1 00:00:04,480 --> 00:00:10,039 Speaker 1: Brains love to look at faces, But why and what 2 00:00:10,200 --> 00:00:15,400 Speaker 1: is face blindness? And what is a super recognizer? Why 3 00:00:15,400 --> 00:00:19,079 Speaker 1: does your electrical plug sometimes look to you like a 4 00:00:19,120 --> 00:00:24,160 Speaker 1: little face? Did aliens plant a signal for us on Mars? 5 00:00:24,600 --> 00:00:28,200 Speaker 1: Or are we looking at a quirk of our own brains? 6 00:00:28,520 --> 00:00:30,200 Speaker 1: What does any of this have to do with the 7 00:00:30,640 --> 00:00:35,360 Speaker 1: neurologist Oliver Sachs and his inability to recognize most of 8 00:00:35,400 --> 00:00:38,160 Speaker 1: the people in his life, or looking at a magazine 9 00:00:38,280 --> 00:00:44,480 Speaker 1: upside down? Or what the mistakes in computerized face recognition technology? 10 00:00:45,080 --> 00:00:51,199 Speaker 1: Tell us Welcome to Inner Cosmos with me David Eagleman. 11 00:00:51,560 --> 00:00:55,000 Speaker 1: I'm a neuroscientist and an author at Stanford, and in 12 00:00:55,040 --> 00:00:59,600 Speaker 1: these episodes we dive deeply into our three pound universe 13 00:01:00,080 --> 00:01:08,880 Speaker 1: to understand some of the most surprising aspects of our lives. 14 00:01:12,440 --> 00:01:17,640 Speaker 1: Today's episode is about faces. Now, let's say I posed 15 00:01:18,080 --> 00:01:22,679 Speaker 1: the question to you, how does the brain recognize faces? 16 00:01:23,280 --> 00:01:26,600 Speaker 1: You may reasonably think, why is that even a question? 17 00:01:27,120 --> 00:01:32,440 Speaker 1: Recognizing people's faces isn't really that hard, But think about 18 00:01:32,480 --> 00:01:37,200 Speaker 1: how similar faces are to one another. They're insanely similar. 19 00:01:37,360 --> 00:01:42,160 Speaker 1: You have two eyes, nose, mouth, chin, ears, everyone's face 20 00:01:42,200 --> 00:01:46,840 Speaker 1: looks pretty much the same people's faces are pretty much 21 00:01:46,959 --> 00:01:51,240 Speaker 1: very tiny variations on a theme. The only reason you 22 00:01:51,320 --> 00:01:55,200 Speaker 1: are able to distinguish faces, literally thousands of faces from 23 00:01:55,240 --> 00:01:59,840 Speaker 1: one another, is because your brain devotes an enormous amount 24 00:01:59,880 --> 00:02:03,280 Speaker 1: of circuitry to it. It puts so much effort in 25 00:02:03,920 --> 00:02:07,360 Speaker 1: that it feels effortless to you. And we have all 26 00:02:07,440 --> 00:02:14,120 Speaker 1: this circuitry because face recognition is extremely salient. It's how 27 00:02:14,560 --> 00:02:19,359 Speaker 1: we recognize this package of identity from this other. Oh 28 00:02:19,440 --> 00:02:23,880 Speaker 1: that's my mother, that's my neighbor, that's my wife. There's 29 00:02:24,440 --> 00:02:27,440 Speaker 1: Tom Cruise and his face is similar but different from 30 00:02:27,800 --> 00:02:32,840 Speaker 1: Dwayne Johnson, and that's different from Joe Biden. It's massively 31 00:02:32,880 --> 00:02:37,400 Speaker 1: important for us as a social species to keep track 32 00:02:37,880 --> 00:02:42,200 Speaker 1: of identities. Your brain wants to know who it is 33 00:02:42,280 --> 00:02:45,679 Speaker 1: and everything that entails, like is that a friend or 34 00:02:45,720 --> 00:02:50,240 Speaker 1: an enemy? What are the risks and opportunities here? And 35 00:02:50,280 --> 00:02:53,040 Speaker 1: the answer to that is massively different. If you're looking 36 00:02:53,440 --> 00:02:57,160 Speaker 1: at the face of Ted Bundy versus your best pal, 37 00:02:57,480 --> 00:03:00,040 Speaker 1: and keep in mind that to a space alien and 38 00:03:00,080 --> 00:03:04,680 Speaker 1: these faces would be essentially indistinguishable. Now, when you're trying 39 00:03:04,680 --> 00:03:09,040 Speaker 1: to distinguish faces, there are lots of other cues like 40 00:03:09,520 --> 00:03:13,800 Speaker 1: hairline and hairstyle and height, and details of their skin 41 00:03:14,040 --> 00:03:18,519 Speaker 1: from color to smoothness, and even cues like how they 42 00:03:18,639 --> 00:03:23,600 Speaker 1: walk and their clothing choice. And context also plays a 43 00:03:23,680 --> 00:03:28,600 Speaker 1: large role. You've probably experienced some time when you saw 44 00:03:28,680 --> 00:03:31,320 Speaker 1: somebody you know well, like a neighbor or a colleague, 45 00:03:31,639 --> 00:03:35,640 Speaker 1: but it's in an unfamiliar setting and you struggle to 46 00:03:35,720 --> 00:03:39,720 Speaker 1: recognize them immediately. Why does that happen Because our brains 47 00:03:39,880 --> 00:03:44,760 Speaker 1: use contextual cues when identifying people, So our brains rely 48 00:03:45,000 --> 00:03:49,960 Speaker 1: not just on facial features, but also on surrounding context 49 00:03:49,960 --> 00:03:54,760 Speaker 1: of all sorts to identify people. But despite all these 50 00:03:54,800 --> 00:03:58,920 Speaker 1: other cues that we use, just note that facial recognition 51 00:03:59,200 --> 00:04:02,640 Speaker 1: is possible even over zoom, where many of those other 52 00:04:02,800 --> 00:04:06,160 Speaker 1: cues are gone. And even if you've got a bunch 53 00:04:06,240 --> 00:04:09,400 Speaker 1: of movie stars to put on wigs and change their 54 00:04:09,440 --> 00:04:12,440 Speaker 1: hairlines and so on, you'd still be pretty good at 55 00:04:12,440 --> 00:04:17,840 Speaker 1: distinguishing them just based on the details of the face itself. 