1 00:00:00,000 --> 00:00:03,848 [DIGITAL EFFECTS] 2 00:00:09,620 --> 00:00:11,280 NANCY KANWISHER: So to remind you, 3 00:00:11,280 --> 00:00:15,090 we've been talking last week about doing two things 4 00:00:15,090 --> 00:00:18,137 at once-- asking all sort of questions of what we might want 5 00:00:18,137 --> 00:00:20,220 to know about face perception in the brain-- there 6 00:00:20,220 --> 00:00:21,450 are some questions. 7 00:00:21,450 --> 00:00:22,957 But at the same time, the agenda has 8 00:00:22,957 --> 00:00:24,540 been to consider the different methods 9 00:00:24,540 --> 00:00:27,030 available in human cognitive neuroscience and what kinds 10 00:00:27,030 --> 00:00:29,400 of questions each one can answer. 11 00:00:29,400 --> 00:00:31,422 So last week, we talked about a bunch of them, 12 00:00:31,422 --> 00:00:32,880 and today, we're going to wrap this 13 00:00:32,880 --> 00:00:36,245 up talking about TMS and animal studies. 14 00:00:36,245 --> 00:00:38,370 But first, I just want to remind you very briefly-- 15 00:00:38,370 --> 00:00:40,720 I won't go through in excruciating detail-- 16 00:00:40,720 --> 00:00:42,690 we talked about behavioral methods, which 17 00:00:42,690 --> 00:00:45,360 are great for characterizing internal representations, 18 00:00:45,360 --> 00:00:49,410 as you saw with face inversion effects and some 19 00:00:49,410 --> 00:00:51,510 of the other behavioral data, they 20 00:00:51,510 --> 00:00:54,330 have major disadvantages, which is that with behavior, you're 21 00:00:54,330 --> 00:00:55,650 just measuring the output. 22 00:00:55,650 --> 00:00:57,000 It's pretty sparse. 23 00:00:57,000 --> 00:00:59,880 And from that, you have to infer all the stuff that happened in 24 00:00:59,880 --> 00:01:03,754 between the retina-- or whatever your sensory modality is-- 25 00:01:03,754 --> 00:01:04,379 and the output. 26 00:01:04,379 --> 00:01:05,970 All that internal mental stuff you 27 00:01:05,970 --> 00:01:07,780 have to infer just from the output. 28 00:01:07,780 --> 00:01:09,572 So it's amazing that that works at all, 29 00:01:09,572 --> 00:01:11,280 and you have to be really smart to do it. 30 00:01:11,280 --> 00:01:13,890 And lots of people have been doing that for a long time, 31 00:01:13,890 --> 00:01:15,270 but it's challenging. 32 00:01:15,270 --> 00:01:16,980 So why not look inside? 33 00:01:16,980 --> 00:01:18,940 And one of the best ways to do that, of course, 34 00:01:18,940 --> 00:01:19,690 is functional MRI. 35 00:01:19,690 --> 00:01:21,420 It has the best spatial resolution 36 00:01:21,420 --> 00:01:23,880 available for normal subjects. 37 00:01:23,880 --> 00:01:25,890 But as you guys all seem to pick up on, 38 00:01:25,890 --> 00:01:29,040 its temporal resolution is lousy, 39 00:01:29,040 --> 00:01:33,390 and its ability to tell you whether the neural activity 40 00:01:33,390 --> 00:01:35,970 you're looking at is causally involved in behavior 41 00:01:35,970 --> 00:01:38,457 is like nil. 42 00:01:38,457 --> 00:01:39,540 Now, at least one of you-- 43 00:01:39,540 --> 00:01:41,700 I only read a few of the assignments-- 44 00:01:41,700 --> 00:01:43,200 but at least one of you was confused 45 00:01:43,200 --> 00:01:45,690 about which causal role we're talking about. 46 00:01:45,690 --> 00:01:48,130 And this is actually really important. 47 00:01:48,130 --> 00:01:50,700 So let's take a moment to talk about this. 48 00:01:50,700 --> 00:01:52,680 Causality, the idea of causality-- 49 00:01:52,680 --> 00:01:56,550 if x causes y, that means, essentially, y wouldn't have 50 00:01:56,550 --> 00:01:59,580 happened without x, or y happened more 51 00:01:59,580 --> 00:02:02,520 because x happened, when x happens 52 00:02:02,520 --> 00:02:04,050 than when x doesn't happen. 53 00:02:04,050 --> 00:02:06,270 So that's pretty basic. 54 00:02:06,270 --> 00:02:09,330 So that means if you want to test the causal role of x on y, 55 00:02:09,330 --> 00:02:11,460 you have to mess with x. 56 00:02:11,460 --> 00:02:13,950 That's the key challenge. 57 00:02:13,950 --> 00:02:17,790 OK, so with that in mind, here's this whole causal chain. 58 00:02:17,790 --> 00:02:19,500 A stimulus lands on a retina. 59 00:02:19,500 --> 00:02:21,270 A bunch of neural activity happens, 60 00:02:21,270 --> 00:02:22,950 and some behavioral output happens. 61 00:02:22,950 --> 00:02:24,720 So there's a whole causal chain there. 62 00:02:24,720 --> 00:02:28,620 Now, let's consider what kind of causality we're talking about. 63 00:02:28,620 --> 00:02:31,830 There's one kind which is that the stimulus causes 64 00:02:31,830 --> 00:02:34,530 neural activity in the brain. 65 00:02:34,530 --> 00:02:37,290 That's a kind of causality that we can absolutely 66 00:02:37,290 --> 00:02:39,810 test, even if we're measuring that with functional MRI, 67 00:02:39,810 --> 00:02:41,790 because we can mess with the stimulus. 68 00:02:41,790 --> 00:02:43,800 We can present different stimuli and produce 69 00:02:43,800 --> 00:02:45,960 different neural activity, OK? 70 00:02:45,960 --> 00:02:49,680 So in that case, we can look at the causal effect 71 00:02:49,680 --> 00:02:51,510 of the stimulus on the neural activity. 72 00:02:51,510 --> 00:02:52,380 No problem. 73 00:02:52,380 --> 00:02:53,170 That's standard. 74 00:02:53,170 --> 00:02:57,870 That's what we do, pretty much, in every experiment with ERPs, 75 00:02:57,870 --> 00:02:59,310 or functional MRI, or so forth. 76 00:02:59,310 --> 00:03:00,600 Is that clear? 77 00:03:00,600 --> 00:03:03,510 OK, on the other hand, if we want 78 00:03:03,510 --> 00:03:07,350 to know this kind of causality from some neural activity we 79 00:03:07,350 --> 00:03:10,680 measure in the brain to either a behavioral response, 80 00:03:10,680 --> 00:03:13,680 or a subjective feeling reported by a behavioral response, 81 00:03:13,680 --> 00:03:16,290 or something like that, that's the challenging part. 82 00:03:16,290 --> 00:03:19,080 That's the kind of causality that we can't infer 83 00:03:19,080 --> 00:03:21,030 from ERPs or functional MRI. 84 00:03:21,030 --> 00:03:23,010 Everyone got that? 85 00:03:23,010 --> 00:03:28,170 Yeah, OK, it's sort of obvious and not obvious. 86 00:03:28,170 --> 00:03:32,128 OK, so let's talk a little bit more 87 00:03:32,128 --> 00:03:33,420 about that temporal resolution. 88 00:03:33,420 --> 00:03:35,280 I know I kept saying the temporal resolution 89 00:03:35,280 --> 00:03:38,190 of functional MRI is lousy, but I had run out of time 90 00:03:38,190 --> 00:03:40,410 and skipped through the key slide, So let me back up 91 00:03:40,410 --> 00:03:41,770 and do that here. 92 00:03:41,770 --> 00:03:45,910 This is the BOLD or MRI response as a function of time. 93 00:03:45,910 --> 00:03:47,430 This is an idealized version of it, 94 00:03:47,430 --> 00:03:49,830 but it looks kind of like that. 95 00:03:49,830 --> 00:03:53,430 And sorry these things are tiny here, but these are seconds-- 96 00:03:53,430 --> 00:03:55,360 5 seconds, 10 seconds. 97 00:03:55,360 --> 00:03:57,240 So let me show you what that means. 98 00:03:57,240 --> 00:04:00,300 If you're recording neural activity, 99 00:04:00,300 --> 00:04:03,720 back here, in the first stage of visual processing in the cortex 100 00:04:03,720 --> 00:04:05,640 coming up from the eyes-- which is where? 101 00:04:11,850 --> 00:04:13,110 AUDIENCE: The occipital lobe. 102 00:04:13,110 --> 00:04:14,818 NANCY KANWISHER: The occipital lobe, yes. 103 00:04:14,818 --> 00:04:15,560 What area? 104 00:04:15,560 --> 00:04:16,560 AUDIENCE: Primary visual cortex. 105 00:04:16,560 --> 00:04:18,560 NANCY KANWISHER: Primary visual cortex, exactly. 106 00:04:18,560 --> 00:04:21,990 OK, so suppose that we stuck an electrode 107 00:04:21,990 --> 00:04:25,470 in my primary visual cortex, and we flashed up 108 00:04:25,470 --> 00:04:27,570 a very brief visual display. 109 00:04:27,570 --> 00:04:30,900 OK, here's a visual stimulus, on for maybe a tenth of a second-- 110 00:04:30,900 --> 00:04:33,390 bright, flashing thing V1 loves-- 111 00:04:33,390 --> 00:04:36,210 V1, also primary visual cortex, right? 112 00:04:36,210 --> 00:04:39,090 OK, the neural activity would happen in less than 1/10 113 00:04:39,090 --> 00:04:41,170 of a second-- super fast. 114 00:04:41,170 --> 00:04:43,020 We know that from work in animals 115 00:04:43,020 --> 00:04:44,820 and even some work in humans, OK? 116 00:04:44,820 --> 00:04:47,715 So super fast after the stimulus. 117 00:04:47,715 --> 00:04:50,370 It just goes straight up from the retina, the LGN V1-- 118 00:04:50,370 --> 00:04:51,630 boom, there it is. 119 00:04:51,630 --> 00:04:54,180 So all the neural activity happens right there, 120 00:04:54,180 --> 00:04:56,850 and it ends right there. 121 00:04:56,850 --> 00:05:00,810 But the MRI response is five, six seconds later-- 122 00:05:00,810 --> 00:05:04,650 this big, sloppy, slow thing as the blood slashes 123 00:05:04,650 --> 00:05:10,260 into V1 many seconds after the relevant neural activity. 124 00:05:10,260 --> 00:05:13,140 So what's relevant here is not just that it's delayed, 125 00:05:13,140 --> 00:05:15,630 but it's big and sloppy. 126 00:05:15,630 --> 00:05:18,480 And so both of those things are the reasons 127 00:05:18,480 --> 00:05:23,340 why functional MRI responses aren't good for distinguishing 128 00:05:23,340 --> 00:05:26,940 what happens on a fine temporal scale of, say, events less 129 00:05:26,940 --> 00:05:28,890 than a second. 130 00:05:28,890 --> 00:05:34,120 All right, OK, in contrast, as I mentioned, 131 00:05:34,120 --> 00:05:35,790 when you glue electrodes on the scalp, 132 00:05:35,790 --> 00:05:39,270 or stick these fancy magnetic sensors in the big hairdryer 133 00:05:39,270 --> 00:05:41,700 device around your head, there you 134 00:05:41,700 --> 00:05:43,620 get beautiful temporal resolution, 135 00:05:43,620 --> 00:05:45,720 but it's like the Heisenberg principle 136 00:05:45,720 --> 00:05:46,950 of cognitive neuroscience. 137 00:05:46,950 --> 00:05:49,200 You want time, you don't get space. 138 00:05:49,200 --> 00:05:53,700 OK, and similarly, here, we can measure 139 00:05:53,700 --> 00:05:59,910 the causal effect on scalp response neural responses, 140 00:05:59,910 --> 00:06:04,470 but not the causal role of those neural responses on behavior. 141 00:06:04,470 --> 00:06:06,200 Everyone clear with this? 142 00:06:06,200 --> 00:06:08,030 OK. 143 00:06:08,030 --> 00:06:11,490 OK, then I talked about these rare cases 144 00:06:11,490 --> 00:06:14,550 where we can record directly from the surface 145 00:06:14,550 --> 00:06:17,130 of the human brain with electrical activity, where 146 00:06:17,130 --> 00:06:22,020 we now get both space and time at the same time. 147 00:06:22,020 --> 00:06:23,950 And the key disadvantage there, of course, 148 00:06:23,950 --> 00:06:25,710 is that it's extremely invasive. 149 00:06:25,710 --> 00:06:29,220 You have to take a big piece of skull off to get in there. 150 00:06:29,220 --> 00:06:30,960 And, of course, that would only happen 151 00:06:30,960 --> 00:06:34,410 in the case of people who are already in pretty 152 00:06:34,410 --> 00:06:37,500 serious medical circumstances. 153 00:06:37,500 --> 00:06:40,620 OK, so now, when we have this incredible opportunity 154 00:06:40,620 --> 00:06:43,980 to record this amazing data from the center of the brain, 155 00:06:43,980 --> 00:06:46,950 does that enable us to make this kind of causal inference 156 00:06:46,950 --> 00:06:48,525 from neural activity to behavior? 157 00:06:52,860 --> 00:06:53,910 Yes? 158 00:06:53,910 --> 00:06:55,255 What do you think, yes? 159 00:06:55,255 --> 00:06:55,755 No? 160 00:07:00,360 --> 00:07:01,140 Isabelle? 161 00:07:01,140 --> 00:07:02,010 Is that Isabelle? 162 00:07:02,010 --> 00:07:03,000 Yes. 163 00:07:03,000 --> 00:07:04,936 Why are you shaking your head? 164 00:07:04,936 --> 00:07:06,720 AUDIENCE: Because it just tells us 165 00:07:06,720 --> 00:07:09,170 which neurons are responsible for [INAUDIBLE].. 166 00:07:09,170 --> 00:07:10,420 NANCY KANWISHER: That's right. 167 00:07:10,420 --> 00:07:13,340 It's cooler, it's fancier, it's more 168 00:07:13,340 --> 00:07:17,000 impressive than functional MRI or ERPs, 169 00:07:17,000 --> 00:07:18,630 but it's still the same deal. 170 00:07:18,630 --> 00:07:22,520 We're just recording responses, OK? 171 00:07:22,520 --> 00:07:25,670 So we can do this causality, from the stimulus 172 00:07:25,670 --> 00:07:27,260 to those neural responses, but it 173 00:07:27,260 --> 00:07:29,420 doesn't tell us which of those responses 174 00:07:29,420 --> 00:07:31,640 are related to behavior yet. 175 00:07:31,640 --> 00:07:34,730 I showed you other methods that do, but this one alone doesn't. 176 00:07:34,730 --> 00:07:36,230 Everybody got that? 177 00:07:36,230 --> 00:07:41,390 All right, so then I talked about studying patients 178 00:07:41,390 --> 00:07:43,880 with focal brain damage. 