 --> 00:56:21,460 It says you need that bit to recognize faces. 1353 00:56:21,460 --> 00:56:23,700 But also says something else. 1354 00:56:23,700 --> 00:56:24,450 What else does it? 1355 00:56:24,450 --> 00:56:26,490 AUDIENCE: That you don't need it for recognizing objects. 1356 00:56:26,490 --> 00:56:28,210 NANCY KANWISHER: You don't need it for recognizing objects. 1357 00:56:28,210 --> 00:56:30,240 So this is actually really strong evidence 1358 00:56:30,240 --> 00:56:34,800 that that bit of brain is very specialized for face 1359 00:56:34,800 --> 00:56:36,360 recognition. 1360 00:56:36,360 --> 00:56:39,155 Specialized and necessary for face recognition. 1361 00:56:41,820 --> 00:56:43,125 OK, so-- 1362 00:56:46,770 --> 00:56:49,230 AUDIENCE: Can that person still detect faces? 1363 00:56:49,230 --> 00:56:50,272 NANCY KANWISHER: Oh, yes. 1364 00:56:50,272 --> 00:56:51,780 Good question, absolutely. 1365 00:56:51,780 --> 00:56:55,860 OK, so let me just distinguish-- this person here has 1366 00:56:55,860 --> 00:56:59,100 prosopagnosia-- that means a selective deficit in face 1367 00:56:59,100 --> 00:57:00,180 recognition-- 1368 00:57:00,180 --> 00:57:03,300 like Jacob Hodes, who I described yesterday, 1369 00:57:03,300 --> 00:57:05,210 who has no brain damage whatsoever 1370 00:57:05,210 --> 00:57:07,710 but has just never been able to recognize faces at any point 1371 00:57:07,710 --> 00:57:08,800 in his life. 1372 00:57:08,800 --> 00:57:11,910 So this syndrome can arise just from some weird developmental 1373 00:57:11,910 --> 00:57:14,850 thing where you're atypical and you're just really bad at it, 1374 00:57:14,850 --> 00:57:17,940 or it can result from damage to that part of the brain. 1375 00:57:17,940 --> 00:57:21,300 So now we're talking about the case of damage, 1376 00:57:21,300 --> 00:57:24,600 but in both cases, people with prosopagnosia 1377 00:57:24,600 --> 00:57:27,150 have no problem knowing that a face is a face. 1378 00:57:27,150 --> 00:57:29,880 They just don't know who it is. 1379 00:57:29,880 --> 00:57:30,922 Yeah? 1380 00:57:30,922 --> 00:57:32,940 AUDIENCE: Has there ever been a case 1381 00:57:32,940 --> 00:57:37,100 of problems of people who can't recognize faces who 1382 00:57:37,100 --> 00:57:38,090 [INAUDIBLE]? 1383 00:57:41,060 --> 00:57:42,350 NANCY KANWISHER: Indeed. 1384 00:57:42,350 --> 00:57:43,130 Indeed. 1385 00:57:43,130 --> 00:57:45,950 Jacob Hodes, who I talked about last time who is just 1386 00:57:45,950 --> 00:57:48,140 absolutely awful at face recognition, 1387 00:57:48,140 --> 00:57:50,420 including family members, close friends, 1388 00:57:50,420 --> 00:57:52,910 can't do it, like not at all. 1389 00:57:52,910 --> 00:57:56,750 He has a very normal looking fusiform face area. 1390 00:57:56,750 --> 00:57:59,360 So after I told you that I had that conversation with him 1391 00:57:59,360 --> 00:58:03,440 a dozen years ago or something like that, I scanned him. 1392 00:58:03,440 --> 00:58:06,980 And he had a beautiful fusiform face area, like textbook. 1393 00:58:06,980 --> 00:58:11,000 It looked-- well, looked like mine, which is a damn fine one 1394 00:58:11,000 --> 00:58:13,470 if I do say so myself. 