56 00:04:18,360 --> 00:04:20,800 Speaker 1: And I think there are at least two ways to 57 00:04:20,880 --> 00:04:26,479 Speaker 1: appreciate how hard a challenge face recognition is to the brain. 58 00:04:26,880 --> 00:04:30,640 Speaker 1: The first is to understand how difficult it is to 59 00:04:30,680 --> 00:04:35,960 Speaker 1: get computers to recognize and distinguish human faces. Now you 60 00:04:36,040 --> 00:04:40,600 Speaker 1: might think, aren't we pretty good at this with modern technology. Yes, 61 00:04:41,000 --> 00:04:45,400 Speaker 1: but this has been a sixty year slog by very 62 00:04:45,440 --> 00:04:48,400 Speaker 1: smart people in the computer science world. Starting in the 63 00:04:48,480 --> 00:04:53,000 Speaker 1: nineteen sixties, people realized that face recognition was a really 64 00:04:53,040 --> 00:04:56,839 Speaker 1: hard problem, and they've been climbing that hill ever since. 65 00:04:57,279 --> 00:05:01,000 Speaker 1: So the way that researchers have gotten face recognition to 66 00:05:01,120 --> 00:05:05,800 Speaker 1: work on computers nowadays is by busting the problem up 67 00:05:06,040 --> 00:05:12,960 Speaker 1: into different computational stages. So first, you have algorithms just 68 00:05:13,000 --> 00:05:16,479 Speaker 1: for detecting a face, like is there a face out there? 69 00:05:17,200 --> 00:05:20,240 Speaker 1: Then you have a second step where you segment this 70 00:05:20,400 --> 00:05:23,880 Speaker 1: face from the background and figure out how the face 71 00:05:24,080 --> 00:05:26,480 Speaker 1: is aligned to tell you about the size of the 72 00:05:26,520 --> 00:05:29,760 Speaker 1: face and the illumination on the face and even the 73 00:05:29,839 --> 00:05:32,920 Speaker 1: pose of the face. Then you have a third step 74 00:05:33,160 --> 00:05:37,159 Speaker 1: where you extract the facial features, like here's the length 75 00:05:37,200 --> 00:05:39,960 Speaker 1: of the nose and the distance between the eyes, and 76 00:05:40,000 --> 00:05:42,520 Speaker 1: the shape of the mouth and the position of the ears. 77 00:05:42,880 --> 00:05:46,160 Speaker 1: All these things you pinpoint and you measure them, and 78 00:05:46,200 --> 00:05:50,279 Speaker 1: then in the fourth step, you match those numbers against 79 00:05:50,320 --> 00:05:53,719 Speaker 1: a database of faces to see if you can identify 80 00:05:53,800 --> 00:05:57,320 Speaker 1: this one. So this is a really complicated algorithm and 81 00:05:57,400 --> 00:06:01,160 Speaker 1: it's instructive to see how computer get things wrong. I 82 00:06:01,200 --> 00:06:05,400 Speaker 1: saw a great photograph where it's a family sitting on 83 00:06:05,440 --> 00:06:08,960 Speaker 1: their porch at Halloween, and the computer has identified each 84 00:06:09,000 --> 00:06:12,160 Speaker 1: of their faces and put a little label next to it, 85 00:06:12,279 --> 00:06:16,159 Speaker 1: like Henry and John and Sarah, and then it identifies 86 00:06:16,480 --> 00:06:21,040 Speaker 1: another face and says unknown person. But that face is 87 00:06:21,200 --> 00:06:23,880 Speaker 1: the Jack O lantern, the pumpkin that's sitting there on 88 00:06:23,920 --> 00:06:27,400 Speaker 1: the porch. Now, a human observer would never make that 89 00:06:27,560 --> 00:06:30,479 Speaker 1: kind of mistake. But this goes to show that it's 90 00:06:30,560 --> 00:06:36,760 Speaker 1: a very computationally challenging problem. So some version of this 91 00:06:36,880 --> 00:06:41,640 Speaker 1: very complicated algorithm, probably not exactly the same, but equally complex. 92 00:06:42,120 --> 00:06:44,960 Speaker 1: This is what your brain is doing under the hood 93 00:06:45,600 --> 00:06:49,280 Speaker 1: when you just glanced at some billboard and you think, oh, 94 00:06:49,400 --> 00:06:52,440 Speaker 1: that's Paul Giamati. I love that guy. And we'll come 95 00:06:52,520 --> 00:06:54,920 Speaker 1: back to what the brain is doing in just a moment, 96 00:06:55,320 --> 00:06:57,280 Speaker 1: but I just want to say for now that quite 97 00:06:57,279 --> 00:07:02,520 Speaker 1: often we find that the tasks that seem most effortless 98 00:07:02,560 --> 00:07:08,039 Speaker 1: to us are the things underpinned by the most brain circuitry. 99 00:07:08,440 --> 00:07:10,760 Speaker 1: And I want to look at a second way to 100 00:07:10,880 --> 00:07:15,360 Speaker 1: appreciate how hard a challenge face recognition is for the brain. 101 00:07:16,120 --> 00:07:20,920 Speaker 1: Just look at a bunch of Golden Retriever faces. If 102 00:07:20,960 --> 00:07:23,480 Speaker 1: you were in a situation where you were hanging out 103 00:07:23,520 --> 00:07:27,880 Speaker 1: with hundreds or thousands of Golden Retriever dogs, do they 104 00:07:27,920 --> 00:07:33,080 Speaker 1: really look that different to you? Or consider horse faces 105 00:07:33,400 --> 00:07:39,560 Speaker 1: or squirrel faces or cowfaces. Obviously we can imagine some 106 00:07:39,760 --> 00:07:42,200 Speaker 1: faces at the extreme that look a little bit different, 107 00:07:42,200 --> 00:07:44,960 Speaker 1: But in general, if I dropped you into the middle 108 00:07:44,960 --> 00:07:48,480 Speaker 1: of a giant ranch and you looked at thousands of cows, 109 00:07:48,680 --> 00:07:52,840 Speaker 1: you'd say, okay, well, each face has two eyes and 110 00:07:52,920 --> 00:07:55,640 Speaker 1: a snout and a mouth, and there's just not that 111 00:07:55,760 --> 00:07:59,880 Speaker 1: much difference from one face to another. You wouldn't meet 112 00:08:00,400 --> 00:08:03,400 Speaker 1: recognized