179 00:07:43,880 --> 00:07:47,000 And here, you really can make a strong causal link 180 00:07:47,000 --> 00:07:49,580 between a bit of brain and a behavioral ability. 181 00:07:49,580 --> 00:07:52,100 You lose that bit of brain, you can no longer do that task. 182 00:07:52,100 --> 00:07:56,870 That's a really direct kind of causal role. 183 00:07:56,870 --> 00:07:59,360 I talked about double associations. 184 00:07:59,360 --> 00:08:01,920 I gave it short shrift, but it's actually really important. 185 00:08:01,920 --> 00:08:03,710 You should know it. 186 00:08:03,710 --> 00:08:07,490 A double dissociation is when you have one patient who can do 187 00:08:07,490 --> 00:08:11,240 A but not B-- say, recognize objects but not faces-- 188 00:08:11,240 --> 00:08:14,690 and another patient who can do B but not A-- 189 00:08:14,690 --> 00:08:16,850 say, recognize faces but not objects. 190 00:08:16,850 --> 00:08:19,820 And when you have in the literature two cases like that, 191 00:08:19,820 --> 00:08:22,130 now you're in a really strong position 192 00:08:22,130 --> 00:08:24,770 to infer that there's something fundamentally different 193 00:08:24,770 --> 00:08:26,900 about face recognition and object recognition 194 00:08:26,900 --> 00:08:27,960 in the brain. 195 00:08:27,960 --> 00:08:30,170 OK, so that's really important-- the senses 196 00:08:30,170 --> 00:08:33,350 in which a double association is more inferentially powerful 197 00:08:33,350 --> 00:08:35,280 than a single association. 198 00:08:35,280 --> 00:08:40,277 OK, "more important" means I'm sure to test you on it. 199 00:08:40,277 --> 00:08:42,860 No, it's also important, whether I was going to test you on it 200 00:08:42,860 --> 00:08:43,429 or not. 201 00:08:43,429 --> 00:08:44,510 [LAUGHS] 202 00:08:44,510 --> 00:08:47,540 OK, and so, of course, in focal brain damage, 203 00:08:47,540 --> 00:08:52,430 we can absolutely infer causal role from a bit of brain 204 00:08:52,430 --> 00:08:53,660 to a behavioral ability. 205 00:08:53,660 --> 00:08:57,850 Lose that bit of brain, lose the ability, yeah? 206 00:08:57,850 --> 00:08:59,830 OK. 207 00:08:59,830 --> 00:09:03,850 And the case that I showed you with that amazing movie 208 00:09:03,850 --> 00:09:06,820 of the guy getting stimulated in his fusiform face area 209 00:09:06,820 --> 00:09:09,760 and seeing percepts of faces on top of whatever he looked at, 210 00:09:09,760 --> 00:09:12,580 that's a quintessential beautiful example 211 00:09:12,580 --> 00:09:15,220 of the causal role of neural activity there. 212 00:09:15,220 --> 00:09:18,670 We're basically directly manipulating neural activity. 213 00:09:18,670 --> 00:09:21,490 We're injecting neural activity there electrically 214 00:09:21,490 --> 00:09:26,410 and looking at the behavioral and cognitive result that 215 00:09:26,410 --> 00:09:26,980 occurs-- 216 00:09:26,980 --> 00:09:27,480 The guy. 217 00:09:27,480 --> 00:09:28,570 Sees a hallucinatory face. 218 00:09:31,800 --> 00:09:36,190 OK, now, that is amazing data, but as I mentioned, 219 00:09:36,190 --> 00:09:37,310 they're very rare. 220 00:09:37,310 --> 00:09:38,470 We have no control over it. 221 00:09:38,470 --> 00:09:41,020 When we get those data, we celebrate and are all excited, 222 00:09:41,020 --> 00:09:42,550 but mostly, we don't get those data. 223 00:09:45,670 --> 00:09:48,610 Plus, those people have serious problems with their brains. 224 00:09:48,610 --> 00:09:50,680 That's why their brains are being opened up. 225 00:09:50,680 --> 00:09:54,520 So is there any way to test a causal role 226 00:09:54,520 --> 00:09:57,850 of a particular part of the brain in a normal subject 227 00:09:57,850 --> 00:10:00,430 who doesn't have their skull open for neurosurgery 228 00:10:00,430 --> 00:10:03,530 and who has not had brain damage? 229 00:10:03,530 --> 00:10:05,140 Well, there's one way, and that's 230 00:10:05,140 --> 00:10:08,380 called transcranial magnetic stimulation, OK? 231 00:10:08,380 --> 00:10:10,720 So in transcranial magnetic stimulation, 232 00:10:10,720 --> 00:10:14,650 you take a coil of wire about yea big. 233 00:10:14,650 --> 00:10:18,430 That's a tight-wrapped coil of wire embedded in plastic, 234 00:10:18,430 --> 00:10:21,100 connected to a ginormous capacitor, 235 00:10:21,100 --> 00:10:23,032 and you hold it next to your head. 236 00:10:23,032 --> 00:10:24,490 Of course, that's what you would do 237 00:10:24,490 --> 00:10:25,910 if you were a neuroscientist. 238 00:10:25,910 --> 00:10:29,860 And you discharge and make an enormous current 239 00:10:29,860 --> 00:10:33,670 through that coil that's very, very strong and very brief. 240 00:10:33,670 --> 00:10:36,310 The whole thing lasts less than one millisecond. 241 00:10:36,310 --> 00:10:38,650 And you guys know from 8.02, another case 242 00:10:38,650 --> 00:10:41,560 of the right-hand rule coming to our service. 243 00:10:41,560 --> 00:10:43,690 You have a hell of a current going in a coil. 244 00:10:43,690 --> 00:10:46,282 What's going to happen in brain tissue underneath? 245 00:10:49,495 --> 00:10:51,370 AUDIENCE: Increase the magnetic [INAUDIBLE].. 246 00:10:51,370 --> 00:10:53,630 AUDIENCE: The electric field will [INAUDIBLE].. 247 00:10:53,630 --> 00:10:55,680 NANCY KANWISHER: Yeah, exactly. 248 00:10:55,680 --> 00:10:58,820 And so you'll get electric fields perpendicular 249 00:10:58,820 --> 00:11:01,820 to the coil sticking right into the brain like that. 250 00:11:01,820 --> 00:11:04,490 And what do you think happens if you stick a big, 251 00:11:04,490 --> 00:11:07,220 huge transient electric field-- boom!-- 252 00:11:07,220 --> 00:11:09,290 into your head like that. 253 00:11:09,290 --> 00:11:11,630 AUDIENCE: Isn't that a magnetic field [INAUDIBLE]?? 254 00:11:11,630 --> 00:11:12,380 NANCY KANWISHER: Yeah, you're right. 255 00:11:12,380 --> 00:11:13,460 Right-hand rule is magnetic field. 256 00:11:13,460 --> 00:11:15,252 I was thinking I was misremembering, right? 257 00:11:15,252 --> 00:11:17,490 Electric current makes magnetic field, right? 258 00:11:17,490 --> 00:11:20,510 It was a long time ago I took 8.02. 259 00:11:20,510 --> 00:11:23,900 I did-- just a long time ago. 260 00:11:23,900 --> 00:11:26,120 Anyway, for current purposes, doesn't matter. 261 00:11:26,120 --> 00:11:27,272 Either would do it. 262 00:11:27,272 --> 00:11:28,730 Actually, there's a variant of this 263 00:11:28,730 --> 00:11:30,800 where it's an electric field, but it's 264 00:11:30,800 --> 00:11:32,360 debated how well that works. 265 00:11:32,360 --> 00:11:35,690 OK, anyway, what happens is you affect 266 00:11:35,690 --> 00:11:39,740 neural activity in tissue right underneath the skull, right? 267 00:11:39,740 --> 00:11:45,853 OK, so if you want to see a picture, a video 268 00:11:45,853 --> 00:11:47,270 of that happening, there's a video 269 00:11:47,270 --> 00:11:49,465 of me getting zapped with TMS on my website. 270 00:11:49,465 --> 00:11:50,340 You can check it out. 271 00:11:50,340 --> 00:11:51,440 It's kind of ludicrous. 272 00:11:51,440 --> 00:11:52,123 Yes, question? 273 00:11:52,123 --> 00:11:53,790 AUDIENCE: What's the spatial resolution? 274 00:11:53,790 --> 00:11:54,420 NANCY KANWISHER: Oh, we're getting there. 275 00:11:54,420 --> 00:11:56,210 We're getting there. 276 00:11:56,210 --> 00:11:58,250 OK, here's an early version of this. 277 00:11:58,250 --> 00:12:01,190 To generate these very strong and brief magnetic fields, 278 00:12:01,190 --> 00:12:03,500 they had these stacks of coils like this, 279 00:12:03,500 --> 00:12:04,940 and they rotated them around. 280 00:12:04,940 --> 00:12:06,170 It's a little crazy. 281 00:12:06,170 --> 00:12:07,412 Here's a more recent version. 282 00:12:07,412 --> 00:12:08,870 It looks like a big torture device, 283 00:12:08,870 --> 00:12:10,120 but it's actually no big deal. 284 00:12:10,120 --> 00:12:12,095 The guy's just holding his head on a chin rest 285 00:12:12,095 --> 00:12:13,730 to hold his head still, and there's 286 00:12:13,730 --> 00:12:18,240 a person holding the coil next to his head like that. 287 00:12:18,240 --> 00:12:22,580 And so that enables us to briefly and somewhat 288 00:12:22,580 --> 00:12:25,340 selectively disrupt a little patch of cortex 289 00:12:25,340 --> 00:12:29,180 there by sticking in this big random field. 290 00:12:29,180 --> 00:12:31,970 Now, spatial resolution is not amazing-- 291 00:12:31,970 --> 00:12:35,362 maybe 1, 2 centimeters, something like that, OK? 292 00:12:35,362 --> 00:12:37,070 It's better than you might guess for such 293 00:12:37,070 --> 00:12:38,755 an incredibly crude device-- 294 00:12:38,755 --> 00:12:41,130 like something people would have done hundreds years ago, 295 00:12:41,130 --> 00:12:43,730 and yet we still do it today. 296 00:12:43,730 --> 00:12:48,620 You can also use a lovely method where you scan the subject with 297 00:12:48,620 --> 00:12:51,770 functional MRI first, find a particular functional region 298 00:12:51,770 --> 00:12:54,350 that you're interested in in that person's brain-- remember, 299 00:12:54,350 --> 00:12:56,720 these things can vary in their exact location across 300 00:12:56,720 --> 00:12:57,650 subjects-- 301 00:12:57,650 --> 00:13:02,060 and then find a way to register externally on the scalp, 302 00:13:02,060 --> 00:13:05,630 what is the closest spot to that region you found in their brain 303 00:13:05,630 --> 00:13:07,070 previously with functional MRI? 304 00:13:07,070 --> 00:13:09,950 And stick the coil right there, and exactly titrate 305 00:13:09,950 --> 00:13:13,310 its location with reference to that brain image. 306 00:13:13,310 --> 00:13:17,010 So that makes this whole enterprise more worthwhile. 307 00:13:17,010 --> 00:13:20,930 So what can TMS tell us about face perception? 308 00:13:20,930 --> 00:13:22,580 Well, here's the problem. 309 00:13:22,580 --> 00:13:25,940 Here's my fusiform face area-- that guy right there. 310 00:13:25,940 --> 00:13:29,960 It's a few centimeters in from the scalp-- from the skull. 311 00:13:29,960 --> 00:13:30,980 So that's a drag. 312 00:13:30,980 --> 00:13:33,620 Unless we opened up my head, we can't reach it there 313 00:13:33,620 --> 00:13:34,910 with the TMS coil. 314 00:13:34,910 --> 00:13:36,920 Believe me, the first time I had a chance 315 00:13:36,920 --> 00:13:39,140 to use a TMS coil, the very first thing I did 316 00:13:39,140 --> 00:13:41,720 was stick the coil there, crank it to the max, 317 00:13:41,720 --> 00:13:43,370 and try to see what would happen. 318 00:13:43,370 --> 00:13:44,600 Not a damn thing happened. 319 00:13:44,600 --> 00:13:46,070 It was very disappointing. 320 00:13:46,070 --> 00:13:48,110 I knew lots of friends who tried the same thing. 321 00:13:48,110 --> 00:13:49,360 It was the most obvious thing. 322 00:13:49,360 --> 00:13:51,370 It just doesn't work; it's too medial. 323 00:13:51,370 --> 00:13:51,870 Yeah? 324 00:13:51,870 --> 00:13:54,740 Well, there was a question over here a moment ago? 325 00:13:54,740 --> 00:13:58,428 AUDIENCE: If you use TMS near someone's brainstem-- 326 00:13:58,428 --> 00:14:00,470 NANCY KANWISHER: Yeah, that wouldn't be so smart. 327 00:14:00,470 --> 00:14:03,930 Luckily, the brainstem is kind of deep in there. 328 00:14:03,930 --> 00:14:06,650 So if you were really stupid and stuck it down, 329 00:14:06,650 --> 00:14:09,320 I don't know, way in here, you might be able to cause trouble. 330 00:14:09,320 --> 00:14:11,670 But mostly, people don't stick it back there. 331 00:14:11,670 --> 00:14:13,670 And actually, the subjects won't let you anyway, 332 00:14:13,670 --> 00:14:15,295 because there is a lot of neck muscles, 333 00:14:15,295 --> 00:14:18,980 and it really hurts when you do TMS over muscles. 334 00:14:18,980 --> 00:14:21,620 And so if anybody had such a stupid ideas 335 00:14:21,620 --> 00:14:25,610 to try to zap the brainstem, the subject 336 00:14:25,610 --> 00:14:27,200 would probably object immediately 337 00:14:27,200 --> 00:14:29,600 before they got very far with it, because it would hurt. 338 00:14:29,600 --> 00:14:32,900 [LAUGHS] And you guys are all probably 339 00:14:32,900 --> 00:14:35,480 wondering, how safe is this? 340 00:14:35,480 --> 00:14:36,707 It's not totally clear. 