1395 00:58:13,470 --> 00:58:17,360 And I looked at that and I went, oh shit. 1396 00:58:17,360 --> 00:58:19,610 I better publish this before someone else does. 1397 00:58:19,610 --> 00:58:21,120 And I didn't get my act together, 1398 00:58:21,120 --> 00:58:22,610 and then a whole bunch of papers came out saying, 1399 00:58:22,610 --> 00:58:24,620 oh, people with developmental prosopagnosia 1400 00:58:24,620 --> 00:58:26,120 have normal looking face areas. 1401 00:58:26,120 --> 00:58:27,050 Take that, Kanwisher. 1402 00:58:27,050 --> 00:58:30,360 What do you say about that? 1403 00:58:30,360 --> 00:58:32,780 And it was a little shocking. 1404 00:58:32,780 --> 00:58:35,400 But upon further reflection, it's not really devastating, 1405 00:58:35,400 --> 00:58:35,900 right? 1406 00:58:35,900 --> 00:58:38,660 I mean, it's bracing, it's informative. 1407 00:58:38,660 --> 00:58:42,560 But it tells you that having a face area that 1408 00:58:42,560 --> 00:58:45,500 is a region that responds more to faces and objects 1409 00:58:45,500 --> 00:58:49,610 isn't sufficient for normal face recognition, right? 1410 00:58:49,610 --> 00:58:51,630 You need other stuff. 1411 00:58:51,630 --> 00:58:52,950 What might that other stuff be? 1412 00:58:52,950 --> 00:58:55,670 Well, the circuits in there need to work right. 1413 00:58:55,670 --> 00:58:58,550 It's not enough to just respond more to faces and objects. 1414 00:58:58,550 --> 00:59:01,190 To recognize faces, they need to be able to distinguish faces 1415 00:59:01,190 --> 00:59:01,760 from objects. 1416 00:59:01,760 --> 00:59:03,740 We don't know if that's working right. 1417 00:59:03,740 --> 00:59:05,052 What else do you need? 1418 00:59:05,052 --> 00:59:06,200 AUDIENCE: Memory. 1419 00:59:06,200 --> 00:59:09,110 NANCY KANWISHER: Memory, absolutely. 1420 00:59:09,110 --> 00:59:11,180 Yes, you need to remember faces. 1421 00:59:11,180 --> 00:59:12,044 What else? 1422 00:59:12,044 --> 00:59:23,200 AUDIENCE: [INAUDIBLE] 1423 00:59:23,200 --> 00:59:25,420 NANCY KANWISHER: Could be, but in Jacob's case, 1424 00:59:25,420 --> 00:59:28,113 it was close friends he couldn't recognize. 1425 00:59:28,113 --> 00:59:29,530 So what's another possible account 1426 00:59:29,530 --> 00:59:32,350 of how could he have a normal face area and yeah? 1427 00:59:32,350 --> 00:59:32,850 David? 1428 00:59:32,850 --> 00:59:37,930 AUDIENCE: It might be a gap between recognizing a face 1429 00:59:37,930 --> 00:59:41,133 and connecting that to recognizing a person. 1430 00:59:41,133 --> 00:59:42,050 NANCY KANWISHER: Yeah. 1431 00:59:42,050 --> 00:59:42,550 Yeah. 1432 00:59:42,550 --> 00:59:44,430 Or to put that neuroanatomically, 1433 00:59:44,430 --> 00:59:47,190 you got to get the information out of there. 1434 00:59:47,190 --> 00:59:50,550 maybe, for all we know, that little face area 1435 00:59:50,550 --> 00:59:51,480 is working perfectly. 1436 00:59:51,480 --> 00:59:54,840 Maybe that face area knows who that person is, in a sense. 1437 00:59:54,840 --> 00:59:56,730 But if the connection's out of that brain 1438 00:59:56,730 --> 00:59:59,970 region to the rest of the brain are messed up, 1439 00:59:59,970 --> 01:00:01,957 it doesn't do you any good. 1440 01:00:01,957 --> 01:00:04,290 You need to be able to read that information out and act 1441 01:00:04,290 --> 01:00:06,390 on the basis of it. 1442 01:00:06,390 --> 01:00:08,310 Anyway, that's a big sidebar. 1443 01:00:08,310 --> 01:00:10,110 Point is, you can have prosopagnosia 1444 01:00:10,110 --> 01:00:13,860 either as just a developmental disorder 1445 01:00:13,860 --> 01:00:16,172 or as a result of brain damage. 