if one of those cows had been in a 113 00:08:03,440 --> 00:08:06,760 Speaker 1: movie like oh, that's that famous cow, or if one 114 00:08:06,800 --> 00:08:10,600 Speaker 1: of those cows was wanted by the legal system, like 115 00:08:11,120 --> 00:08:14,200 Speaker 1: wait a minute, I recognize that cow from the poster 116 00:08:14,320 --> 00:08:17,000 Speaker 1: of the post office. And of course, if one of 117 00:08:17,040 --> 00:08:19,960 Speaker 1: these cows was looking at you and a bunch of 118 00:08:19,960 --> 00:08:23,840 Speaker 1: other humans, she would feel exactly the same way. There's 119 00:08:23,920 --> 00:08:28,480 Speaker 1: not that much difference between human faces. Now side note, 120 00:08:28,520 --> 00:08:31,560 Speaker 1: which we'll return to. If you were the rancher, you 121 00:08:31,640 --> 00:08:35,319 Speaker 1: could get better at distinguishing cow faces. This is because 122 00:08:35,360 --> 00:08:40,319 Speaker 1: of brain plasticity and because distinguishing the cows is salient 123 00:08:40,640 --> 00:08:44,200 Speaker 1: to your rancher brain. But for now, what I want 124 00:08:44,200 --> 00:08:49,079 Speaker 1: to surface is just how similar faces are before you 125 00:08:49,160 --> 00:08:54,280 Speaker 1: get good at them from experience. Faces are unbelievably similar. 126 00:08:54,640 --> 00:08:58,400 Speaker 1: And one expression of this difficulty in distinguishing faces comes 127 00:08:58,400 --> 00:09:02,720 Speaker 1: from a psychological effe fact called the other race effect, 128 00:09:03,200 --> 00:09:06,160 Speaker 1: and this refers to the tendency of people to have 129 00:09:06,679 --> 00:09:11,640 Speaker 1: more difficulty recognizing faces that belong to races that are 130 00:09:11,640 --> 00:09:14,200 Speaker 1: not their own race. This phenomenon is also known as 131 00:09:14,240 --> 00:09:18,440 Speaker 1: the cross race effect or the own race bias, and 132 00:09:18,520 --> 00:09:21,400 Speaker 1: it's been widely studied and you see it across all cultures. 133 00:09:21,800 --> 00:09:24,440 Speaker 1: Now people sometimes hear this and they take this as 134 00:09:24,440 --> 00:09:28,679 Speaker 1: evidence for racism. It's not racism in the sense of 135 00:09:28,720 --> 00:09:33,040 Speaker 1: you treating your own group better and other groups more poorly. Instead, 136 00:09:33,120 --> 00:09:36,440 Speaker 1: the other race effect simply results from the fact that 137 00:09:36,480 --> 00:09:39,880 Speaker 1: your brain gets trained up on the faces around you, 138 00:09:40,360 --> 00:09:44,640 Speaker 1: and you are better at recognizing those faces than other faces. 139 00:09:45,000 --> 00:09:48,800 Speaker 1: So if you grew up in Cambodia, you will be 140 00:09:48,960 --> 00:09:53,160 Speaker 1: excellent and distinguishing Cambodian faces, and maybe not so great 141 00:09:53,240 --> 00:09:57,080 Speaker 1: at distinguishing Norwegian faces. If you grew up in Norway, 142 00:09:57,600 --> 00:10:01,480 Speaker 1: you're excellent and distinguishing Norwegian f but not so great 143 00:10:01,559 --> 00:10:06,000 Speaker 1: at distinguishing Zimbabwean faces. And if you grew up in Zimbabwe, 144 00:10:06,080 --> 00:10:09,360 Speaker 1: you're not so good at Intuit Eskimo faces and so on. 145 00:10:09,760 --> 00:10:14,480 Speaker 1: So your brain becomes attuned to the type of faces 146 00:10:14,520 --> 00:10:18,920 Speaker 1: that you see most often, and this heightened sensitivity makes 147 00:10:19,000 --> 00:10:23,679 Speaker 1: us better at distinguishing among those familiar faces. As a 148 00:10:23,720 --> 00:10:27,280 Speaker 1: side note, can you mitigate the other race effect? Sure, 149 00:10:27,360 --> 00:10:31,920 Speaker 1: it's just about exposure. You can improve the ability to 150 00:10:32,000 --> 00:10:37,480 Speaker 1: recognize other race faces through training and exposure. Now, Hollywood 151 00:10:37,520 --> 00:10:39,640 Speaker 1: has been on the forefront of this for a long time, 152 00:10:39,960 --> 00:10:42,960 Speaker 1: making sure that there's a mixture that we all get 153 00:10:42,960 --> 00:10:46,800 Speaker 1: exposed to different sorts of faces, and that's presumably very useful. 154 00:10:47,160 --> 00:10:50,040 Speaker 1: It could be noted that Hollywood has tended to be 155 00:10:50,160 --> 00:10:52,960 Speaker 1: mostly in love with white faces and black faces, and 156 00:10:53,000 --> 00:10:55,080 Speaker 1: that leaves a lot of people out of the mix. 157 00:10:55,480 --> 00:10:58,840 Speaker 1: So we the viewing audience don't necessarily get much better 158 00:10:58,880 --> 00:11:03,320 Speaker 1: at Cambodian face or Norwegian faces, or Inuit Eskimo faces 159 00:11:03,400 --> 00:11:07,680 Speaker 1: or whatever. But it's a start to training our brains 160 00:11:07,760 --> 00:11:11,320 Speaker 1: on a broader diet of faces. So the other race 161 00:11:11,360 --> 00:11:14,920 Speaker 1: effect underscores how our brains can get tuned to the 162 00:11:14,960 --> 00:11:19,040 Speaker 1: social world around us, and it highlights our natural tendency 163 00:11:19,400 --> 00:11:35,280 Speaker 1: to recognize familiar faces more easily. Okay, so where are 164 00:11:35,280 --> 00:11:39,320 Speaker 1: we so far. It's really hard to distinguish faces unless 165 00:11:39,360 --> 00:11:42,360 Speaker 1: your brain has had lots and lots of practice at it. 166 00:11:42,840 --> 00:11:46,920 Speaker 1: But again it seems effortless to us now. Even though 167 00:11:47,120 --> 00:11:50,680 Speaker 1: we're not always