341 00:14:36,707 --> 00:14:38,540 There have been lots of studies in animals-- 342 00:14:38,540 --> 00:14:39,680 [LAUGHTER] 343 00:14:39,680 --> 00:14:43,847 --where they zap a rabbit 100,000 times 344 00:14:43,847 --> 00:14:45,680 or something like that and say, well, rabbit 345 00:14:45,680 --> 00:14:48,824 seems fine, hops around. 346 00:14:48,824 --> 00:14:51,320 And the best they can do in animal studies. 347 00:14:51,320 --> 00:14:54,200 When I first used TMS around 20 years ago, 348 00:14:54,200 --> 00:14:57,770 I read a few basic safety studies, 349 00:14:57,770 --> 00:14:59,540 and I thought, god, I don't know. 350 00:14:59,540 --> 00:15:01,820 But I also realized that if you look at the papers, 351 00:15:01,820 --> 00:15:04,520 the initials of the subjects were the same as the authors. 352 00:15:04,520 --> 00:15:09,080 So I called them up, and I said, hey, tell me honestly, 353 00:15:09,080 --> 00:15:14,090 did you guys ever notice any ill effects from getting zapped? 354 00:15:14,090 --> 00:15:16,100 And the guy I talked to said, yeah, I've 355 00:15:16,100 --> 00:15:19,610 been zapped about 10,000 times, and I never noticed anything 356 00:15:19,610 --> 00:15:21,770 except for one thing. 357 00:15:21,770 --> 00:15:23,870 After a whole hour of getting zapped, 358 00:15:23,870 --> 00:15:26,930 it gave me a hell of a craving for ice cream. 359 00:15:26,930 --> 00:15:29,300 So I decided, OK, I can live with that. 360 00:15:29,300 --> 00:15:31,460 We got it through the human subjects committee, 361 00:15:31,460 --> 00:15:34,760 and we do-- not a lot, but some TMS in my lab. 362 00:15:34,760 --> 00:15:37,233 And I'm probably now been zapped at least as many times 363 00:15:37,233 --> 00:15:39,650 as that kind, and I guess you guys can judge for yourself. 364 00:15:39,650 --> 00:15:41,275 So you don't have the before condition, 365 00:15:41,275 --> 00:15:42,440 so it's a little hard. 366 00:15:42,440 --> 00:15:46,050 Anyway, as far as anybody can tell, it's perfectly safe. 367 00:15:46,050 --> 00:15:46,550 Yes? 368 00:15:46,550 --> 00:15:48,425 AUDIENCE: So there are some contraindications 369 00:15:48,425 --> 00:15:50,220 if you are prone to seizures or if you're 370 00:15:50,220 --> 00:15:50,753 on certain medications. 371 00:15:50,753 --> 00:15:51,560 NANCY KANWISHER: Yes, yes, yes. 372 00:15:51,560 --> 00:15:53,602 AUDIENCE: So if you ever sign up for a TMS study, 373 00:15:53,602 --> 00:15:54,560 read the fine print. 374 00:15:54,560 --> 00:15:57,290 NANCY KANWISHER: Good point, yep. 375 00:15:57,290 --> 00:15:59,570 OK, so back to this. 376 00:15:59,570 --> 00:16:02,090 It would be lovely to zap that guy, 377 00:16:02,090 --> 00:16:04,430 but it's too hard to reach. 378 00:16:04,430 --> 00:16:07,610 OK, so then, this guy David Pitcher came along, 379 00:16:07,610 --> 00:16:09,440 and he had a very good idea. 380 00:16:09,440 --> 00:16:14,810 And my 1970s synopsis of his idea-- 381 00:16:14,810 --> 00:16:17,690 paraphrasing still-- is if you can't 382 00:16:17,690 --> 00:16:20,810 zap the region you love, love the region you can. 383 00:16:20,810 --> 00:16:24,500 And so Pitcher said, hey, what about that other guy there? 384 00:16:24,500 --> 00:16:25,980 We haven't talked a lot about it. 385 00:16:25,980 --> 00:16:28,340 It's sometimes called the occipital face area. 386 00:16:28,340 --> 00:16:31,820 I think of it as a kind of crappy version of the FFA. 387 00:16:31,820 --> 00:16:33,260 It's kind of face-selective. 388 00:16:33,260 --> 00:16:34,820 It's not as face-selective. 389 00:16:34,820 --> 00:16:37,340 It's more variable, so it's not as fun to study, 390 00:16:37,340 --> 00:16:39,350 but it's there in most people. 391 00:16:39,350 --> 00:16:41,720 I have a damn fine one, I have to say-- 392 00:16:41,720 --> 00:16:46,130 many people do-- and it is right out there next to the scalp, 393 00:16:46,130 --> 00:16:47,410 just asking for it. 394 00:16:47,410 --> 00:16:47,960 Right 395 00:16:47,960 --> 00:16:52,070 OK, so here's what David Pitcher did. 396 00:16:52,070 --> 00:16:53,983 He gave subjects a-- 397 00:16:53,983 --> 00:16:55,400 you need a behavioral task, right? 398 00:16:55,400 --> 00:16:58,010 Because in this case, we're testing the causal role 399 00:16:58,010 --> 00:16:59,460 of a bit of brain on behavior. 400 00:16:59,460 --> 00:17:01,020 So we're going to measure behavior. 401 00:17:01,020 --> 00:17:02,300 And so what is our task? 402 00:17:02,300 --> 00:17:04,760 OK, so here's his task. 403 00:17:04,760 --> 00:17:07,339 Sorry, it's a little tiny, but this is one trial. 404 00:17:07,339 --> 00:17:08,630 Time is going this way. 405 00:17:08,630 --> 00:17:09,800 You present a face. 406 00:17:09,800 --> 00:17:10,849 There's a brief interval. 407 00:17:10,849 --> 00:17:12,260 You present another face. 408 00:17:12,260 --> 00:17:15,690 And the task is just, are those two faces same or different? 409 00:17:15,690 --> 00:17:18,290 It's your basic face perception task. 410 00:17:18,290 --> 00:17:22,040 But then, what you can do is you can zap the occipital face 411 00:17:22,040 --> 00:17:24,470 area at different time-- 412 00:17:24,470 --> 00:17:27,500 during presentation of that second face, 413 00:17:27,500 --> 00:17:29,510 and you can do it at different time intervals. 414 00:17:29,510 --> 00:17:32,120 Remember, its effect is very brief. 415 00:17:32,120 --> 00:17:35,360 The actual magnetic change is less than a millisecond. 416 00:17:35,360 --> 00:17:40,100 OK, so here's what David Pitcher found in that study. 417 00:17:40,100 --> 00:17:43,040 This is accuracy at the same different matching 418 00:17:43,040 --> 00:17:47,390 task when you stimulate the right occipital face 419 00:17:47,390 --> 00:17:48,860 area versus vertex-- 420 00:17:48,860 --> 00:17:51,320 that means you stick the coil up here, which is pretty far 421 00:17:51,320 --> 00:17:52,520 away from face regions. 422 00:17:52,520 --> 00:17:55,340 It's a control condition-- not a perfect one, 423 00:17:55,340 --> 00:17:56,840 but better than nothing. 424 00:17:56,840 --> 00:17:59,990 By the way, TMS usually doesn't hurt unless you stick it 425 00:17:59,990 --> 00:18:00,620 over muscles. 426 00:18:00,620 --> 00:18:02,838 You stick it over the frontal lobes, and-- 427 00:18:02,838 --> 00:18:03,380 I don't know. 428 00:18:03,380 --> 00:18:06,140 Every time I try to disrupt my language abilities, 429 00:18:06,140 --> 00:18:08,723 it hurts too much, because there are muscles up there. 430 00:18:08,723 --> 00:18:11,390 But most places, like the top of the head, there aren't muscles, 431 00:18:11,390 --> 00:18:13,012 and it doesn't really hurt. 432 00:18:13,012 --> 00:18:14,720 But it still makes a loud cracking noise, 433 00:18:14,720 --> 00:18:15,890 and it's kind of like somebody went-- 434 00:18:15,890 --> 00:18:16,390 [TAPS SKULL] 435 00:18:16,390 --> 00:18:17,235 --like that. 436 00:18:17,235 --> 00:18:19,610 So you might imagine you need a control condition, right? 437 00:18:19,610 --> 00:18:21,470 Because if people are-- 438 00:18:21,470 --> 00:18:23,228 that also has a TMS pulse, right? 439 00:18:23,228 --> 00:18:24,770 If you bang somebody on the head when 440 00:18:24,770 --> 00:18:26,478 they're trying to do a task, you probably 441 00:18:26,478 --> 00:18:27,735 disrupt their performance. 442 00:18:27,735 --> 00:18:29,360 So you need to bang them somewhere else 443 00:18:29,360 --> 00:18:32,120 to see if it's specific to that location. 444 00:18:32,120 --> 00:18:37,160 OK, so OK, so here's a little effective on the accuracy. 445 00:18:37,160 --> 00:18:39,260 It's not a huge effect size. 446 00:18:39,260 --> 00:18:45,830 So here, it's going from 85% correct to 78% correct 447 00:18:45,830 --> 00:18:49,580 when you zap occipital face area compared to vertex. 448 00:18:49,580 --> 00:18:51,960 Everybody gets what's going on here? 449 00:18:51,960 --> 00:18:53,220 So that's good. 450 00:18:53,220 --> 00:18:54,620 That tells us something. 451 00:18:54,620 --> 00:18:56,720 Zapping here messes up face perception 452 00:18:56,720 --> 00:18:58,580 more than zapping here, OK? 453 00:19:01,180 --> 00:19:04,100 OK, so that tells us something about causal role, 454 00:19:04,100 --> 00:19:06,620 but what else would you want to know? 455 00:19:06,620 --> 00:19:09,890 That's a beginning, but having just learned 456 00:19:09,890 --> 00:19:11,870 what I told you about TMS, what else 457 00:19:11,870 --> 00:19:14,810 could you do that would tell you more? 458 00:19:14,810 --> 00:19:15,460 Yeah. 459 00:19:15,460 --> 00:19:17,227 AUDIENCE: You see the face [INAUDIBLE].. 460 00:19:17,227 --> 00:19:19,310 NANCY KANWISHER: Ah, well, that's a good question. 461 00:19:19,310 --> 00:19:20,300 It wasn't what I was fishing for, 462 00:19:20,300 --> 00:19:21,690 but it's a very good point. 463 00:19:21,690 --> 00:19:25,160 So this shows disruption, but I showed you with that video 464 00:19:25,160 --> 00:19:27,260 before that if you electrically stimulate the FFA, 465 00:19:27,260 --> 00:19:28,700 you see a face. 466 00:19:28,700 --> 00:19:31,310 Well, unfortunately, nobody has reported 467 00:19:31,310 --> 00:19:34,580 that when you zap a face area, you see a percept of a face. 468 00:19:34,580 --> 00:19:37,190 Boy, that would be fun if true, but it doesn't work. 469 00:19:37,190 --> 00:19:39,740 And there's much debate about why. 470 00:19:39,740 --> 00:19:41,330 It probably has to do with the fact 471 00:19:41,330 --> 00:19:43,550 that your ability to target just that region 472 00:19:43,550 --> 00:19:47,960 is less good than it is with direct stimulation. 473 00:19:47,960 --> 00:19:50,510 There are many reports and many published studies 474 00:19:50,510 --> 00:19:56,450 where if you zap V1, you see a flash of light, OK? 475 00:19:56,450 --> 00:19:58,160 I don't see the damn flash of light. 476 00:19:58,160 --> 00:20:00,500 I've tried, and tried, and tried, 477 00:20:00,500 --> 00:20:04,340 and people in my lab who I trust promised me 478 00:20:04,340 --> 00:20:05,330 they actually see it. 479 00:20:05,330 --> 00:20:06,500 It isn't just BS. 480 00:20:06,500 --> 00:20:08,870 But I don't know; I don't see it. 481 00:20:08,870 --> 00:20:12,860 Anyway, so probably, the question 482 00:20:12,860 --> 00:20:15,890 of when you get disruption and when you get a positive percept 483 00:20:15,890 --> 00:20:18,320 is a very interesting, complicated one. 484 00:20:18,320 --> 00:20:20,090 I think it will ultimately have to do 485 00:20:20,090 --> 00:20:24,110 with how those batches of neurons not only 486 00:20:24,110 --> 00:20:26,450 respond to faces, or light, or whatever, 487 00:20:26,450 --> 00:20:28,460 but how they code for that information, 488 00:20:28,460 --> 00:20:34,580 such that when you put a big artifactual, non-biological 489 00:20:34,580 --> 00:20:36,920 signal in there, will it have any meaning 490 00:20:36,920 --> 00:20:38,930 that the subject can interpret? 491 00:20:38,930 --> 00:20:40,940 I don't know if that's helpful. 492 00:20:40,940 --> 00:20:43,400 I think nobody really understands that, when 493 00:20:43,400 --> 00:20:44,607 you get a positive percept. 494 00:20:44,607 --> 00:20:46,190 But I hope you can at least understand 495 00:20:46,190 --> 00:20:48,500 that at least if you mess with it and muck it up, 496 00:20:48,500 --> 00:20:49,550 you can disrupt. 497 00:20:49,550 --> 00:20:51,140 That logic is clear. 498 00:20:51,140 --> 00:20:53,440 When you will be able to actually stick in a signal 499 00:20:53,440 --> 00:20:55,210 and get a positive, coherent percept 500 00:20:55,210 --> 00:21:00,010 is a more subtle thing, OK? 501 00:21:00,010 --> 00:21:01,790 OK, what else would you want to know? 502 00:21:01,790 --> 00:21:02,290 Yes. 503 00:21:02,290 --> 00:21:05,610 AUDIENCE: Whether it messes up object perception, or not? 504 00:21:05,610 --> 00:21:08,250 NANCY KANWISHER: Absolutely, absolutely. 505 00:21:08,250 --> 00:21:10,980 All we're showing here is it's messing up face perception. 506 00:21:10,980 --> 00:21:12,960 Maybe the guy can't see here. 507 00:21:12,960 --> 00:21:14,557 Maybe he's just globally blind. 508 00:21:14,557 --> 00:21:16,890 Maybe he'd have the same problem with object perception, 509 00:21:16,890 --> 00:21:18,060 absolutely. 510 00:21:18,060 --> 00:21:20,040 The assigned reading for Wednesday 511 00:21:20,040 --> 00:21:23,135 shows exactly that experiment, OK? 512 00:21:23,135 --> 00:21:24,510 What else would you want to know? 513 00:21:31,090 --> 00:21:35,320 Remember, a TMS pulse lasts less than a millisecond. 