1446 01:00:16,172 --> 01:00:17,880 Oh, God, I knew this was going to happen. 1447 01:00:17,880 --> 01:00:24,300 All right, so OK, so very briefly, it messes up ability 1448 01:00:24,300 --> 01:00:26,970 to discriminate and recognize faces, not your ability 1449 01:00:26,970 --> 01:00:28,080 to detect a face, right? 1450 01:00:28,080 --> 01:00:29,455 So as [INAUDIBLE] had asked, it's 1451 01:00:29,455 --> 01:00:31,682 not just you can't tell the thing is a face, 1452 01:00:31,682 --> 01:00:32,640 they're fine with that. 1453 01:00:32,640 --> 01:00:36,330 Importantly, they are normal and voice recognition. 1454 01:00:36,330 --> 01:00:37,830 So it's not that they're confused 1455 01:00:37,830 --> 01:00:40,150 about distinguishing one person from another. 1456 01:00:40,150 --> 01:00:43,905 They can do it fine from audition, just not from vision. 1457 01:00:46,740 --> 01:00:48,900 In the rare cases where the lesion is small, 1458 01:00:48,900 --> 01:00:52,650 it can be very specific, leaving object recognition intact. 1459 01:00:52,650 --> 01:00:54,480 More often, there's kind of a blurry mess. 1460 01:00:54,480 --> 01:00:57,930 You have a big lesion and a bunch of things are affected. 1461 01:00:57,930 --> 01:00:59,790 OK, so we've talked about that. 1462 01:00:59,790 --> 01:01:05,250 OK now, it's very important in neuropsychology reasoning-- 1463 01:01:05,250 --> 01:01:07,320 like we want to say, OK, that's really powerful, 1464 01:01:07,320 --> 01:01:08,610 the case of prosopagnosia. 1465 01:01:08,610 --> 01:01:12,900 You lose that bit, you can't recognize faces. 1466 01:01:12,900 --> 01:01:15,990 And that establishes a kind of causality 1467 01:01:15,990 --> 01:01:19,290 that we didn't have before with just functional MRI. 1468 01:01:19,290 --> 01:01:22,950 But is that sufficient to say that that region is specialized 1469 01:01:22,950 --> 01:01:24,390 for face recognition only? 1470 01:01:29,960 --> 01:01:30,560 It's not. 1471 01:01:30,560 --> 01:01:32,400 Whenever I ask this, the answer's no. 1472 01:01:32,400 --> 01:01:35,270 Your task is to say, why? 1473 01:01:35,270 --> 01:01:39,140 How could you have great difficulty at recognizing faces 1474 01:01:39,140 --> 01:01:40,880 and be OK at object recognition? 1475 01:01:40,880 --> 01:01:44,150 And yet, not have a deficit that's specific to faces? 1476 01:01:44,150 --> 01:01:47,030 How might that arise? 1477 01:01:47,030 --> 01:01:49,550 You guys have suggested this hypothesis in different context 1478 01:01:49,550 --> 01:01:50,050 before. 1479 01:01:57,120 --> 01:01:57,780 Yes? 1480 01:01:57,780 --> 01:01:58,738 You look like you know. 1481 01:01:58,738 --> 01:01:59,250 No? 1482 01:01:59,250 --> 01:02:01,910 AUDIENCE: But I just you could do other things. 1483 01:02:01,910 --> 01:02:06,980 It doesn't have to only be for facial recognition. 1484 01:02:06,980 --> 01:02:11,420 Because it response to animals [INAUDIBLE],, right? 1485 01:02:11,420 --> 01:02:13,010 NANCY KANWISHER: Sort of. 1486 01:02:13,010 --> 01:02:16,880 But the question here is-- 1487 01:02:16,880 --> 01:02:18,980 OK, let's just start bare bones. 1488 01:02:18,980 --> 01:02:22,490 You have a lesion, you get around fine in the world, 1489 01:02:22,490 --> 01:02:24,740 you can do everything else but you have a real problem 1490 01:02:24,740 --> 01:02:26,840 recognizing faces. 