great at distinguishing faces from one another, 168 00:11:50,880 --> 00:11:54,800 Speaker 1: unless we've had lots of practice, we are highly programmed 169 00:11:54,880 --> 00:11:58,640 Speaker 1: to identify that there is a face there and to 170 00:11:58,800 --> 00:12:02,640 Speaker 1: zoom in on it. Why because faces carry so much 171 00:12:02,640 --> 00:12:05,840 Speaker 1: information for us, And this is why you sometimes look 172 00:12:05,840 --> 00:12:09,240 Speaker 1: at the electrical plug in your wall and you see 173 00:12:09,240 --> 00:12:12,400 Speaker 1: a little face, or you see a pattern of burn 174 00:12:12,480 --> 00:12:14,800 Speaker 1: marks on a piece of toast and you think, oh, 175 00:12:14,840 --> 00:12:17,679 Speaker 1: that looks just like a face, or you look at 176 00:12:17,760 --> 00:12:20,080 Speaker 1: the surface of the moon and you think, oh, there's 177 00:12:20,200 --> 00:12:23,440 Speaker 1: a man there looking at me. In science, this is 178 00:12:23,480 --> 00:12:30,360 Speaker 1: called paradolia. Paradolia refers to our tendency to perceive specific 179 00:12:30,520 --> 00:12:35,240 Speaker 1: and often meaningful images in random patterns, and the most 180 00:12:35,280 --> 00:12:37,840 Speaker 1: common usage of this word is in the case of 181 00:12:37,960 --> 00:12:42,920 Speaker 1: perceiving the pattern of a face in random unrelated objects. 182 00:12:42,920 --> 00:12:45,760 Speaker 1: And we've all been there where your brain says, oh, 183 00:12:46,000 --> 00:12:47,880 Speaker 1: that looks like a face when you're looking at a 184 00:12:48,280 --> 00:12:53,960 Speaker 1: cloud or a rock formation, or some arrangement of fruits 185 00:12:53,960 --> 00:12:55,920 Speaker 1: on a table. And I don't know if you've ever 186 00:12:55,920 --> 00:12:59,480 Speaker 1: seen this image from the surface of Mars. I'll put 187 00:12:59,480 --> 00:13:03,520 Speaker 1: it on the show. But on the bumpy landscape of Mars, 188 00:13:03,720 --> 00:13:06,120 Speaker 1: there are some bumps that happen to fall into the 189 00:13:06,160 --> 00:13:09,120 Speaker 1: pattern that look like two y's and a nose and 190 00:13:09,160 --> 00:13:12,040 Speaker 1: a mouth. And many people go nuts about this image 191 00:13:12,080 --> 00:13:15,520 Speaker 1: because they suggest that it's a sculpture of a human 192 00:13:15,679 --> 00:13:20,080 Speaker 1: face planted there by aliens. But probably not. This is 193 00:13:20,200 --> 00:13:28,199 Speaker 1: merely our brains experiencing garden variety paradolia. We assign patterns 194 00:13:28,280 --> 00:13:33,560 Speaker 1: to random inputs, and in particular, we love to see faces. Okay, 195 00:13:33,800 --> 00:13:36,920 Speaker 1: but why does this happen to us so commonly. Well, 196 00:13:37,400 --> 00:13:42,000 Speaker 1: it's because we are hardwired to see faces everywhere, and 197 00:13:42,040 --> 00:13:46,920 Speaker 1: the reason is that faces are crucial for social interaction. 198 00:13:47,600 --> 00:13:51,840 Speaker 1: Seeing a face tells us about a person's identity, their 199 00:13:52,000 --> 00:13:55,920 Speaker 1: emotional state, whether they're a friend or a foe. This 200 00:13:56,080 --> 00:14:00,680 Speaker 1: ability helped our ancestors to survive by recognize is saying oh, 201 00:14:00,720 --> 00:14:04,439 Speaker 1: that's a tribe member. Oh, that's a potential threat. In fact, 202 00:14:04,520 --> 00:14:08,360 Speaker 1: even newborn babies prefer to look at faces. They look 203 00:14:08,360 --> 00:14:12,640 Speaker 1: at face like stimuli more than other stimuli. So take 204 00:14:13,040 --> 00:14:16,800 Speaker 1: two little circles and a little square beneath that, and 205 00:14:16,840 --> 00:14:20,560 Speaker 1: a little rectangle underneath that, they'll stare at that as 206 00:14:20,600 --> 00:14:24,800 Speaker 1: opposed to the same shapes in a different configuration. And 207 00:14:24,840 --> 00:14:30,360 Speaker 1: this underscores that brains are primed to focus on faces. 208 00:14:30,800 --> 00:14:35,120 Speaker 1: Paridolia is just an extension of this bias to recognize faces. 209 00:14:35,160 --> 00:14:39,960 Speaker 1: It's our brain's way of staying vigilant because in the wild, 210 00:14:40,160 --> 00:14:44,120 Speaker 1: it is safer to mistakenly see a face where there 211 00:14:44,200 --> 00:14:48,720 Speaker 1: isn't one than to miss a real face. So this 212 00:14:49,080 --> 00:14:53,720 Speaker 1: hypersensitivity to faces means the difference between life and death 213 00:14:53,760 --> 00:14:58,480 Speaker 1: in some situations, like spotting a lurking predator or an 214 00:14:58,520 --> 00:15:02,920 Speaker 1: aggressive intruder. By the way, I'll just mention that paradolia 215 00:15:03,040 --> 00:15:06,720 Speaker 1: is something that artists have always exploited. They turn random 216 00:15:06,760 --> 00:15:12,320 Speaker 1: patterns or abstract shapes into recognizable faces. Just as one example, 217 00:15:12,440 --> 00:15:15,880 Speaker 1: one of my favorite artists, Salvador Dhli. He very often 218 00:15:15,920 --> 00:15:20,800 Speaker 1: takes advantage of paradolia to create double images that play 219 00:15:20,880 --> 00:15:24,280 Speaker 1: with the viewer's perception. You can see this as a 220 00:15:24,520 --> 00:15:30,200 Speaker 1: rock formation or as a face. So paradolia highlights how 221 00:15:30,200 --> 00:15:32,520 Speaker 1: our brains are constantly trying to make sense of the 222 00:15:32,560 --> 00:15:37,200 Speaker 1: world around us, even when there isn't anything immediately recognizable. 