514 00:21:35,320 --> 00:21:40,280 That enables us to ask a whole interesting kind of question. 515 00:21:40,280 --> 00:21:41,960 What else could we find out? 516 00:21:41,960 --> 00:21:43,060 Yeah. 517 00:21:43,060 --> 00:21:45,070 AUDIENCE: Oh, just [INAUDIBLE]. 518 00:21:45,070 --> 00:21:46,390 NANCY KANWISHER: Yeah, yeah. 519 00:21:46,390 --> 00:21:46,890 I'm sorry. 520 00:21:46,890 --> 00:21:48,710 It's probably-- I don't mean to be insulting your intelligence. 521 00:21:48,710 --> 00:21:50,570 You're probably sitting there saying, this is too obvious. 522 00:21:50,570 --> 00:21:51,830 That's what I'm talking about. 523 00:21:51,830 --> 00:21:54,260 You can zap at different times and ask, 524 00:21:54,260 --> 00:21:56,390 when is the information going through there? 525 00:21:56,390 --> 00:21:59,240 When is that region playing a causal role in behavior? 526 00:21:59,240 --> 00:22:02,000 And here is a very beautiful data that David got. 527 00:22:02,000 --> 00:22:03,980 And there's basically no effect at any point 528 00:22:03,980 --> 00:22:11,030 other than that interval between 60 and 100 milliseconds, OK? 529 00:22:11,030 --> 00:22:12,020 So that's cool. 530 00:22:12,020 --> 00:22:14,210 Tells you that's when that region is likely 531 00:22:14,210 --> 00:22:15,860 engaged in processing. 532 00:22:15,860 --> 00:22:17,540 Make sense? 533 00:22:17,540 --> 00:22:22,910 OK, and is it Shardul? 534 00:22:22,910 --> 00:22:24,507 Yes, already made the point. 535 00:22:24,507 --> 00:22:26,090 I was going to ask you guys, does this 536 00:22:26,090 --> 00:22:27,620 tell us this region is specifically 537 00:22:27,620 --> 00:22:28,880 involved in face perception? 538 00:22:28,880 --> 00:22:29,630 Absolutely not. 539 00:22:29,630 --> 00:22:31,370 We'd have to test other things. 540 00:22:31,370 --> 00:22:34,050 It could affect every visual percept. 541 00:22:36,590 --> 00:22:39,710 OK, so you can read more about that. 542 00:22:39,710 --> 00:22:42,680 All right, just so to collect all the advantages, 543 00:22:42,680 --> 00:22:44,630 it gives you strong causal evidence 544 00:22:44,630 --> 00:22:46,310 that a particular part of the brain 545 00:22:46,310 --> 00:22:49,370 is involved in perception or behavior. 546 00:22:49,370 --> 00:22:51,320 It has good temporal information, 547 00:22:51,320 --> 00:22:54,890 unlike studying patients with focal brain damage. 548 00:22:54,890 --> 00:22:57,080 And it is the only disruption method 549 00:22:57,080 --> 00:22:59,960 that can be used in normal humans, OK? 550 00:22:59,960 --> 00:23:03,050 So that's why, even though it's so crude and rudimentary, 551 00:23:03,050 --> 00:23:07,940 we use it, because it's the only thing that fills that niche. 552 00:23:07,940 --> 00:23:09,620 A couple other unimportant things. 553 00:23:09,620 --> 00:23:11,750 Spatial resolution isn't as good as we'd like, 554 00:23:11,750 --> 00:23:15,740 but it's surprising how much you can learn nonetheless. 555 00:23:15,740 --> 00:23:18,140 And it doesn't reach very far below the scalp, 556 00:23:18,140 --> 00:23:20,060 although Ed Boyden-- the amazing Ed Boyden-- 557 00:23:20,060 --> 00:23:24,230 is working on a crazy new version of it that might. 558 00:23:24,230 --> 00:23:29,670 OK, so where have all this menagerie of methods gotten us 559 00:23:29,670 --> 00:23:30,170 to? 560 00:23:30,170 --> 00:23:31,820 I won't go through all this in detail. 561 00:23:31,820 --> 00:23:33,860 We listed all these questions before. 562 00:23:33,860 --> 00:23:36,920 I gave you some of the answers from previous methods. 563 00:23:36,920 --> 00:23:39,870 The ones we've just talked about show, for example, 564 00:23:39,870 --> 00:23:42,780 that the fusiform face area-- or the occipital face area. 565 00:23:42,780 --> 00:23:44,390 In the case of TMS-- 566 00:23:44,390 --> 00:23:47,360 are causally involved in face perception, apparently 567 00:23:47,360 --> 00:23:49,640 not in object perception, pending 568 00:23:49,640 --> 00:23:51,962 the paper you're going to read. 569 00:23:51,962 --> 00:23:53,420 And so that's important, because it 570 00:23:53,420 --> 00:23:56,210 says when we try to come up with theories 571 00:23:56,210 --> 00:23:58,003 of how face recognition works, we 572 00:23:58,003 --> 00:24:00,170 might think about having a different theory for face 573 00:24:00,170 --> 00:24:04,250 recognition from our theory of object recognition. 574 00:24:04,250 --> 00:24:08,390 OK, so this is all magnificent and wonderful, 575 00:24:08,390 --> 00:24:13,670 but I finessed this list of questions so that the methods 576 00:24:13,670 --> 00:24:16,700 would be able to address them-- at least a little bit-- 577 00:24:16,700 --> 00:24:20,510 and I sneakily left off a whole suite of other questions 578 00:24:20,510 --> 00:24:22,940 that are extremely important-- arguably more 579 00:24:22,940 --> 00:24:26,420 important-- that those methods don't address, OK? 580 00:24:26,420 --> 00:24:29,900 So we want to know not just that a region responds to faces. 581 00:24:29,900 --> 00:24:32,060 We want to know exactly what is represented 582 00:24:32,060 --> 00:24:35,600 in that region or other regions that respond to other things. 583 00:24:35,600 --> 00:24:37,775 We want to know, what is the neural code for faces? 584 00:24:40,520 --> 00:24:43,730 We want to know, what are the actual computations that go on 585 00:24:43,730 --> 00:24:46,530 in a given region, how do they unfold over time, 586 00:24:46,530 --> 00:24:48,830 and how do those computations produce 587 00:24:48,830 --> 00:24:51,050 the representations and behavioral abilities 588 00:24:51,050 --> 00:24:53,360 that we measure? 589 00:24:53,360 --> 00:24:56,293 We want to know, what are the actual anatomical connections? 590 00:24:56,293 --> 00:24:57,710 I showed you that little occipital 591 00:24:57,710 --> 00:25:01,760 face area right nearby but discontiguous from the fusiform 592 00:25:01,760 --> 00:25:02,840 face area. 593 00:25:02,840 --> 00:25:05,990 I've wanted to know for 20 years whether those damn things are 594 00:25:05,990 --> 00:25:07,910 connected anatomically. 595 00:25:07,910 --> 00:25:12,320 Shockingly, we still don't know that. 596 00:25:12,320 --> 00:25:14,127 We want to know what is the causal role 597 00:25:14,127 --> 00:25:15,335 of each region in perception. 598 00:25:15,335 --> 00:25:17,900 And I showed you a few ways that we get little bits of data-- 599 00:25:17,900 --> 00:25:22,760 kind of, sort of-- but there's a lot of cases where we don't. 600 00:25:22,760 --> 00:25:25,280 And we want to know, how does all this stuff get wired up 601 00:25:25,280 --> 00:25:27,320 over development, right? 602 00:25:27,320 --> 00:25:28,880 What is the role of experience? 603 00:25:28,880 --> 00:25:31,580 Do you need to see faces to wire up the phase region, 604 00:25:31,580 --> 00:25:35,910 or is it there at birth before you ever see a face? 605 00:25:35,910 --> 00:25:38,580 The sad truth is, for the most part, 606 00:25:38,580 --> 00:25:40,440 we don't have good methods to answer 607 00:25:40,440 --> 00:25:43,530 these questions in humans. 608 00:25:43,530 --> 00:25:46,710 So that's just a big bummer, but it's true. 609 00:25:46,710 --> 00:25:48,600 Most of these questions can only be 610 00:25:48,600 --> 00:25:51,810 answered by research in animals, or can be best answered 611 00:25:51,810 --> 00:25:54,750 by research in animals. 612 00:25:54,750 --> 00:25:56,550 So I'm going to take a moment to talk 613 00:25:56,550 --> 00:25:58,770 about ethical issues in animal research, 614 00:25:58,770 --> 00:26:02,850 just to note that I think there is an issue. 615 00:26:02,850 --> 00:26:04,890 And I'll say that it's not unreasonable 616 00:26:04,890 --> 00:26:05,730 if you have qualms. 617 00:26:05,730 --> 00:26:07,320 I noticed in an earlier lecture, I 618 00:26:07,320 --> 00:26:10,145 started talking about recording from animal brains, 619 00:26:10,145 --> 00:26:12,270 and I didn't have time at that moment to mark this, 620 00:26:12,270 --> 00:26:13,800 but I do think it's important. 621 00:26:13,800 --> 00:26:16,863 If it makes you uneasy, that's totally legitimate. 622 00:26:16,863 --> 00:26:18,780 You should think about that, and respect that, 623 00:26:18,780 --> 00:26:20,880 and think hard about whether that's-- 624 00:26:20,880 --> 00:26:23,070 what you make of that. 625 00:26:23,070 --> 00:26:26,250 Unambiguously, causing animals pointless suffering 626 00:26:26,250 --> 00:26:28,660 is just completely unacceptable, OK? 627 00:26:28,660 --> 00:26:31,360 So I think we can all agree on that. 628 00:26:31,360 --> 00:26:33,900 And I think there's a very difficult trade-off 629 00:26:33,900 --> 00:26:38,250 between avoiding suffering in animals and research 630 00:26:38,250 --> 00:26:39,960 that has saved countless lives. 631 00:26:39,960 --> 00:26:42,150 So people can legitimately come down 632 00:26:42,150 --> 00:26:45,930 on different sides of this, but many lives have been saved-- 633 00:26:45,930 --> 00:26:48,330 including mine-- based on animal research 634 00:26:48,330 --> 00:26:52,380 that enabled treatments that were life-saving. 635 00:26:52,380 --> 00:26:55,380 And a few things to think about to help you 636 00:26:55,380 --> 00:26:58,860 inform how you handle that trade-off. 637 00:26:58,860 --> 00:27:01,740 First of all, know that animal research in the United States 638 00:27:01,740 --> 00:27:04,650 is very heavily regulated, OK? 639 00:27:04,650 --> 00:27:09,060 So animals receive excellent vet care-- 640 00:27:09,060 --> 00:27:10,830 shockingly, better than probably lots 641 00:27:10,830 --> 00:27:12,040 of citizens of this country. 642 00:27:12,040 --> 00:27:15,270 That's another topic. 643 00:27:15,270 --> 00:27:20,130 Also, there's a very major emphasis on avoiding pain. 644 00:27:20,130 --> 00:27:22,320 So I think it's probably generally true 645 00:27:22,320 --> 00:27:25,710 that it's infrequent that lab animals suffer a lot of pain. 646 00:27:25,710 --> 00:27:29,260 Researchers and vets are very careful to avoid that. 647 00:27:29,260 --> 00:27:31,290 So the bigger issue is not so much 648 00:27:31,290 --> 00:27:34,350 are the animals physically suffering from pain, per se, 649 00:27:34,350 --> 00:27:36,840 but what kind of life is it to live in a lab 650 00:27:36,840 --> 00:27:38,370 and be a lab animal? 651 00:27:38,370 --> 00:27:41,730 And I think that's a legitimate question. 652 00:27:41,730 --> 00:27:45,120 For monkeys, at least, where I, at least-- and maybe 653 00:27:45,120 --> 00:27:48,570 in my speciesist bias, being more sympathetic to similar 654 00:27:48,570 --> 00:27:49,530 species. 655 00:27:49,530 --> 00:27:54,180 I don't know if that's legitimate, but it's a natural. 656 00:27:54,180 --> 00:27:58,020 There are increasing efforts to improve the quality of life 657 00:27:58,020 --> 00:28:00,180 for monkeys in labs. 658 00:28:00,180 --> 00:28:03,210 Many monkeys are now housed in social groups 659 00:28:03,210 --> 00:28:05,640 where they can hang out with their families, 660 00:28:05,640 --> 00:28:10,000 and that certainly improves their quality of life. 661 00:28:10,000 --> 00:28:12,780 Many monkeys basically play video games all day. 662 00:28:12,780 --> 00:28:14,987 In DiCarlo Lab, they're studying visual perception, 663 00:28:14,987 --> 00:28:15,820 and what do they do? 664 00:28:15,820 --> 00:28:18,720 They get the monkeys in there basically doing visual tasks 665 00:28:18,720 --> 00:28:21,240 in exchange for juice rewards. 666 00:28:21,240 --> 00:28:22,830 Not all that different from what, 667 00:28:22,830 --> 00:28:25,920 probably, lots of you guys do. 668 00:28:25,920 --> 00:28:28,500 Now, maybe they'd be happier in nature. 669 00:28:28,500 --> 00:28:31,050 Probably, much of the time, they'd be happier in nature. 670 00:28:31,050 --> 00:28:32,640 But I think that's complicated, too. 671 00:28:32,640 --> 00:28:34,450 Nature can be pretty nasty. 672 00:28:34,450 --> 00:28:39,450 So it's not totally obvious that quality of life in a random lab 673 00:28:39,450 --> 00:28:43,200 is worse than quality of life in nature. 674 00:28:43,200 --> 00:28:46,080 The third point, I'd say, is that the benefits of research 675 00:28:46,080 --> 00:28:46,920 are forever. 676 00:28:46,920 --> 00:28:49,350 You discover something major about how brains work, 677 00:28:49,350 --> 00:28:50,543 that's forever, right? 678 00:28:50,543 --> 00:28:51,960 So you've got to amortize whatever 679 00:28:51,960 --> 00:28:54,780 cost of animal suffering there is against the forever-ness 680 00:28:54,780 --> 00:28:57,540 of that insight. 