1491 01:02:26,840 --> 01:02:32,750 Does that mean that the region lesioned is specialized 1492 01:02:32,750 --> 01:02:35,405 for face recognition per se? 1493 01:02:35,405 --> 01:02:38,315 AUDIENCE: There might be other [INAUDIBLE].. 1494 01:02:41,132 --> 01:02:42,340 NANCY KANWISHER: That's true. 1495 01:02:42,340 --> 01:02:44,530 There could be other things going on, absolutely. 1496 01:02:44,530 --> 01:02:45,580 But let's suppose they're not. 1497 01:02:45,580 --> 01:02:47,290 Let's suppose you had good reason to think there weren't. 1498 01:02:47,290 --> 01:02:48,010 Yeah? 1499 01:02:48,010 --> 01:02:50,620 AUDIENCE: It could be-- 1500 01:02:50,620 --> 01:02:52,390 it'd be a path-- 1501 01:02:52,390 --> 01:02:54,280 it could be one point in a pathway. 1502 01:02:54,280 --> 01:02:54,850 NANCY KANWISHER: That's true. 1503 01:02:54,850 --> 01:02:56,470 It could be totally a point in a pathway. 1504 01:02:56,470 --> 01:02:57,928 Absolutely, that's another account. 1505 01:02:57,928 --> 01:02:59,416 What else? 1506 01:02:59,416 --> 01:03:01,720 AUDIENCE: Well, I couldn't get last comment, so. 1507 01:03:01,720 --> 01:03:03,887 NANCY KANWISHER: He said maybe you damage a pathway. 1508 01:03:06,740 --> 01:03:07,610 Yeah? 1509 01:03:07,610 --> 01:03:09,402 AUDIENCE: Maybe there's some other function 1510 01:03:09,402 --> 01:03:10,790 we haven't tested in that person. 1511 01:03:10,790 --> 01:03:11,900 NANCY KANWISHER: All these are very good 1512 01:03:11,900 --> 01:03:12,860 alternative hypotheses. 1513 01:03:12,860 --> 01:03:14,152 You guys are very good at this. 1514 01:03:14,152 --> 01:03:16,010 The one I'm fishing for is, maybe 1515 01:03:16,010 --> 01:03:19,670 face recognition is just harder than object recognition. 1516 01:03:19,670 --> 01:03:23,570 Maybe the part that's damaged is just generically involved 1517 01:03:23,570 --> 01:03:26,180 in object recognition, but you damage 1518 01:03:26,180 --> 01:03:28,940 part of the object recognition system and face recognition 1519 01:03:28,940 --> 01:03:31,820 takes a bigger hit because it's harder. 1520 01:03:31,820 --> 01:03:32,330 Right? 1521 01:03:32,330 --> 01:03:33,810 Does that make sense? 1522 01:03:33,810 --> 01:03:35,780 Do you see how the case of prosopagnosia 1523 01:03:35,780 --> 01:03:37,200 is consistent with that? 1524 01:03:37,200 --> 01:03:40,910 So that means we cannot infer from these data alone that that 1525 01:03:40,910 --> 01:03:43,490 region's specialized for face recognition. 1526 01:03:43,490 --> 01:03:45,380 Now, we can do various things like test 1527 01:03:45,380 --> 01:03:48,470 them on really hard versions of object recognition. 1528 01:03:48,470 --> 01:03:49,940 And people have done that. 1529 01:03:49,940 --> 01:03:53,570 But there's another kind of data that are really powerful here. 1530 01:03:53,570 --> 01:03:57,500 And that's when we have the opposite syndrome. 1531 01:03:57,500 --> 01:04:00,560 So there's only a couple of cases of this. 1532 01:04:00,560 --> 01:04:04,177 The best one is called CK, published in a paper in 1997. 1533 01:04:04,177 --> 01:04:05,510 You don't need to remember that. 1534 01:04:05,510 --> 01:04:07,490 The point about this is that this guy 1535 01:04:07,490 --> 01:04:09,630 has the opposite syndrome. 1536 01:04:09,630 --> 01:04:12,290 He's severely impaired at object recognition. 