223 00:15:37,720 --> 00:15:39,640 Speaker 1: And I want to give just one other angle on 224 00:15:39,760 --> 00:15:43,600 Speaker 1: paradolia here. If you're a regular listener to this podcast, 225 00:15:44,040 --> 00:15:46,480 Speaker 1: you know that I often talk about how the brain's 226 00:15:46,720 --> 00:15:50,600 Speaker 1: main job is to build an internal model of the 227 00:15:50,680 --> 00:15:56,720 Speaker 1: outside world, and our model influences our perception. So whatever 228 00:15:56,800 --> 00:16:00,920 Speaker 1: you specialize in, you'll see lots of that world. We're 229 00:16:00,960 --> 00:16:05,480 Speaker 1: all specialists at faces, and we often see those where 230 00:16:05,480 --> 00:16:09,480 Speaker 1: there isn't actually one, but it generalizes beyond that. So 231 00:16:09,640 --> 00:16:11,680 Speaker 1: for me personally, when I look around the world, I 232 00:16:11,760 --> 00:16:15,800 Speaker 1: tend to impose shapes that look like brains. I think, oh, 233 00:16:15,800 --> 00:16:18,880 Speaker 1: there's a brain, or there's a midsadgital section, or there's 234 00:16:18,920 --> 00:16:23,320 Speaker 1: a cerebellum. My friends who are pilots tend to see 235 00:16:23,520 --> 00:16:28,440 Speaker 1: airplane shapes and runways. My collogist friends see mushroom shapes, 236 00:16:28,840 --> 00:16:32,680 Speaker 1: my dermatologists friends see melanomas, and so on. Whatever you 237 00:16:32,840 --> 00:16:37,360 Speaker 1: specialize in makes you an expert in detecting that in 238 00:16:37,440 --> 00:16:41,840 Speaker 1: the world, and you impose that interpretation on lots of things. 239 00:16:42,280 --> 00:16:48,240 Speaker 1: It's paridolia writ large. Okay, so back to faces in particular. 240 00:16:48,680 --> 00:16:51,600 Speaker 1: Now we're ready to turn to how the brain does 241 00:16:51,640 --> 00:16:55,600 Speaker 1: its magic with face recognition. One of the key areas 242 00:16:55,640 --> 00:16:58,400 Speaker 1: in our brain that does this is in the temporal lobe. 243 00:16:58,440 --> 00:17:02,760 Speaker 1: It's a little region called the fusiform face area or 244 00:17:03,160 --> 00:17:07,000 Speaker 1: FFA fusiform face area. Now, we can see in brain 245 00:17:07,040 --> 00:17:12,439 Speaker 1: imaging that this area is particularly sensitive to faces. But 246 00:17:12,680 --> 00:17:15,239 Speaker 1: what's interesting is it's not just any part of the 247 00:17:15,280 --> 00:17:19,720 Speaker 1: face that gets attention from the FFA, it's the holistic 248 00:17:19,880 --> 00:17:25,080 Speaker 1: view of the face. So the FFA responds strongly to 249 00:17:25,200 --> 00:17:30,200 Speaker 1: the overall configuration of a face rather than individual features. 250 00:17:30,480 --> 00:17:34,720 Speaker 1: So that allows us to quickly recognize faces and familiar faces, 251 00:17:34,880 --> 00:17:37,040 Speaker 1: even if we only get a quick glimpse of them. 252 00:17:37,520 --> 00:17:40,639 Speaker 1: When you look at these brain imaging studies, for example, 253 00:17:40,720 --> 00:17:46,400 Speaker 1: with fMRI, you find that this fusiform face area responds 254 00:17:46,560 --> 00:17:51,200 Speaker 1: more to familiar faces than to unfamiliar faces. And given 255 00:17:51,240 --> 00:17:54,240 Speaker 1: what we saw a moment ago with the other race effect, 256 00:17:54,800 --> 00:17:57,960 Speaker 1: the FFA is more active when you're viewing faces of 257 00:17:58,000 --> 00:18:01,800 Speaker 1: your own race compared to faces of other races. So again, 258 00:18:02,080 --> 00:18:06,120 Speaker 1: just like recognizing the distinction between different cows if you're 259 00:18:06,200 --> 00:18:10,200 Speaker 1: the rancher, or distinguishing faces that you grew up with 260 00:18:10,240 --> 00:18:13,200 Speaker 1: as opposed to other cultures that you rarely see. Your 261 00:18:13,320 --> 00:18:18,600 Speaker 1: visual system is specialized to efficiently process faces that you 262 00:18:18,680 --> 00:18:21,960 Speaker 1: see a lot more. Now back to this issue about 263 00:18:22,040 --> 00:18:27,159 Speaker 1: recognizing the face holistically rather than detail by detail. This 264 00:18:27,280 --> 00:18:31,480 Speaker 1: is what underlies the phenomenon that's called the inversion effect, 265 00:18:31,600 --> 00:18:34,840 Speaker 1: which is when you see a face that is upside down, 266 00:18:35,480 --> 00:18:39,359 Speaker 1: it's really hard to recognize. So try this. Just open 267 00:18:39,400 --> 00:18:44,280 Speaker 1: a magazine upside down and flip through the pages and 268 00:18:44,280 --> 00:18:47,159 Speaker 1: see if you can recognize the people in it. And 269 00:18:47,200 --> 00:18:49,240 Speaker 1: then when you flip it right side up, it's a 270 00:18:49,280 --> 00:18:52,440 Speaker 1: whole different experience. And this is because this part of 271 00:18:52,480 --> 00:18:57,080 Speaker 1: the brain FFA is super specialized on faces, and it's 272 00:18:57,119 --> 00:19:00,960 Speaker 1: always seen faces in a particular way, and it's recognizing 273 00:19:00,960 --> 00:19:04,320 Speaker 1: the whole big picture with two eyes above the nose, 274 00:19:04,600 --> 00:19:08,160 Speaker 1: above the mouth, and if you throw a ranch in that, 275 00:19:08,600 --> 00:19:11,280 Speaker 1: it doesn't function so well. The key thing is that 276 00:19:11,359 --> 00:19:15,840 Speaker 1: this inversion effect is much stronger for faces than for 277 00:19:15,960 --> 00:19:22,800 Speaker 1: other objects, which highlights this very specialized nature of face processing. Now, 278 00:19:22,880 --> 00:19:25,479 Speaker 1: what's wild is that this part of the brain is 279 00:19:25,600 --> 00:19:29,320 Speaker 1: trying to see faces everywhere, as we saw in paridolia. 