681 00:28:57,540 --> 00:28:59,160 And so in my view-- 682 00:28:59,160 --> 00:29:00,340 not that you need to agree-- 683 00:29:00,340 --> 00:29:04,200 but in my view, animal research is vastly more justifiable 684 00:29:04,200 --> 00:29:06,300 than things like eating meat or buying 685 00:29:06,300 --> 00:29:10,620 leather, which is just transient entertainment or convenience, 686 00:29:10,620 --> 00:29:12,180 right? 687 00:29:12,180 --> 00:29:14,607 So anyway, you guys, I encourage you all 688 00:29:14,607 --> 00:29:17,190 to think hard about this and to come to different conclusions. 689 00:29:17,190 --> 00:29:21,840 I just wanted to note that these are issues that are worth 690 00:29:21,840 --> 00:29:24,990 thinking about. 691 00:29:24,990 --> 00:29:29,490 That said, the methods in animal research are breathtaking, 692 00:29:29,490 --> 00:29:32,580 and they get more and more breathtaking every day. 693 00:29:32,580 --> 00:29:34,500 In this building, people are constantly 694 00:29:34,500 --> 00:29:37,140 inventing astonishing new ways to answer 695 00:29:37,140 --> 00:29:38,200 all kinds of questions. 696 00:29:38,200 --> 00:29:40,033 And I wanted to give you just a gist of some 697 00:29:40,033 --> 00:29:41,970 of the kind of stuff that you can 698 00:29:41,970 --> 00:29:44,400 do to answer that list of questions 699 00:29:44,400 --> 00:29:46,420 that I said we can't really answer in humans. 700 00:29:46,420 --> 00:29:47,428 So just very briefly-- 701 00:29:47,428 --> 00:29:49,470 this used to be a whole lecture, but I've decided 702 00:29:49,470 --> 00:29:50,760 to cut it to one slide-- 703 00:29:50,760 --> 00:29:54,570 very briefly, about 10-plus years ago, 704 00:29:54,570 --> 00:29:57,810 these two amazing people, Doris Tsao and Winrich Freiwald-- 705 00:29:57,810 --> 00:30:00,990 who, mark my words, will get a Nobel Prize someday, 706 00:30:00,990 --> 00:30:04,800 or at least they should, and they might-- 707 00:30:04,800 --> 00:30:07,200 they popped a monkey in the scanner 708 00:30:07,200 --> 00:30:10,560 and did the very same experiment that we do on humans, OK? 709 00:30:10,560 --> 00:30:12,160 So here's a monkey brain. 710 00:30:12,160 --> 00:30:15,630 Again, the cortex is unfolded so you can see the whole surface. 711 00:30:15,630 --> 00:30:18,450 The dark bits are the bits that used to be inside a fold. 712 00:30:18,450 --> 00:30:20,430 The little yellow patches are the patches 713 00:30:20,430 --> 00:30:22,950 that respond more to faces than objects-- 714 00:30:22,950 --> 00:30:25,450 just analogous to the FFA in humans, 715 00:30:25,450 --> 00:30:28,080 but there are six little patches in monkeys. 716 00:30:28,080 --> 00:30:30,992 OK, so that's so far, that's like, OK, fine, 717 00:30:30,992 --> 00:30:31,950 monkeys have them, too. 718 00:30:31,950 --> 00:30:33,240 That's cool. 719 00:30:33,240 --> 00:30:35,570 But the thing is, because that's a monkey, 720 00:30:35,570 --> 00:30:39,590 you can then stick electrodes straight into that region 721 00:30:39,590 --> 00:30:42,170 right there, and you can record from hundreds 722 00:30:42,170 --> 00:30:43,370 of neurons in that region. 723 00:30:43,370 --> 00:30:45,140 And you can record the response of each 724 00:30:45,140 --> 00:30:47,450 of those hundreds of neurons to hundreds or thousands 725 00:30:47,450 --> 00:30:48,350 of stimuli. 726 00:30:48,350 --> 00:30:50,300 You can characterize the neural code 727 00:30:50,300 --> 00:30:54,770 for faces in monkeys in a way that you just can't for humans. 728 00:30:54,770 --> 00:30:57,320 In fact, Doris now published a paper last year 729 00:30:57,320 --> 00:30:58,850 called "The Neural Code for Faces," 730 00:30:58,850 --> 00:31:00,680 based on a decade of this research. 731 00:31:00,680 --> 00:31:02,660 It's quite breathtaking. 732 00:31:02,660 --> 00:31:08,090 OK, second, you can watch those representations change, 733 00:31:08,090 --> 00:31:10,980 those neural population codes change over time. 734 00:31:10,980 --> 00:31:15,140 You can see, at one time point, what the code seems 735 00:31:15,140 --> 00:31:17,840 to be saying here, and then here, and then here, 736 00:31:17,840 --> 00:31:19,340 and you can watch that-- 737 00:31:19,340 --> 00:31:21,673 those codes-- change over time in each of those regions. 738 00:31:21,673 --> 00:31:23,382 And you can see different representations 739 00:31:23,382 --> 00:31:24,530 in each of those regions. 740 00:31:24,530 --> 00:31:27,398 It's quite breathtaking. 741 00:31:27,398 --> 00:31:28,940 You can answer this question of, what 742 00:31:28,940 --> 00:31:31,215 are the anatomical connections between these regions, 743 00:31:31,215 --> 00:31:32,840 with a whole bunch of different methods 744 00:31:32,840 --> 00:31:34,048 that I won't go through here. 745 00:31:34,048 --> 00:31:37,585 But you can actually answer what's connected to what. 746 00:31:37,585 --> 00:31:38,960 And what these guys have found is 747 00:31:38,960 --> 00:31:41,090 that all of those yellow face patches 748 00:31:41,090 --> 00:31:43,640 are connected to each other by long-range connections that 749 00:31:43,640 --> 00:31:47,030 go through the white matter underneath the gray matter. 750 00:31:47,030 --> 00:31:49,790 Those regions are not connected at all 751 00:31:49,790 --> 00:31:52,980 to the intervening other patches of cortex. 752 00:31:52,980 --> 00:31:55,490 So that set of six little regions 753 00:31:55,490 --> 00:31:59,270 is like a computational unit with different hubs 754 00:31:59,270 --> 00:32:02,480 that talk to each other. 755 00:32:02,480 --> 00:32:04,730 And you can see all that in monkeys in a way 756 00:32:04,730 --> 00:32:08,270 that we still don't know in humans. 757 00:32:08,270 --> 00:32:11,390 You can electrically stimulate, or disrupt with other methods, 758 00:32:11,390 --> 00:32:13,790 any one of those patches one at a time. 759 00:32:13,790 --> 00:32:16,280 You can disrupt them for 50 milliseconds 760 00:32:16,280 --> 00:32:20,810 here, 200 milliseconds there, whatever you like. 761 00:32:20,810 --> 00:32:23,960 And you can study this whole system over development. 762 00:32:23,960 --> 00:32:26,300 How does it change from shortly after birth 763 00:32:26,300 --> 00:32:29,210 to monkey adolescence? 764 00:32:29,210 --> 00:32:32,270 And you can control experience during development. 765 00:32:32,270 --> 00:32:33,710 You can raise monkeys without ever 766 00:32:33,710 --> 00:32:36,320 letting them see faces, and ask whether seeing a face 767 00:32:36,320 --> 00:32:38,390 is necessary for the development of that region. 768 00:32:38,390 --> 00:32:40,400 We'll talk about that in a few lectures. 769 00:32:40,400 --> 00:32:43,250 My point is just that with animal research 770 00:32:43,250 --> 00:32:47,000 you can answer vastly richer, more sophisticated 771 00:32:47,000 --> 00:32:49,080 questions than you could ever answer in humans, 772 00:32:49,080 --> 00:32:50,120 and that's just life. 773 00:32:50,120 --> 00:32:51,140 Yes, what's your name? 774 00:32:51,140 --> 00:32:51,860 AUDIENCE: I'm Esther. 775 00:32:51,860 --> 00:32:52,560 NANCY KANWISHER: Esther, hi. 776 00:32:52,560 --> 00:32:54,080 AUDIENCE: So in these experiments, 777 00:32:54,080 --> 00:32:55,860 they showed them monkey faces, right? 778 00:32:55,860 --> 00:32:56,520 Not humans? 779 00:32:56,520 --> 00:32:58,228 NANCY KANWISHER: Done all different ways. 780 00:32:58,228 --> 00:33:00,110 Remember, monkeys see other monkeys, 781 00:33:00,110 --> 00:33:01,730 but they see a lot of humans, too. 782 00:33:01,730 --> 00:33:04,580 And monkey face patches respond pretty similarly to human 783 00:33:04,580 --> 00:33:06,410 faces and monkey faces. 784 00:33:06,410 --> 00:33:09,020 Human faces respond pretty similarly to human faces 785 00:33:09,020 --> 00:33:10,220 and monkey faces, too-- 786 00:33:10,220 --> 00:33:12,440 even if you don't work in a monkey lab. 787 00:33:15,230 --> 00:33:18,080 OK, so just to say that there are loads of other methods, 788 00:33:18,080 --> 00:33:21,050 and we'll get these later in the course. 789 00:33:21,050 --> 00:33:27,440 OK, so that snake assignment, I hope that seemed-- 790 00:33:27,440 --> 00:33:29,840 I thought you guys, for the most part, did very well 791 00:33:29,840 --> 00:33:33,927 and did exactly the kind of things that we had in mind. 792 00:33:33,927 --> 00:33:36,260 And I just want to go through a few bits of terminology, 793 00:33:36,260 --> 00:33:38,990 because I realized, some of you who messed up the wording, 794 00:33:38,990 --> 00:33:42,080 I hadn't really fully explained what the different words mean. 795 00:33:42,080 --> 00:33:44,960 OK, so first of all, there's this incredibly boring words 796 00:33:44,960 --> 00:33:47,270 of independent variables and dependent variables. 797 00:33:47,270 --> 00:33:49,370 And frankly, I didn't know which was 798 00:33:49,370 --> 00:33:52,850 which until I started teaching this stuff a few years ago. 799 00:33:52,850 --> 00:33:54,980 But the concept is really important. 800 00:33:54,980 --> 00:33:58,760 An independent variable, that's a factor that you, 801 00:33:58,760 --> 00:34:01,850 the experimentalist, manipulate and change, 802 00:34:01,850 --> 00:34:04,610 so that you can then measure what effect it 803 00:34:04,610 --> 00:34:07,610 has on a brain or behavior. 804 00:34:07,610 --> 00:34:10,732 The effect you measure is the dependent variable. 805 00:34:10,732 --> 00:34:13,190 The independent variable is called the independent variable 806 00:34:13,190 --> 00:34:14,600 because you, the experimentalist, 807 00:34:14,600 --> 00:34:17,270 get to mess with it, get to manipulate it, OK? 808 00:34:17,270 --> 00:34:19,639 The dependent one, you're measuring its dependence 809 00:34:19,639 --> 00:34:21,350 on the independent one. 810 00:34:21,350 --> 00:34:23,540 So just basically, in the experiment, 811 00:34:23,540 --> 00:34:25,820 you muck with something in the world, 812 00:34:25,820 --> 00:34:27,320 and you measure the consequences. 813 00:34:27,320 --> 00:34:29,929 The thing you muck with is the independent variable. 814 00:34:29,929 --> 00:34:32,750 The muckee, the thing you measure the effect on, 815 00:34:32,750 --> 00:34:34,460 is the dependent variable. 816 00:34:34,460 --> 00:34:35,489 Make sense? 817 00:34:35,489 --> 00:34:35,989 OK. 818 00:34:38,540 --> 00:34:40,762 All right, so for example, the bold response, 819 00:34:40,762 --> 00:34:43,429 that's a dependent variable, and pretty much all the experiments 820 00:34:43,429 --> 00:34:45,409 we'll talk about here. 821 00:34:45,409 --> 00:34:47,870 All right, the hypothesis, most of you got that. 822 00:34:47,870 --> 00:34:50,943 The hypothesis is the statement about the world 823 00:34:50,943 --> 00:34:52,610 that you're trying to figure out if it's 824 00:34:52,610 --> 00:34:56,840 true in your experiment, OK? 825 00:34:56,840 --> 00:34:59,240 A prediction-- most of you got this, but let me just say, 826 00:34:59,240 --> 00:35:01,340 a prediction is supposed to be extremely precise. 827 00:35:01,340 --> 00:35:03,380 It's the exact statement of what you 828 00:35:03,380 --> 00:35:06,420 will see when you measure your dependent variable 829 00:35:06,420 --> 00:35:08,083 if the hypothesis is true. 830 00:35:08,083 --> 00:35:09,500 What is the crucial thing you have 831 00:35:09,500 --> 00:35:12,350 to look for in the data you measure that tells you 832 00:35:12,350 --> 00:35:13,895 if the hypothesis is true or not? 833 00:35:13,895 --> 00:35:16,160 And the prediction is what you will find 834 00:35:16,160 --> 00:35:20,750 if the hypothesis is true, OK? 835 00:35:20,750 --> 00:35:22,700 Confound-- we haven't talked about this yet. 836 00:35:22,700 --> 00:35:26,360 A confound is a difference between your conditions 837 00:35:26,360 --> 00:35:29,180 that you're manipulating other than the one you 838 00:35:29,180 --> 00:35:32,600 intend to manipulate. 839 00:35:32,600 --> 00:35:36,230 And hence, confounds give you alternative accounts. 840 00:35:36,230 --> 00:35:38,690 Case in point, we compare the response 841 00:35:38,690 --> 00:35:42,260 in the brain when people look at faces versus when they look 842 00:35:42,260 --> 00:35:44,810 at a bunch of random objects. 843 00:35:44,810 --> 00:35:47,450 The fact that the faces have more curvy surfaces, 844 00:35:47,450 --> 00:35:50,840 or are animate, or are more interesting, 845 00:35:50,840 --> 00:35:54,440 those are all confounds with respect to the hypothesis 846 00:35:54,440 --> 00:35:57,650 that that region is responding specifically to faces. 