1537 01:04:12,290 --> 01:04:14,390 He can't tell a chair from a table 1538 01:04:14,390 --> 01:04:16,910 from a car from a toaster, but he's 1539 01:04:16,910 --> 01:04:19,790 100% normal at face recognition. 1540 01:04:19,790 --> 01:04:21,800 Totally normal at face recognition. 1541 01:04:21,800 --> 01:04:23,030 In fact, better than average. 1542 01:04:25,833 --> 01:04:27,250 Do you see how that's in some ways 1543 01:04:27,250 --> 01:04:31,030 even more powerful evidence that face recognition goes 1544 01:04:31,030 --> 01:04:32,890 on in specialized brain machinery 1545 01:04:32,890 --> 01:04:35,500 than the case of prosopagnosia? 1546 01:04:35,500 --> 01:04:37,930 Face recognition isn't even a special thing 1547 01:04:37,930 --> 01:04:40,120 that sits on top of normal object recognition. 1548 01:04:40,120 --> 01:04:41,830 It's a totally different pathway. 1549 01:04:41,830 --> 01:04:44,020 You can have no ability to recognize objects 1550 01:04:44,020 --> 01:04:45,860 and your OK a face recognition. 1551 01:04:45,860 --> 01:04:48,010 Does everybody see how that's really powerful? 1552 01:04:48,010 --> 01:04:50,110 And how those two kinds of evidence 1553 01:04:50,110 --> 01:04:55,270 together are vastly more powerful than either one alone. 1554 01:04:55,270 --> 01:04:59,110 Well, that's called a double association. 1555 01:04:59,110 --> 01:05:00,610 We'll skip all of that for now. 1556 01:05:00,610 --> 01:05:03,400 Doubled associations are particularly powerful 1557 01:05:03,400 --> 01:05:05,230 examples-- 1558 01:05:05,230 --> 01:05:07,960 powerful forms of evidence in cognitive 1559 01:05:07,960 --> 01:05:10,810 neuroscience where we have opposite syndromes that 1560 01:05:10,810 --> 01:05:13,480 collectively make it really hard to wiggle out and come up 1561 01:05:13,480 --> 01:05:15,250 with alternative accounts other than 1562 01:05:15,250 --> 01:05:18,460 that there's a bit of brain that's really specialized 1563 01:05:18,460 --> 01:05:19,450 for face recognition. 1564 01:05:19,450 --> 01:05:21,730 It's not just that face recognition is harder, 1565 01:05:21,730 --> 01:05:25,540 or else you'd never get this syndrome. 1566 01:05:25,540 --> 01:05:27,550 All right, I just wanted to finish that point. 1567 01:05:27,550 --> 01:05:29,785 OK now, how much time do I have until the quiz? 1568 01:05:33,930 --> 01:05:35,423 15 minutes, OK good. 1569 01:05:35,423 --> 01:05:36,310 AUDIENCE: 13. 1570 01:05:36,310 --> 01:05:36,780 NANCY KANWISHER: OK, good. 1571 01:05:36,780 --> 01:05:38,190 We're going to skip over TMS. 1572 01:05:38,190 --> 01:05:39,060 I'm sorry about that, guys. 1573 01:05:39,060 --> 01:05:40,980 Someday I'll learn to time things in a lecture. 1574 01:05:40,980 --> 01:05:43,105 Actually, I knew this was going to happen, I just-- 1575 01:05:43,105 --> 01:05:44,880 we'll get back to TMS later. 1576 01:05:44,880 --> 01:05:49,020 And we will skip to the most amazing method 1577 01:05:49,020 --> 01:05:52,680 in all of cognitive neuroscience for which-- we're 1578 01:05:52,680 --> 01:05:55,530 going to come back to this dude who you met before who 1579 01:05:55,530 --> 01:05:59,100 has the face selective responses in that part of his brain. 1580 01:05:59,100 --> 01:06:01,710 Remember how I said that even though these data are 1581 01:06:01,710 --> 01:06:03,840 gorgeous and spectacular and the only way 1582 01:06:03,840 --> 01:06:06,300 we can get high spatial and temporal resolution together, 1583 01:06:06,300 --> 01:06:08,860 but they don't tell us causality? 