280 00:19:29,800 --> 00:19:32,520 Speaker 1: And so if someone puts an electrode into this part 281 00:19:32,520 --> 00:19:35,520 Speaker 1: of your brain, then you might look at, let's say, 282 00:19:35,520 --> 00:19:38,600 Speaker 1: a soccer ball, and when the electrode turns on and 283 00:19:38,680 --> 00:19:42,760 Speaker 1: stimulates this area with electricity, the hexagons of the soccer 284 00:19:42,800 --> 00:19:46,480 Speaker 1: ball seem to you to become the eyes and the 285 00:19:46,560 --> 00:19:49,879 Speaker 1: mouth of a face. So activity in this part of 286 00:19:49,880 --> 00:19:54,040 Speaker 1: the brain is always working over time to detect faces 287 00:19:54,040 --> 00:19:57,280 Speaker 1: in the world, and sometimes it does this by imposing 288 00:19:57,320 --> 00:20:02,240 Speaker 1: the interpretation onto the canvas of the world. Now, consistent 289 00:20:02,240 --> 00:20:06,679 Speaker 1: with what I mentioned before, the FFA isn't specialized just 290 00:20:06,800 --> 00:20:11,000 Speaker 1: for faces. It also plays a role in recognizing other 291 00:20:11,119 --> 00:20:15,920 Speaker 1: things that we're highly familiar with, like cars for car enthusiasts, 292 00:20:16,119 --> 00:20:20,480 Speaker 1: or birds for bird watchers. So that tells us slightly 293 00:20:20,520 --> 00:20:24,280 Speaker 1: more generally that this part of the brain specializes in 294 00:20:24,320 --> 00:20:28,159 Speaker 1: recognizing very detailed things that are relevant and that we 295 00:20:28,200 --> 00:20:31,920 Speaker 1: have a lot of experience seeing. So back to faces, 296 00:20:32,600 --> 00:20:35,640 Speaker 1: the FFA, it's not just that it's sensitive to faces. 297 00:20:36,040 --> 00:20:38,600 Speaker 1: It's actually critical if you want to see a face. 298 00:20:38,760 --> 00:20:42,240 Speaker 1: If you get damaged to the FFA or to its connections, 299 00:20:42,480 --> 00:20:47,159 Speaker 1: then you become unable to recognize faces. And will return 300 00:20:47,200 --> 00:20:49,440 Speaker 1: to that in a moment, but first I just want 301 00:20:49,480 --> 00:20:52,879 Speaker 1: to make one more thing clear. Like almost everything in 302 00:20:52,920 --> 00:20:56,840 Speaker 1: the brain, it's not just one area that's involved. Face 303 00:20:56,920 --> 00:21:02,480 Speaker 1: recognition uses a broader, specialized network. Beyond the FFA, you 304 00:21:02,560 --> 00:21:06,600 Speaker 1: have other areas involved, like the occipital face area and 305 00:21:06,640 --> 00:21:09,919 Speaker 1: the superior temporal sulcus. Don't worry about the details, but 306 00:21:09,960 --> 00:21:11,919 Speaker 1: the thing I want to surface here is that some 307 00:21:12,080 --> 00:21:15,320 Speaker 1: areas are more focused on the detailed features of the 308 00:21:15,359 --> 00:21:19,840 Speaker 1: face and others are sensitive to facial expressions and movements. 309 00:21:20,200 --> 00:21:24,119 Speaker 1: And the key is that together these areas form a 310 00:21:24,320 --> 00:21:29,360 Speaker 1: larger network that allows us to recognize faces and interpret 311 00:21:29,400 --> 00:21:32,080 Speaker 1: their expressions well. Your brain's also doing a lot of 312 00:21:32,119 --> 00:21:37,560 Speaker 1: work to gauge a person's emotions and then navigating social 313 00:21:37,560 --> 00:21:42,680 Speaker 1: interactions by carefully watching the reactions of someone's facial expressions. 314 00:21:43,200 --> 00:21:47,160 Speaker 1: We humans are super specialists at this because our well 315 00:21:47,200 --> 00:21:53,040 Speaker 1: being generally depends on our ability to recognize and understand 316 00:21:53,080 --> 00:21:55,439 Speaker 1: the people around us. And just to give you a 317 00:21:55,480 --> 00:21:59,080 Speaker 1: sense of how all these networks interact, there are studies 318 00:21:59,119 --> 00:22:04,000 Speaker 1: showing that you're interpretation of facial expressions is influenced by 319 00:22:04,040 --> 00:22:08,080 Speaker 1: the broader context. So just as an example, in one study, 320 00:22:08,400 --> 00:22:12,640 Speaker 1: you're shown faces with neutral expressions, but you're given different 321 00:22:13,119 --> 00:22:16,639 Speaker 1: contextual information about the person. So if you're told the 322 00:22:16,720 --> 00:22:21,879 Speaker 1: person is happy, you perceive the neutral face as slightly smiling, 323 00:22:22,160 --> 00:22:25,520 Speaker 1: and if you're told this person is sad, you perceive 324 00:22:25,600 --> 00:22:29,320 Speaker 1: that same face as slightly frowning. So, in other words, 325 00:22:29,320 --> 00:22:33,720 Speaker 1: your brain combines the facial information with the context of 326 00:22:33,800 --> 00:22:39,240 Speaker 1: the larger situation whenever you're interpreting emotions and for that matter, intentions. 