847 00:35:57,650 --> 00:35:59,150 Everybody got that? 848 00:35:59,150 --> 00:36:02,510 OK, it's very common, amid undergraduates, 849 00:36:02,510 --> 00:36:05,780 to use confound to mean anything bad about an experiment. 850 00:36:05,780 --> 00:36:06,710 That's not right. 851 00:36:06,710 --> 00:36:09,050 A confound is a very particular thing. 852 00:36:09,050 --> 00:36:11,780 It's another dimension that co-varies with the thing 853 00:36:11,780 --> 00:36:14,090 that you care about. 854 00:36:14,090 --> 00:36:15,650 It's like a nuisance variable that's 855 00:36:15,650 --> 00:36:17,930 correlated with the thing you're manipulating, 856 00:36:17,930 --> 00:36:22,520 and hence is giving you a difficulty inferring 857 00:36:22,520 --> 00:36:27,180 a clean inference from your data. 858 00:36:27,180 --> 00:36:30,270 All right, a contrast. 859 00:36:30,270 --> 00:36:32,220 We talked about activations in the brain, 860 00:36:32,220 --> 00:36:34,637 like those little yellow patches I showed in monkey brains 861 00:36:34,637 --> 00:36:35,280 a moment ago. 862 00:36:35,280 --> 00:36:38,790 That shows you the bits that responded more 863 00:36:38,790 --> 00:36:40,620 in functional MRI when that monkey was 864 00:36:40,620 --> 00:36:42,540 looking at faces than objects. 865 00:36:42,540 --> 00:36:46,010 The contrast is faces versus objects, right? 866 00:36:46,010 --> 00:36:48,240 It's looking for a higher response 867 00:36:48,240 --> 00:36:49,980 in one condition than another. 868 00:36:49,980 --> 00:36:50,550 Make sense? 869 00:36:50,550 --> 00:36:52,110 OK, these should all be fairly clear. 870 00:36:52,110 --> 00:36:55,050 I just know that not everybody got this. 871 00:36:55,050 --> 00:37:01,000 OK, now, the point of a contrast is to isolate a mental process, 872 00:37:01,000 --> 00:37:01,500 OK? 873 00:37:01,500 --> 00:37:05,140 So let's talk about that for a second. 874 00:37:05,140 --> 00:37:08,760 So how do we decide what contrasts to use? 875 00:37:08,760 --> 00:37:10,740 OK, well, first thing you have to do 876 00:37:10,740 --> 00:37:12,420 is get clear about your hypothesis. 877 00:37:12,420 --> 00:37:13,590 State it explicitly. 878 00:37:13,590 --> 00:37:16,590 Most of you guys did that really well. 879 00:37:16,590 --> 00:37:19,230 Often, your hypothesis-- with functional MRI, at least-- 880 00:37:19,230 --> 00:37:23,520 will concern a particular mental process that you're studying-- 881 00:37:23,520 --> 00:37:25,890 like face recognition. 882 00:37:25,890 --> 00:37:27,720 Now, remember, importantly-- I said 883 00:37:27,720 --> 00:37:29,340 this briefly way back-- functional 884 00:37:29,340 --> 00:37:32,100 MRI can only tell you about differences 885 00:37:32,100 --> 00:37:34,110 between two conditions. 886 00:37:34,110 --> 00:37:36,750 The absolute number, you're going to measure the MR signal 887 00:37:36,750 --> 00:37:38,850 intensity in one condition-- say, 888 00:37:38,850 --> 00:37:40,320 when people are looking at faces-- 889 00:37:40,320 --> 00:37:44,040 and it's going to be something like 726. 890 00:37:44,040 --> 00:37:46,260 And it's totally meaningless. 891 00:37:46,260 --> 00:37:49,020 That's just how strong the MRI signal is from that point. 892 00:37:49,020 --> 00:37:52,080 It doesn't mean a damn thing on its own. 893 00:37:52,080 --> 00:37:55,860 But then, if we also measure, in that same part of the brain, 894 00:37:55,860 --> 00:37:59,010 the MR signal intensity when the subject is looking at objects, 895 00:37:59,010 --> 00:38:04,620 and it's 720, then now we're in business, OK? 896 00:38:04,620 --> 00:38:07,270 All right, so everything is a difference. 897 00:38:07,270 --> 00:38:09,450 So that means that in any imaging experiment, 898 00:38:09,450 --> 00:38:12,840 you'll need to compare two or more conditions. 899 00:38:12,840 --> 00:38:16,170 One condition will never get you anything. 900 00:38:16,170 --> 00:38:18,750 And if you want to isolate a particular mental process, 901 00:38:18,750 --> 00:38:22,770 you need to turn that mental process on or off, 902 00:38:22,770 --> 00:38:25,715 or you need to vary how strongly it's turned on. 903 00:38:25,715 --> 00:38:27,090 So this is all in the service of, 904 00:38:27,090 --> 00:38:29,090 how are we going to decide what contrast to use? 905 00:38:29,090 --> 00:38:32,430 That's our goal, is to turn on or off one little thing. 906 00:38:32,430 --> 00:38:35,800 OK, and here's the problem. 907 00:38:35,800 --> 00:38:39,240 If I told you, OK, look at my face, 908 00:38:39,240 --> 00:38:43,890 and don't process low-level visual information, 909 00:38:43,890 --> 00:38:45,540 and don't think about what I'm saying. 910 00:38:45,540 --> 00:38:47,520 Just see my face. 911 00:38:47,520 --> 00:38:48,270 It's like, what? 912 00:38:48,270 --> 00:38:49,830 You can't do that, right? 913 00:38:49,830 --> 00:38:51,210 There's a whole processing chain. 914 00:38:51,210 --> 00:38:55,320 You can't just do one little mental process at a time. 915 00:38:55,320 --> 00:38:58,110 And so that means we can't just have 916 00:38:58,110 --> 00:39:00,600 a task where you do only mental process x, 917 00:39:00,600 --> 00:39:03,060 and a task where you don't do mental process x, 918 00:39:03,060 --> 00:39:06,150 If you're not doing other stuff. 919 00:39:06,150 --> 00:39:10,650 So what that means is we need to choose two tasks, each of which 920 00:39:10,650 --> 00:39:14,400 has lots of mental processes, but that differ in only one. 921 00:39:17,460 --> 00:39:19,530 And then, we can compare those two. 922 00:39:19,530 --> 00:39:22,680 So this is called subtraction logic, 923 00:39:22,680 --> 00:39:24,780 and it comes from work over 100 years 924 00:39:24,780 --> 00:39:27,250 ago in cognitive psychology and people 925 00:39:27,250 --> 00:39:28,830 who were just measuring behavior. 926 00:39:28,830 --> 00:39:31,740 This dude, Donders, he's a Dutch physiologist, 927 00:39:31,740 --> 00:39:35,670 and he invented the subtraction method to measure reaction 928 00:39:35,670 --> 00:39:38,400 times in humans, way back. 929 00:39:38,400 --> 00:39:41,170 And so with functional MRI, we're doing the same thing. 930 00:39:41,170 --> 00:39:44,070 So we're going to come up with two different tasks which 931 00:39:44,070 --> 00:39:47,310 involve the whole suite, from input, to mental processing, 932 00:39:47,310 --> 00:39:49,140 to output. 933 00:39:49,140 --> 00:39:51,630 And yet, we're going to try to make them differ in just 934 00:39:51,630 --> 00:39:53,520 one particular mental process. 935 00:39:53,520 --> 00:39:54,930 Everybody with the program here? 936 00:39:54,930 --> 00:40:00,870 OK, all right, so what you aspire toward in the contrasts 937 00:40:00,870 --> 00:40:04,470 that you choose is something called a minimal pair, right? 938 00:40:04,470 --> 00:40:07,290 So the idea is we're going to have these two tasks that 939 00:40:07,290 --> 00:40:10,860 are identical in every respect, except for that one thing 940 00:40:10,860 --> 00:40:12,000 we care about, OK? 941 00:40:12,000 --> 00:40:14,170 So here's a task, and here's a task. 942 00:40:14,170 --> 00:40:16,350 This one involves snake perception, 943 00:40:16,350 --> 00:40:18,640 and this one is identical to this one, 944 00:40:18,640 --> 00:40:22,310 except for snake perception. 945 00:40:22,310 --> 00:40:23,720 That's what we want. 946 00:40:23,720 --> 00:40:27,260 OK, and if you get those two things, that's 947 00:40:27,260 --> 00:40:29,030 called a minimal pair. 948 00:40:29,030 --> 00:40:30,890 And this is the single most important thing 949 00:40:30,890 --> 00:40:32,150 in experimental design. 950 00:40:32,150 --> 00:40:33,530 All the other stuff-- 951 00:40:33,530 --> 00:40:36,260 like how you arrange your stimuli over time 952 00:40:36,260 --> 00:40:37,650 and all that kind of stuff-- 953 00:40:37,650 --> 00:40:40,317 OK, it matters a little bit, but this is the crux of the matter. 954 00:40:40,317 --> 00:40:41,960 What are those conditions, and are they 955 00:40:41,960 --> 00:40:44,030 the right kind of minimal pair? 956 00:40:44,030 --> 00:40:47,450 And you guys got the gist, but I felt like most of you 957 00:40:47,450 --> 00:40:49,470 didn't really engage. 958 00:40:49,470 --> 00:40:52,068 OK, what exactly were those non-snake conditions? 959 00:40:52,068 --> 00:40:53,735 So that's really the crux of the matter. 960 00:40:56,750 --> 00:40:59,870 So the most common problem with imaging experiments 961 00:40:59,870 --> 00:41:02,930 is not that the scanner wasn't as fancy as it could have been, 962 00:41:02,930 --> 00:41:05,790 or they didn't use the latest cutting-edge analysis method. 963 00:41:05,790 --> 00:41:08,570 The most common problem is that people's contrasts-- 964 00:41:08,570 --> 00:41:10,070 their conditions-- were not designed 965 00:41:10,070 --> 00:41:13,550 beautifully enough to isolate a single mental process, OK? 966 00:41:16,160 --> 00:41:19,160 That is that the conditions were not minimal pairs. 967 00:41:19,160 --> 00:41:21,500 Any other difference between the two conditions 968 00:41:21,500 --> 00:41:23,735 other than the one you intend is a confound. 969 00:41:26,270 --> 00:41:28,970 All right, so let's engage on this. 970 00:41:28,970 --> 00:41:32,540 Now, if we ran a whole experiment only 971 00:41:32,540 --> 00:41:36,690 on male subjects, is that a confound? 972 00:41:36,690 --> 00:41:37,740 No. 973 00:41:37,740 --> 00:41:39,448 Why not, Isabelle? 974 00:41:39,448 --> 00:41:41,740 AUDIENCE: Because it's not a difference between the two 975 00:41:41,740 --> 00:41:42,940 experimental conditions. 976 00:41:42,940 --> 00:41:45,190 NANCY KANWISHER: Yeah, it's just a bad design feature, 977 00:41:45,190 --> 00:41:47,620 or something that limits your ability to draw inferences. 978 00:41:47,620 --> 00:41:52,230 Again, sub-optimal design it's not the same as a confound. 979 00:41:52,230 --> 00:41:54,280 A confound is this very particular thing. 980 00:41:54,280 --> 00:41:58,630 OK, if all the snake pictures have grassy backgrounds, 981 00:41:58,630 --> 00:42:03,520 and all the non-snake conditions do not, is that a confound? 982 00:42:03,520 --> 00:42:07,000 Yeah, exactly a confound, right. 983 00:42:07,000 --> 00:42:10,360 OK, so I just said all this, so I'll stop boring you. 984 00:42:10,360 --> 00:42:13,570 OK, and the reason that the grassy background 985 00:42:13,570 --> 00:42:16,222 thing is a confound is it gives you an alternative account 986 00:42:16,222 --> 00:42:16,930 of that contrast. 987 00:42:16,930 --> 00:42:18,550 Maybe it's grassiness, not snake-ness, 988 00:42:18,550 --> 00:42:20,110 that's the key difference. 989 00:42:20,110 --> 00:42:21,100 You don't know. 990 00:42:21,100 --> 00:42:23,560 OK, all right. 991 00:42:23,560 --> 00:42:26,140 OK, so all of that said, minimum pairs 992 00:42:26,140 --> 00:42:29,290 are like a platonic ideal of experimental design. 993 00:42:29,290 --> 00:42:33,100 What you aspire toward, but you can never really do it. 994 00:42:33,100 --> 00:42:37,300 If the two conditions were identical except for this one 995 00:42:37,300 --> 00:42:39,250 little thing, they'd be identical. 996 00:42:39,250 --> 00:42:41,320 You can never totally pull it off, 997 00:42:41,320 --> 00:42:44,110 but you can track the little ways in which you fail, 998 00:42:44,110 --> 00:42:48,930 and you can test them one at a time in later experiments, OK? 999 00:42:48,930 --> 00:42:50,667 All right, good. 1000 00:42:50,667 --> 00:42:52,500 All right, so here's what we're going to do. 1001 00:42:52,500 --> 00:42:53,910 We're going to break into groups, 1002 00:42:53,910 --> 00:42:56,535 and you guys are going to think how to take the kind of designs 1003 00:42:56,535 --> 00:42:58,590 that you already put together and turn them 1004 00:42:58,590 --> 00:43:02,885 into actual experiments-- which is going to require deciding 1005 00:43:02,885 --> 00:43:04,260 on a whole bunch of other things, 1006 00:43:04,260 --> 00:43:06,760 and then we're going to discuss the things you come up with. 1007 00:43:06,760 --> 00:43:08,730 OK, what are the exact conditions 1008 00:43:08,730 --> 00:43:10,210 you'll run in your experiment? 1009 00:43:10,210 --> 00:43:13,230 So we could spend a whole class talking about this. 1010 00:43:13,230 --> 00:43:15,180 So I'd love to hear your best-ofs, 1011 00:43:15,180 --> 00:43:18,210 but I don't want to engage on that for a whole class. 1012 00:43:18,210 --> 00:43:19,680 A lot of the keys, some of you guys 1013 00:43:19,680 --> 00:43:23,220 had very clever non-snake conditions 1014 00:43:23,220 --> 00:43:26,310 to test to get close to minimal pairs. 