1584 01:06:08,860 --> 01:06:09,360 Right? 1585 01:06:09,360 --> 01:06:11,040 That's true here? 1586 01:06:11,040 --> 01:06:12,990 Resolution doesn't get you causality. 1587 01:06:12,990 --> 01:06:14,790 To test the causal role of something, 1588 01:06:14,790 --> 01:06:17,670 you need to mess with it. 1589 01:06:17,670 --> 01:06:21,870 So it turns out that sometimes the neurosurgeons electrically 1590 01:06:21,870 --> 01:06:24,600 stimulate through those same electrodes. 1591 01:06:24,600 --> 01:06:28,260 And they do that to test the function of those regions 1592 01:06:28,260 --> 01:06:29,970 causally. 1593 01:06:29,970 --> 01:06:34,260 They also do it to test their hypotheses about the location 1594 01:06:34,260 --> 01:06:36,960 of the seizure foci. 1595 01:06:36,960 --> 01:06:39,090 So in those rare cases, where you 1596 01:06:39,090 --> 01:06:41,280 have a patient like this with selective electrodes 1597 01:06:41,280 --> 01:06:44,670 like that where the clinicians decide that they are going 1598 01:06:44,670 --> 01:06:46,350 to electrically stimulate through some 1599 01:06:46,350 --> 01:06:48,270 of those electrodes, then we're in a position 1600 01:06:48,270 --> 01:06:50,493 to kind of have it all scientifically, right? 1601 01:06:50,493 --> 01:06:51,660 I don't mean to be so crude. 1602 01:06:51,660 --> 01:06:54,360 This is a horrible situation for that lovely guy to be in, 1603 01:06:54,360 --> 01:06:57,220 but scientifically, it's extremely powerful. 1604 01:06:57,220 --> 01:07:00,570 So I'm going to show you-- 1605 01:07:00,570 --> 01:07:02,430 we did in fact have an opportunity. 1606 01:07:02,430 --> 01:07:04,980 The same guys in Japan emailed me 1607 01:07:04,980 --> 01:07:07,880 and said, OK, we're going to be stimulating that electrode. 1608 01:07:07,880 --> 01:07:10,230 What do we do? 1609 01:07:10,230 --> 01:07:12,750 And I said, OK, have him look at faces 1610 01:07:12,750 --> 01:07:14,310 and have them look at other objects 1611 01:07:14,310 --> 01:07:15,932 and ask him if anything changes. 1612 01:07:15,932 --> 01:07:17,640 And I'm going to show you a video of what 1613 01:07:17,640 --> 01:07:19,680 happens when that goes on. 1614 01:07:19,680 --> 01:07:23,040 OK, here we go. 1615 01:07:23,040 --> 01:07:24,570 Oh, and I need to turn on the audio. 1616 01:07:27,360 --> 01:07:31,320 OK, he's getting stimulated right there and he says-- 1617 01:07:31,320 --> 01:07:32,310 [VIDEO PLAYBACK] 1618 01:07:32,310 --> 01:07:36,270 - [NON-ENGLISH SPEECH] 1619 01:07:53,340 --> 01:07:56,250 NANCY KANWISHER: He's such a good subject, this guy. 1620 01:07:56,250 --> 01:07:57,188 - One more time. 1621 01:08:09,638 --> 01:08:12,550 - [NON-ENGLISH SPEECH] 1622 01:08:20,740 --> 01:08:23,210 - His eyes. 1623 01:08:23,210 --> 01:08:24,700 - [NON-ENGLISH SPEECH] 1624 01:08:30,833 --> 01:08:32,250 NANCY KANWISHER: OK, that tells us 1625 01:08:32,250 --> 01:08:34,890 that that region is causally involved in face perception. 1626 01:08:34,890 --> 01:08:36,660 Is it causally involved in perception 1627 01:08:36,660 --> 01:08:38,609 of things that aren't faces? 1628 01:08:38,609 --> 01:08:40,529 He's getting stimulated in the same electrode. 1629 01:08:40,529 --> 01:08:42,479 He doesn't know that there's a face area. 1630 01:08:42,479 --> 01:08:43,537 - [NON-ENGLISH SPEECH] 1631 01:08:43,537 --> 01:08:45,120 NANCY KANWISHER: He doesn't know which 1632 01:08:45,120 --> 01:08:47,855 electrode is being stimulated. 