327 00:22:39,720 --> 00:22:43,480 Speaker 1: So putting together the giant networks in the brain involved 328 00:22:43,480 --> 00:22:47,760 Speaker 1: in identifying and interpreting faces, the thing that becomes clear 329 00:22:47,840 --> 00:22:52,840 Speaker 1: is that we are massively finely tuned to process faces 330 00:22:53,320 --> 00:22:58,639 Speaker 1: in a way that benefits are social interactions and are survival. Now, 331 00:22:58,960 --> 00:23:02,119 Speaker 1: if you've been listening to other episodes of this podcast, 332 00:23:02,160 --> 00:23:05,560 Speaker 1: you'll know that I'm obsessed by the differences in the 333 00:23:05,600 --> 00:23:09,280 Speaker 1: internal life from person to person, from head to head. 334 00:23:09,800 --> 00:23:13,080 Speaker 1: And one place this really comes up is with individual 335 00:23:13,200 --> 00:23:18,560 Speaker 1: variability in face recognition. So, while most people are fairly 336 00:23:18,560 --> 00:23:23,080 Speaker 1: good at recognizing faces, there's a shocking amount of variation 337 00:23:23,520 --> 00:23:26,359 Speaker 1: in this ability. Some people are exceptionally good at it, 338 00:23:26,600 --> 00:23:29,600 Speaker 1: and other people struggle. And this brings us to the 339 00:23:29,680 --> 00:23:34,919 Speaker 1: concepts of super recognizers on the one hand, and face 340 00:23:35,160 --> 00:23:38,520 Speaker 1: blindness on the other hand. So let's start with the 341 00:23:38,880 --> 00:23:43,440 Speaker 1: super recognizers. These are people who have an extraordinary ability 342 00:23:43,760 --> 00:23:48,680 Speaker 1: to recognize faces. They can remember and identify faces even 343 00:23:48,720 --> 00:23:52,760 Speaker 1: after just brief encounters or after long periods of time. 344 00:23:53,040 --> 00:23:57,159 Speaker 1: Super recognizers are terrific at picking out their acquaintances and 345 00:23:57,320 --> 00:24:01,240 Speaker 1: really large crowds, and they sometimes end up getting employed 346 00:24:01,240 --> 00:24:05,879 Speaker 1: in security or law enforcement to identify perpetrators. Like they'll 347 00:24:05,880 --> 00:24:09,040 Speaker 1: be hired to watch the video of let's say a 348 00:24:09,080 --> 00:24:12,280 Speaker 1: subway entrance and you'll see thousands of people flowing in 349 00:24:12,320 --> 00:24:15,040 Speaker 1: and out, and they'll just stare at the feed for 350 00:24:15,160 --> 00:24:18,400 Speaker 1: hours or days, and then they'll say, oh, there's the guy. Now. 351 00:24:18,440 --> 00:24:20,920 Speaker 1: On the other end of the spectrum, we have what's 352 00:24:20,960 --> 00:24:25,000 Speaker 1: known as face blindness, and this is known as prosopagnosia. 353 00:24:25,320 --> 00:24:28,320 Speaker 1: This word comes from the Greek pro sopon meaning face, 354 00:24:28,480 --> 00:24:33,800 Speaker 1: and agnosia meaning lack of knowledge. So people with prosopagnosia 355 00:24:34,359 --> 00:24:38,879 Speaker 1: have a very hard time recognizing faces, even faces of 356 00:24:39,080 --> 00:24:42,120 Speaker 1: close family and friends. Now, by the way, this isn't 357 00:24:42,200 --> 00:24:46,360 Speaker 1: all or none. The condition ranges from mild to severe, 358 00:24:46,600 --> 00:24:52,640 Speaker 1: like difficulty recognizing familiar faces to an inability to distinguish 359 00:24:52,720 --> 00:24:55,240 Speaker 1: any faces at all. And at that far end of 360 00:24:55,280 --> 00:24:59,760 Speaker 1: the spectrum, just imagine not being able to recognize your 361 00:24:59,760 --> 00:25:03,880 Speaker 1: own spouse or child if you see them out of context. 362 00:25:04,320 --> 00:25:08,600 Speaker 1: And this is the reality for many people with prosopagnosia. Now, 363 00:25:08,800 --> 00:25:13,239 Speaker 1: how does prosopagnosia happen? Usually you're born with it. This 364 00:25:13,320 --> 00:25:17,160 Speaker 1: is called congenital prosopagnosia, just meaning it's present from birth. 365 00:25:17,800 --> 00:25:22,760 Speaker 1: In rarer cases, you can have acquired prosopagnosia, which just 366 00:25:22,840 --> 00:25:26,520 Speaker 1: means that you get it because of a brain injury 367 00:25:26,760 --> 00:25:29,760 Speaker 1: later in life, like a stroke or a traumatic brain injury. 368 00:25:30,000 --> 00:25:33,960 Speaker 1: And in these acquired cases, people notice the sudden change 369 00:25:33,960 --> 00:25:38,000 Speaker 1: in their ability to recognize faces, which is obviously distressing 370 00:25:38,040 --> 00:25:41,439 Speaker 1: and isolating. But what's fascinating is that people born with 371 00:25:41,600 --> 00:25:44,720 Speaker 1: this usually don't realize they have it because they've never 372 00:25:44,800 --> 00:25:48,280 Speaker 1: known anything different, and so they just get along by 373 00:25:48,359 --> 00:25:52,399 Speaker 1: using other cues like people's voices or their clothing or 374 00:25:52,440 --> 00:25:56,119 Speaker 1: the way they walk, and this is how they recognize people. 375 00:25:56,400 --> 00:25:59,679 Speaker 1: One example of a person who had prosopagnosia was the 376 00:25:59,800 --> 00:26:03,119 Speaker 1: neurologist Oliver Sacks, and he wrote an article in The 377 00:26:03,119 --> 00:26:05,560 Speaker 1: New Yorker in twenty ten on this. He talked about 378 00:26:05,560 --> 00:26:09,600 Speaker 1: his personal lifelong struggle with face blindness, but he never 379 00:26:09,720 --> 00:26:13,680 Speaker 1: realized there was something unusual about this until his middle age. 380 00:26:13,760 --> 00:26:16,520 Speaker 1: He'd always just been really bad with recognizing faces, but 381 00:26:16,600 --> 00:26:19,560 Speaker 1: that was just the way it was. But he went 382 00:26:19,600 --> 00:26:22,120 Speaker 1: to visit an older brother in Australia and they got 383 00:26:22,160 --> 00:26:24,720 Speaker 1: to talking and he realized that his brother had the 384 00:26:24,800 --> 00:26:27,919 Speaker 1: same problems with faces that he did, and it dawned 385 00:26:27,920 --> 00:26:31,520 Speaker 1: on him that