1015 00:43:26,310 --> 00:43:28,260 I want to hear about those. 1016 00:43:28,260 --> 00:43:30,060 But then, beyond that, here's something 1017 00:43:30,060 --> 00:43:32,220 that probably none of you mentioned. 1018 00:43:32,220 --> 00:43:34,620 It's understandable; I don't think I said much about it. 1019 00:43:34,620 --> 00:43:37,680 What are subjects doing in the scanner? 1020 00:43:37,680 --> 00:43:40,020 Are they just lying there, and the stimuli are just 1021 00:43:40,020 --> 00:43:42,530 flashing up, and they're going dumdy-dumdy-dum? 1022 00:43:42,530 --> 00:43:44,280 Are they doing something with the stimuli? 1023 00:43:44,280 --> 00:43:47,040 Go think about what you would want to have happen, OK? 1024 00:43:47,040 --> 00:43:49,680 So what is the task? 1025 00:43:49,680 --> 00:43:52,920 Third, some of you mentioned baseline conditions 1026 00:43:52,920 --> 00:43:54,960 but didn't really say what they are. 1027 00:43:54,960 --> 00:43:56,640 What would a baseline condition be? 1028 00:43:56,640 --> 00:43:59,820 And do you want them, or is it a waste of scan time? 1029 00:43:59,820 --> 00:44:00,810 Think about that. 1030 00:44:03,450 --> 00:44:05,790 OK, next, suppose you get to scan 1031 00:44:05,790 --> 00:44:08,580 10 subjects for one hour each. 1032 00:44:08,580 --> 00:44:11,940 Now, think about how that design is actually going to go. 1033 00:44:11,940 --> 00:44:14,040 Are you going to assign different conditions 1034 00:44:14,040 --> 00:44:16,770 to different subjects-- so these five people will 1035 00:44:16,770 --> 00:44:19,110 see all the snake images, and these five people will 1036 00:44:19,110 --> 00:44:21,532 see all the non-snake images? 1037 00:44:21,532 --> 00:44:23,490 Or, are you going to have snakes and non-snakes 1038 00:44:23,490 --> 00:44:26,520 within each subject? 1039 00:44:26,520 --> 00:44:33,090 Next, it's nice to not make the subject do their task 1040 00:44:33,090 --> 00:44:34,097 non-stop for an hour. 1041 00:44:34,097 --> 00:44:35,430 We usually give subjects breaks. 1042 00:44:35,430 --> 00:44:39,660 So we break an experiment into pieces of 3 to 10 minutes-- 1043 00:44:39,660 --> 00:44:41,700 or whatever I wrote, yeah. 1044 00:44:41,700 --> 00:44:43,710 And so those are called runs. 1045 00:44:43,710 --> 00:44:46,470 So think about how you want to allocate 1046 00:44:46,470 --> 00:44:48,120 those conditions to runs. 1047 00:44:51,600 --> 00:44:53,543 And how many runs will you include? 1048 00:44:53,543 --> 00:44:54,960 And then, think about what's going 1049 00:44:54,960 --> 00:44:56,163 to happen within each run. 1050 00:44:56,163 --> 00:44:58,080 So if you're going to have multiple conditions 1051 00:44:58,080 --> 00:45:01,148 within a run, are you going to stick all of the snake 1052 00:45:01,148 --> 00:45:03,690 conditions in the first half and all the non-snake conditions 1053 00:45:03,690 --> 00:45:04,710 in the second half? 1054 00:45:04,710 --> 00:45:08,080 If not, why not? 1055 00:45:08,080 --> 00:45:11,133 And if there are multiple conditions within a run, 1056 00:45:11,133 --> 00:45:13,050 yeah, are you going to clump them all together 1057 00:45:13,050 --> 00:45:19,050 or interleave them randomly, and what are the trade-offs there? 1058 00:45:19,050 --> 00:45:24,750 And, what is the order of conditions within a run? 1059 00:45:24,750 --> 00:45:27,150 And we won't get to number 10 for the moment. 1060 00:45:27,150 --> 00:45:30,998 OK, so we're going to break you guys into four groups, 1061 00:45:30,998 --> 00:45:32,790 and you're going to talk amongst yourselves 1062 00:45:32,790 --> 00:45:34,498 and try to come up with your best answers 1063 00:45:34,498 --> 00:45:37,988 to these in five, 10, minutes, something like that. 1064 00:45:37,988 --> 00:45:39,780 And then, we're going to pull your thoughts 1065 00:45:39,780 --> 00:45:43,590 on this when we get back, OK? 1066 00:45:43,590 --> 00:45:46,950 OK, so part of my agenda in doing this 1067 00:45:46,950 --> 00:45:49,960 is just to break up the monotony of me going blah, 1068 00:45:49,960 --> 00:45:53,080 blah, blah, because experimental design is like, it's important, 1069 00:45:53,080 --> 00:45:54,910 but it's not the most riveting thing. 1070 00:45:54,910 --> 00:45:57,750 The other thing is, experimental design is basically 1071 00:45:57,750 --> 00:45:59,440 just organized common sense. 1072 00:45:59,440 --> 00:46:00,997 And so most of this stuff, you guys 1073 00:46:00,997 --> 00:46:02,580 just answered all these questions just 1074 00:46:02,580 --> 00:46:03,580 by thinking about them. 1075 00:46:03,580 --> 00:46:06,000 You need to know a few things about the methods, 1076 00:46:06,000 --> 00:46:10,448 but really, in experimental design, the biggest, the best 1077 00:46:10,448 --> 00:46:12,990 guideline, the best way to think about design is think about, 1078 00:46:12,990 --> 00:46:14,040 OK you're the subject. 1079 00:46:14,040 --> 00:46:15,207 You're lying in the scanner. 1080 00:46:15,207 --> 00:46:16,560 You're doing that. 1081 00:46:16,560 --> 00:46:17,425 Does that work? 1082 00:46:17,425 --> 00:46:19,050 Are you actually going to be doing what 1083 00:46:19,050 --> 00:46:20,217 you're supposed to be doing? 1084 00:46:20,217 --> 00:46:23,393 Are you going to be selectively turning on and off this one 1085 00:46:23,393 --> 00:46:24,935 little mental process you care about, 1086 00:46:24,935 --> 00:46:26,602 or are you doing a million other things, 1087 00:46:26,602 --> 00:46:29,738 like falling asleep, and getting bored, and all of that, 1088 00:46:29,738 --> 00:46:31,530 and predicting what's going to happen next, 1089 00:46:31,530 --> 00:46:32,880 and all that kind of stuff? 1090 00:46:32,880 --> 00:46:36,990 OK, all right, so let's just take a few examples. 1091 00:46:39,780 --> 00:46:42,660 What were some good kinds of control conditions-- that 1092 00:46:42,660 --> 00:46:47,970 is, non-snake stimuli that are good to compare to snakes that 1093 00:46:47,970 --> 00:46:49,980 maybe aren't perfect minimal pairs, 1094 00:46:49,980 --> 00:46:52,050 but that get partway there? 1095 00:46:52,050 --> 00:46:54,990 I saw a few, just in the few papers that I looked at. 1096 00:46:54,990 --> 00:46:56,542 Yeah, I've got a-- 1097 00:46:56,542 --> 00:46:58,500 I'm sorry, I've asked your name like six times. 1098 00:46:58,500 --> 00:46:59,250 But I'm going to-- 1099 00:46:59,250 --> 00:47:01,470 on my trusty sheet, tell me again how you say it? 1100 00:47:01,470 --> 00:47:01,940 AUDIENCE: Achay. 1101 00:47:01,940 --> 00:47:02,898 NANCY KANWISHER: Achay. 1102 00:47:02,898 --> 00:47:03,810 OK. 1103 00:47:03,810 --> 00:47:07,097 AUDIENCE: So for ours, we compared snakes to worms. 1104 00:47:07,097 --> 00:47:08,430 NANCY KANWISHER: To worms, yeah. 1105 00:47:08,430 --> 00:47:10,800 AUDIENCE: Because they have really similar shapes. 1106 00:47:10,800 --> 00:47:13,560 NANCY KANWISHER: Awesome, and they're both animate. 1107 00:47:13,560 --> 00:47:15,690 That's great, love it. 1108 00:47:15,690 --> 00:47:17,010 What else? 1109 00:47:17,010 --> 00:47:19,306 Who else had a good control condition? 1110 00:47:22,778 --> 00:47:25,500 Or who had an interest in control condition? 1111 00:47:25,500 --> 00:47:27,972 Yes, sorry, your name is-- 1112 00:47:27,972 --> 00:47:28,680 AUDIENCE: Lauren. 1113 00:47:28,680 --> 00:47:30,330 NANCY KANWISHER: Yes, OK. 1114 00:47:30,330 --> 00:47:33,000 AUDIENCE: Yep, our group had pretty much the same baseline 1115 00:47:33,000 --> 00:47:35,130 background, and we would just superimpose 1116 00:47:35,130 --> 00:47:38,620 images of different objects on it so that remained consistent 1117 00:47:38,620 --> 00:47:39,120 throughout. 1118 00:47:39,120 --> 00:47:40,920 NANCY KANWISHER: Uh-huh, and the background was like what? 1119 00:47:40,920 --> 00:47:41,878 AUDIENCE: Forest floor. 1120 00:47:41,913 --> 00:47:43,080 NANCY KANWISHER: Uh-huh, OK. 1121 00:47:43,080 --> 00:47:45,930 So you stick a toaster on the forest floor 1122 00:47:45,930 --> 00:47:48,360 or something like that, versus a snake or something, yeah? 1123 00:47:48,360 --> 00:47:51,120 AUDIENCE: The idea was more like other animals, 1124 00:47:51,120 --> 00:47:53,212 or stuff that would make more sense. 1125 00:47:53,212 --> 00:47:54,420 NANCY KANWISHER: That's good. 1126 00:47:54,420 --> 00:47:56,880 That deals with the grass confound problem, right? 1127 00:47:56,880 --> 00:47:58,230 Absolutely, very good. 1128 00:47:58,230 --> 00:48:00,360 What else? 1129 00:48:00,360 --> 00:48:02,090 David, you had interesting ideas. 1130 00:48:02,090 --> 00:48:05,160 AUDIENCE: Well, we were talking a lot 1131 00:48:05,160 --> 00:48:07,300 about animate versus inanimate things, 1132 00:48:07,300 --> 00:48:09,020 so like comparing to a garden hose. 1133 00:48:09,020 --> 00:48:10,560 NANCY KANWISHER: Yes, garden hose! 1134 00:48:10,560 --> 00:48:11,370 Love it! 1135 00:48:11,370 --> 00:48:13,600 I actually ran this experiment a bunch of years ago, 1136 00:48:13,600 --> 00:48:15,270 and we used a garden hose-- 1137 00:48:15,270 --> 00:48:17,768 or a bunch of garden hoses, coiled up in the grass. 1138 00:48:17,768 --> 00:48:19,560 We tried to make them slither and all that. 1139 00:48:19,560 --> 00:48:21,940 Anyway, but garden hose is great. 1140 00:48:21,940 --> 00:48:22,440 Say more. 1141 00:48:22,440 --> 00:48:24,480 You had other good ideas in your-- 1142 00:48:24,480 --> 00:48:27,840 AUDIENCE: Yeah, we also-- we talked about some of my ideas 1143 00:48:27,840 --> 00:48:31,378 were looking at videos, with motion. 1144 00:48:31,378 --> 00:48:32,253 NANCY KANWISHER: Why? 1145 00:48:32,253 --> 00:48:33,000 AUDIENCE: What'd you say? 1146 00:48:33,000 --> 00:48:33,875 NANCY KANWISHER: Why? 1147 00:48:33,875 --> 00:48:38,950 AUDIENCE: Oh, because when you get a snake, 1148 00:48:38,950 --> 00:48:41,800 it kind of slithers and has this very distinctive thing, 1149 00:48:41,800 --> 00:48:44,320 that it feels like the motion is what creeps me 1150 00:48:44,320 --> 00:48:45,490 out when I see a snake. 1151 00:48:45,490 --> 00:48:46,540 NANCY KANWISHER: Totally. 1152 00:48:46,540 --> 00:48:47,170 Me, too. 1153 00:48:47,170 --> 00:48:49,630 AUDIENCE: And if you have a rigid thing that 1154 00:48:49,630 --> 00:48:53,560 looked like a snake, but it was just sliding rigidly, 1155 00:48:53,560 --> 00:48:55,338 then it wouldn't really creep me out. 1156 00:48:55,338 --> 00:48:56,380 NANCY KANWISHER: Exactly. 1157 00:48:56,380 --> 00:48:57,940 This is a key insight, right? 1158 00:48:57,940 --> 00:49:01,120 So think about if we're interested in how you perceive 1159 00:49:01,120 --> 00:49:04,180 snakes, we want to know not just how you do it in some weird lab 1160 00:49:04,180 --> 00:49:04,700 environment. 1161 00:49:04,700 --> 00:49:06,070 We want to know how you'd actually do that. 1162 00:49:06,070 --> 00:49:07,690 The whole reason to choose snakes 1163 00:49:07,690 --> 00:49:10,360 is it seems like something that could be biologically relevant. 1164 00:49:10,360 --> 00:49:12,730 There might be special hardware. 1165 00:49:12,730 --> 00:49:15,410 When I'm out hiking and I see even a curved stick, 1166 00:49:15,410 --> 00:49:17,967 I, like, jump and shriek before I can censor myself. 1167 00:49:17,967 --> 00:49:18,550 It's horrible. 1168 00:49:18,550 --> 00:49:19,660 I find it very embarrassing. 1169 00:49:19,660 --> 00:49:21,285 It's not consistent with my self-image. 1170 00:49:21,285 --> 00:49:23,810 But I have no control over it; it just happens. 1171 00:49:23,810 --> 00:49:25,310 And so I've thought for a long time, 1172 00:49:25,310 --> 00:49:27,768 there's some damn bit of my brain that's making me do that, 1173 00:49:27,768 --> 00:49:29,830 and it pisses me off, and I'm going to find it. 1174 00:49:29,830 --> 00:49:31,288 Well, we looked and didn't find it. 1175 00:49:31,288 --> 00:49:34,600 But anyway, you go from those intuitions. 1176 00:49:34,600 --> 00:49:37,037 Often, your own introspections are very informative, 1177 00:49:37,037 --> 00:49:38,870 and I think your intuition is exactly right. 1178 00:49:38,870 --> 00:49:41,920 There's a very characteristic motion that snakes have, 1179 00:49:41,920 --> 00:49:43,690 and it could be that that's the cue. 1180 00:49:43,690 --> 00:49:46,975 So then, the trick is you have slithery motion versus what? 1181 00:49:49,628 --> 00:49:50,170 I don't know. 1182 00:49:50,170 --> 00:49:50,670 That's hard. 1183 00:49:50,670 --> 00:49:53,170 What other kinds of motions could you have? 1184 00:49:53,170 --> 00:49:56,200 AUDIENCE: Right, so you could have just even rigid motion, 1185 00:49:56,200 --> 00:50:01,320 where it's not slithering, or it's not changing shape. 