1633 01:08:47,855 --> 01:08:50,069 - [NON-ENGLISH SPEECH] 1634 01:09:26,703 --> 01:09:29,120 NANCY KANWISHER: This is a Kanji character on a card here. 1635 01:09:29,120 --> 01:09:32,599 - [NON-ENGLISH SPEECH] 1636 01:09:39,557 --> 01:09:41,048 - One more time. 1637 01:09:50,520 --> 01:09:52,611 - [NON-ENGLISH SPEECH] 1638 01:09:59,720 --> 01:10:00,630 [END PLAYBACK] 1639 01:10:00,630 --> 01:10:01,880 NANCY KANWISHER: Awesome, huh? 1640 01:10:01,880 --> 01:10:02,990 What did we just learn? 1641 01:10:05,750 --> 01:10:07,020 AUDIENCE: You can trigger it. 1642 01:10:07,020 --> 01:10:08,770 NANCY KANWISHER: You can trigger it, yeah. 1643 01:10:08,770 --> 01:10:09,560 Yeah. 1644 01:10:09,560 --> 01:10:12,060 So what does that tell us about the function of that region? 1645 01:10:19,480 --> 01:10:22,120 Why is this-- I mean, it's amazing to see, no question, 1646 01:10:22,120 --> 01:10:24,235 but what does it tell us scientifically? 1647 01:10:24,235 --> 01:10:26,200 AUDIENCE: It's specific. 1648 01:10:26,200 --> 01:10:27,190 NANCY KANWISHER: Yeah. 1649 01:10:27,190 --> 01:10:29,380 How does it tell us that it's specific? 1650 01:10:29,380 --> 01:10:31,900 AUDIENCE: Because when you stimulate it, 1651 01:10:31,900 --> 01:10:35,630 it particularly sees a face. 1652 01:10:35,630 --> 01:10:36,853 NANCY KANWISHER: Yeah. 1653 01:10:36,853 --> 01:10:38,270 And what happens when he's looking 1654 01:10:38,270 --> 01:10:42,090 at things that aren't faces? 1655 01:10:42,090 --> 01:10:43,520 AUDIENCE: [INAUDIBLE] 1656 01:10:43,520 --> 01:10:44,690 NANCY KANWISHER: Yeah. 1657 01:10:44,690 --> 01:10:47,900 So if that region was causally involved 1658 01:10:47,900 --> 01:10:50,240 in perception of things that aren't faces, 1659 01:10:50,240 --> 01:10:53,000 you might think that it would distort-- 1660 01:10:53,000 --> 01:10:55,670 the box would look different or the ball would look different 1661 01:10:55,670 --> 01:10:57,120 or the Kanji would look different. 1662 01:10:57,120 --> 01:10:59,510 It doesn't, there's just a face on top. 1663 01:10:59,510 --> 01:11:01,310 So I think that's very strong evidence 1664 01:11:01,310 --> 01:11:03,260 that that region is not only causally involved 1665 01:11:03,260 --> 01:11:07,160 in face perception, but very specifically causally involved 1666 01:11:07,160 --> 01:11:08,960 in face perception only. 1667 01:11:08,960 --> 01:11:10,400 Everybody get that? 1668 01:11:10,400 --> 01:11:11,810 Do I have to stop? 1669 01:11:11,810 --> 01:11:13,160 OK. 1670 01:11:13,160 --> 01:11:14,900 OK, I have another video. 1671 01:11:14,900 --> 01:11:17,660 Consider-- and we'll get back to this later-- 1672 01:11:17,660 --> 01:11:22,475 consider other alternative hypotheses to this. 1673 01:11:22,475 --> 01:11:23,905 This is pretty powerful. 1674 01:11:23,905 --> 01:11:26,030 This is more powerful than most of the other things 1675 01:11:26,030 --> 01:11:28,245 I showed you, but there's always ways 1676 01:11:28,245 --> 01:11:29,870 to come up with alternative hypotheses, 1677 01:11:29,870 --> 01:11:31,740 and that's the business we're in here. 1678 01:11:31,740 --> 01:11:34,070 So be percolating on what other control 1679 01:11:34,070 --> 01:11:36,680 conditions you'd want from this guy 1680 01:11:36,680 --> 01:11:39,490 to really believe these data.