1186 00:50:01,320 --> 00:50:03,192 It's just sliding or rotating. 1187 00:50:03,192 --> 00:50:04,150 NANCY KANWISHER: Right. 1188 00:50:04,150 --> 00:50:05,320 Right, exactly. 1189 00:50:05,320 --> 00:50:07,570 Anyway, all those are all good ideas. 1190 00:50:07,570 --> 00:50:12,850 Good, so what should the subject do in the scanner? 1191 00:50:12,850 --> 00:50:15,760 Should they lie there and go dumdy-dumdy-dum? 1192 00:50:15,760 --> 00:50:17,150 Should they do a task? 1193 00:50:17,150 --> 00:50:18,085 If so, what task? 1194 00:50:22,960 --> 00:50:24,580 Oh, if you guys don't volunteer, I'm 1195 00:50:24,580 --> 00:50:26,860 going to start calling on people-- 1196 00:50:26,860 --> 00:50:29,020 even though, as the Jenkins study, 1197 00:50:29,020 --> 00:50:32,170 showed it's nearly impossible to look at these damn photographs 1198 00:50:32,170 --> 00:50:35,290 and figure out who's who. 1199 00:50:35,290 --> 00:50:37,737 David, in the back. 1200 00:50:37,737 --> 00:50:38,320 AUDIENCE: So-- 1201 00:50:38,320 --> 00:50:38,890 NANCY KANWISHER: Task. 1202 00:50:38,890 --> 00:50:39,800 Task, or no task? 1203 00:50:39,800 --> 00:50:40,300 What task? 1204 00:50:40,300 --> 00:50:44,920 AUDIENCE: Yes, we talked about having the subjects find a way 1205 00:50:44,920 --> 00:50:49,270 to indicate that they're paying attention and not just dozing 1206 00:50:49,270 --> 00:50:49,900 off. 1207 00:50:49,900 --> 00:50:50,858 NANCY KANWISHER: Right. 1208 00:50:50,858 --> 00:50:56,110 AUDIENCE: So one idea was to have them essentially indicate 1209 00:50:56,110 --> 00:50:59,080 the source, make [INAUDIBLE] think about a bunch of problems 1210 00:50:59,080 --> 00:51:01,610 like, well, we don't want them thinking about snakes 1211 00:51:01,610 --> 00:51:02,860 for the entire our experiment. 1212 00:51:02,860 --> 00:51:04,070 So-- 1213 00:51:04,070 --> 00:51:05,150 NANCY KANWISHER: Also, if they're going to tell you 1214 00:51:05,150 --> 00:51:07,370 they're seeing a snake, maybe by pushing a button, 1215 00:51:07,370 --> 00:51:09,495 then on the snake trails, they're pushing a button, 1216 00:51:09,495 --> 00:51:11,203 and on the non-snake trials, they're not. 1217 00:51:11,203 --> 00:51:12,870 AUDIENCE: Well, on the non-snake trials, 1218 00:51:12,870 --> 00:51:14,330 they would push another button. 1219 00:51:14,330 --> 00:51:15,950 NANCY KANWISHER: Ah, but then they have two different motor 1220 00:51:15,950 --> 00:51:16,450 responses. 1221 00:51:16,450 --> 00:51:19,460 AUDIENCE: Yeah, so then we would have them run the experiment 1222 00:51:19,460 --> 00:51:21,203 again, but switch buttons. 1223 00:51:21,203 --> 00:51:22,370 NANCY KANWISHER: Good, good. 1224 00:51:22,370 --> 00:51:24,620 Smart, very nice. 1225 00:51:24,620 --> 00:51:26,390 AUDIENCE: Ideally, we might just have 1226 00:51:26,390 --> 00:51:30,632 them perform a task that's completely unrelated to looking 1227 00:51:30,632 --> 00:51:32,090 at snakes or thinking about snakes, 1228 00:51:32,090 --> 00:51:34,750 just so that they're not affecting-- 1229 00:51:34,750 --> 00:51:36,500 NANCY KANWISHER: Absolutely, and the point 1230 00:51:36,500 --> 00:51:37,670 you made before is a good one. 1231 00:51:37,670 --> 00:51:39,670 If you're looking for snakes all the time, maybe 1232 00:51:39,670 --> 00:51:43,147 even if it's apples and dogs, you're thinking, is it a snake? 1233 00:51:43,147 --> 00:51:43,730 Is it a snake? 1234 00:51:43,730 --> 00:51:46,640 And maybe you're using that region, and it's a mess, right? 1235 00:51:46,640 --> 00:51:48,860 Absolutely, yeah. 1236 00:51:48,860 --> 00:51:50,527 All right, so this is a common challenge 1237 00:51:50,527 --> 00:51:52,193 in experimental design, and these things 1238 00:51:52,193 --> 00:51:53,490 don't have clear right answers. 1239 00:51:53,490 --> 00:51:55,532 What I want you to do is just see the trade-offs. 1240 00:51:55,532 --> 00:51:58,670 On the one hand, just passive viewing, lying there, 1241 00:51:58,670 --> 00:51:59,720 is good in a way. 1242 00:51:59,720 --> 00:52:02,420 The things are just impinging on your sensoria, 1243 00:52:02,420 --> 00:52:03,980 and it's doing whatever it will do. 1244 00:52:03,980 --> 00:52:06,320 But the downside is subjects fall asleep and get bored, 1245 00:52:06,320 --> 00:52:07,890 and you don't know if they're awake. 1246 00:52:07,890 --> 00:52:08,723 So that's a problem. 1247 00:52:11,400 --> 00:52:14,420 OK, but the key thing is whatever the task is, 1248 00:52:14,420 --> 00:52:17,250 you don't want the task to engage asymmetrically 1249 00:52:17,250 --> 00:52:19,250 with the stimulus condition, because then you're 1250 00:52:19,250 --> 00:52:21,830 building in a confound, right? 1251 00:52:21,830 --> 00:52:24,590 So in the group I was in, we were talking about, well, you 1252 00:52:24,590 --> 00:52:25,640 could have people-- 1253 00:52:25,640 --> 00:52:27,590 well, we were talking, actually, about faces and objects 1254 00:52:27,590 --> 00:52:28,132 in that case. 1255 00:52:28,132 --> 00:52:30,522 So you could have people name the things, 1256 00:52:30,522 --> 00:52:32,480 but if they're naming snakes versus non-snakes, 1257 00:52:32,480 --> 00:52:34,230 it's not very good if they're going snake, 1258 00:52:34,230 --> 00:52:38,220 snake, snake, snake, snake, dog, toaster, apple. 1259 00:52:38,220 --> 00:52:41,210 One is easier than the other and more repetitive. 1260 00:52:41,210 --> 00:52:43,730 There are all kinds of problems there. 1261 00:52:43,730 --> 00:52:45,590 All right, baseline conditions. 1262 00:52:45,590 --> 00:52:47,690 I didn't really say what a baseline condition was. 1263 00:52:47,690 --> 00:52:49,700 Sorry about that. 1264 00:52:49,700 --> 00:52:52,520 What I meant by a baseline is different from a control 1265 00:52:52,520 --> 00:52:53,120 condition. 1266 00:52:53,120 --> 00:52:55,400 The control condition would be like non-snakes 1267 00:52:55,400 --> 00:52:57,620 contrasted with the snakes. 1268 00:52:57,620 --> 00:53:01,040 Baseline tends to be like a minimalist condition that's 1269 00:53:01,040 --> 00:53:03,710 supposed to turn the brain off. 1270 00:53:03,710 --> 00:53:05,870 Can we turn the brain off? 1271 00:53:05,870 --> 00:53:09,050 No, of course not, but we can aspire toward it. 1272 00:53:09,050 --> 00:53:11,240 We can go partway out there. 1273 00:53:11,240 --> 00:53:13,760 We can say, OK, if we're studying vision, 1274 00:53:13,760 --> 00:53:18,950 let's minimize activity in the visual system as best we can, 1275 00:53:18,950 --> 00:53:19,880 OK? 1276 00:53:19,880 --> 00:53:21,950 So you could just have a blank screen 1277 00:53:21,950 --> 00:53:23,660 that feels like a pretty minimal thing. 1278 00:53:23,660 --> 00:53:25,880 You can have people close their eyes. 1279 00:53:25,880 --> 00:53:27,710 The reason that, in vision experiments, 1280 00:53:27,710 --> 00:53:30,290 people tend to have fixation, where there's a tiny dot 1281 00:53:30,290 --> 00:53:33,170 and subjects are supposed to hold their eyes on it, 1282 00:53:33,170 --> 00:53:36,210 is that in natural- left of their own devices, 1283 00:53:36,210 --> 00:53:38,810 people move their eyes a lot-- several times a second. 1284 00:53:38,810 --> 00:53:41,720 And moving your eyes produces all kinds of activity and lots 1285 00:53:41,720 --> 00:53:42,980 of neurons. 1286 00:53:42,980 --> 00:53:46,370 And so it's a very active visual thing, 1287 00:53:46,370 --> 00:53:48,120 even if there's nothing on the screen. 1288 00:53:48,120 --> 00:53:51,440 And so staring at dot is closer to shutting off 1289 00:53:51,440 --> 00:53:54,710 your visual system, even though it's not shutting it off. 1290 00:53:54,710 --> 00:53:58,640 OK, so given that most of the contrasts we've talked about 1291 00:53:58,640 --> 00:54:01,340 are like faces versus objects or snakes 1292 00:54:01,340 --> 00:54:04,610 versus non-snakes, and all the activations that I've shown you 1293 00:54:04,610 --> 00:54:09,320 guys are contrast between an experimental condition 1294 00:54:09,320 --> 00:54:12,753 and a control condition, why are we bothering with baseline? 1295 00:54:12,753 --> 00:54:14,420 It doesn't even figure in that contrast. 1296 00:54:17,670 --> 00:54:18,780 Yes, Jimmy? 1297 00:54:18,780 --> 00:54:20,460 AUDIENCE: Well, if the region is truly 1298 00:54:20,460 --> 00:54:24,770 selective for only snakes, you could use the baseline as, 1299 00:54:24,770 --> 00:54:27,180 in this sense, like a control, because you can compare 1300 00:54:27,180 --> 00:54:28,470 the other control to it. 1301 00:54:28,470 --> 00:54:30,450 If it's really selective for snakes, 1302 00:54:30,450 --> 00:54:33,940 then the non-snake object should respond in the same as the 1303 00:54:33,940 --> 00:54:34,830 [? minimal. ?] 1304 00:54:34,830 --> 00:54:36,705 NANCY KANWISHER: Awesome, everybody get that? 1305 00:54:36,705 --> 00:54:38,310 So that was exactly right. 1306 00:54:38,310 --> 00:54:40,760 And this is, I think, a very interesting point. 1307 00:54:40,760 --> 00:54:42,300 So suppose we have-- 1308 00:54:42,300 --> 00:54:44,440 remember, with MRI, you just have two numbers. 1309 00:54:44,440 --> 00:54:47,580 So here's the snake response, and here's 1310 00:54:47,580 --> 00:54:50,170 the non-snake response. 1311 00:54:50,170 --> 00:54:52,770 If we don't have a baseline, that's all we have-- 1312 00:54:52,770 --> 00:54:55,920 two numbers, OK? 1313 00:54:55,920 --> 00:54:57,150 And that's fine. 1314 00:54:57,150 --> 00:55:00,400 If we run enough subjects, that could be significant. 1315 00:55:00,400 --> 00:55:03,570 But now, let's think what else we know if we have a baseline. 1316 00:55:03,570 --> 00:55:06,510 Suppose we have a baseline of staring at dot. 1317 00:55:06,510 --> 00:55:09,038 And that's down here. 1318 00:55:09,038 --> 00:55:10,080 We'll call that fixation. 1319 00:55:13,040 --> 00:55:13,790 Are you impressed? 1320 00:55:13,790 --> 00:55:15,498 And you've run enough subjects, so that's 1321 00:55:15,498 --> 00:55:16,760 significantly different. 1322 00:55:16,760 --> 00:55:17,660 Are you impressed? 1323 00:55:23,970 --> 00:55:24,470 Yeah. 1324 00:55:24,470 --> 00:55:27,110 AUDIENCE: Less so than if the fixation were higher up. 1325 00:55:27,110 --> 00:55:28,760 NANCY KANWISHER: Exactly! 1326 00:55:28,760 --> 00:55:29,870 Why? 1327 00:55:29,870 --> 00:55:32,720 AUDIENCE: Because then, if the results are higher up-- 1328 00:55:32,720 --> 00:55:35,840 or if the fixation is the second one that you just drew-- 1329 00:55:35,840 --> 00:55:39,418 then the response to a snake is twice as much as non-snake. 1330 00:55:39,418 --> 00:55:40,460 NANCY KANWISHER: Exactly. 1331 00:55:40,460 --> 00:55:42,835 Does everybody see how-- yeah, that might be significant, 1332 00:55:42,835 --> 00:55:44,090 but who cares, right? 1333 00:55:44,090 --> 00:55:47,150 Some tiny little ratty-ass effect, 1334 00:55:47,150 --> 00:55:51,830 versus if it's like here, or even-- 1335 00:55:51,830 --> 00:55:54,680 this is the case Jimmy was talking about, like that. 1336 00:55:54,680 --> 00:55:57,290 No response at all more than staring at a 1337 00:55:57,290 --> 00:56:01,237 dot to the non-snakes, and yet this response to the snakes. 1338 00:56:01,237 --> 00:56:02,570 That would even more impressive. 1339 00:56:02,570 --> 00:56:05,990 So there are different degrees of selectivity, right? 1340 00:56:05,990 --> 00:56:07,970 Not just does it respond differentially, 1341 00:56:07,970 --> 00:56:09,740 but how selective is it? 1342 00:56:09,740 --> 00:56:11,210 Oh, boy, I'm going way over time. 1343 00:56:11,210 --> 00:56:13,370 I'm sorry. 1344 00:56:13,370 --> 00:56:17,127 So you guys did great thinking through these things. 1345 00:56:17,127 --> 00:56:19,460 And, of course, I didn't get halfway through my lecture. 1346 00:56:19,460 --> 00:56:21,590 That's OK, we'll roll over the best parts for later, 1347 00:56:21,590 --> 00:56:22,965 and the ones that aren't that fun 1348 00:56:22,965 --> 00:56:25,430 will just go by the wayside. 1349 00:56:25,430 --> 00:56:27,810 I will put notes on the rest of some of these things, 1350 00:56:27,810 --> 00:56:29,605 but I think all of you guys pulled out-- 1351 00:56:29,605 --> 00:56:31,730 just thinking hard about it and using common sense, 1352 00:56:31,730 --> 00:56:33,890 you can see that a lot of experimental design 1353 00:56:33,890 --> 00:56:35,370 is common sense. 1354 00:56:35,370 --> 00:56:37,990 All right, see you guys on Wednesday.