this was something beyond normal variation, and 386 00:26:31,560 --> 00:26:35,320 Speaker 1: that they both had this trait of face blindness, and 387 00:26:35,480 --> 00:26:39,040 Speaker 1: he guessed that there was probably some distinctive genetic basis 388 00:26:39,040 --> 00:26:42,359 Speaker 1: to it. And very informally, I've noticed among people that 389 00:26:42,440 --> 00:26:47,680 Speaker 1: I know that those with prosopagnosia often don't enjoy movies 390 00:26:47,800 --> 00:26:50,720 Speaker 1: as much as other people, because in order to follow 391 00:26:50,800 --> 00:26:54,520 Speaker 1: a plot, especially if it's switching between an A story 392 00:26:54,520 --> 00:26:56,720 Speaker 1: and a B story and a C story, in order 393 00:26:56,760 --> 00:26:59,639 Speaker 1: to follow that plot, you really need to get it 394 00:27:00,040 --> 00:27:03,040 Speaker 1: when this actor comes into the scene who you haven't 395 00:27:03,040 --> 00:27:06,280 Speaker 1: seen for fifteen minutes and who's now wearing different clothes, 396 00:27:06,760 --> 00:27:09,679 Speaker 1: but this is just a continuation of his plot, and 397 00:27:09,760 --> 00:27:13,480 Speaker 1: everyone else in the audience recognizes him and immediately remembers 398 00:27:13,520 --> 00:27:15,399 Speaker 1: what was going on with him in the last act. 399 00:27:16,000 --> 00:27:19,000 Speaker 1: But just try to imagine how difficult it makes things 400 00:27:19,320 --> 00:27:22,840 Speaker 1: to follow along if you don't immediately recognize the person. 401 00:27:23,160 --> 00:27:25,639 Speaker 1: And by the way, if you don't have prosopagnosia, you 402 00:27:25,720 --> 00:27:28,240 Speaker 1: might be thinking what would that be like? How could 403 00:27:28,320 --> 00:27:32,960 Speaker 1: you not recognize a person? A common analogy is to 404 00:27:33,000 --> 00:27:36,440 Speaker 1: think about what it's like to walk through a forest 405 00:27:36,720 --> 00:27:40,200 Speaker 1: and just imagine if you had to remember and distinguish 406 00:27:40,520 --> 00:27:43,240 Speaker 1: every tree that you see. Imagine you had to have 407 00:27:43,280 --> 00:27:45,280 Speaker 1: a name for every tree, and you were quizzed on 408 00:27:45,359 --> 00:27:48,560 Speaker 1: it later. That sounds impossible to most of us, but 409 00:27:48,640 --> 00:27:51,840 Speaker 1: this is what daily life is like for the person 410 00:27:51,920 --> 00:27:56,160 Speaker 1: with face blindness. There wandering through a forest of people, 411 00:27:56,720 --> 00:28:01,320 Speaker 1: all of whom look essentially indistinguish. Now, I just want 412 00:28:01,320 --> 00:28:05,320 Speaker 1: to note beyond issues of following movie plots, prosopagnosia can 413 00:28:05,359 --> 00:28:10,080 Speaker 1: have pretty significant social effects like anxiety and embarrassment and 414 00:28:10,119 --> 00:28:15,040 Speaker 1: frustration because of and inability to recognize other people, And 415 00:28:15,160 --> 00:28:19,120 Speaker 1: for children with face blindness, social development can be tough 416 00:28:19,160 --> 00:28:22,479 Speaker 1: because you need to recognize your peers and your teacher, 417 00:28:23,040 --> 00:28:27,480 Speaker 1: and occasionally kids will get misdiagnosed with other learning or 418 00:28:27,480 --> 00:28:31,480 Speaker 1: behavioral disorders, which complicates things for them and an adults, 419 00:28:31,560 --> 00:28:36,520 Speaker 1: face blindness can lead to social isolation or in difficulties 420 00:28:36,560 --> 00:28:41,880 Speaker 1: in forming and maintaining relationships. So super recognizers and people 421 00:28:41,880 --> 00:28:48,040 Speaker 1: with prosopagnosia they represent extreme ends of the face recognition spectrum. 422 00:28:48,080 --> 00:28:51,840 Speaker 1: Most people fall somewhere in between, but with varying degrees 423 00:28:51,960 --> 00:28:57,040 Speaker 1: of ability and understanding the spectrum in face recognition. This 424 00:28:57,120 --> 00:29:00,880 Speaker 1: is how we can get insight into the differences between 425 00:29:01,080 --> 00:29:15,120 Speaker 1: people's realities. Okay, now I just want to say lest 426 00:29:15,200 --> 00:29:18,640 Speaker 1: you think that these issues about recognizing faces just show 427 00:29:18,720 --> 00:29:21,960 Speaker 1: up in podcasts, In fact, they show up in courtrooms 428 00:29:22,000 --> 00:29:25,680 Speaker 1: all the time, most notably in the very stormy world 429 00:29:26,040 --> 00:29:30,400 Speaker 1: of eyewitness identification. If you check out my episode nineteen, 430 00:29:30,880 --> 00:29:34,120 Speaker 1: I talked all about the difficulties of eyewitness identification, which 431 00:29:34,160 --> 00:29:38,920 Speaker 1: is generally very difficult because of fundamental flaws in memory. 432 00:29:39,240 --> 00:29:41,360 Speaker 1: But there are other challenges to this as well, and 433 00:29:41,400 --> 00:29:44,640 Speaker 1: the main one of interest here is when people are 434 00:29:44,800 --> 00:29:48,600 Speaker 1: asked to reproduce faces, as in who did you see 435 00:29:48,680 --> 00:29:51,480 Speaker 1: at the scene of the crime. So what police did 436 00:29:51,480 --> 00:29:55,280 Speaker 1: for centuries was to have a trained artist who sketches 437 00:29:55,360 --> 00:29:58,600 Speaker 1: what you describe. But of course it's really hard to 438 00:29:58,680 --> 00:30:01,840 Speaker 1: describe a face, and so by the late nineteen fifties, 439 00:30:01,920 --> 00:30:05,600 Speaker 1: the Los Angeles Police Department introduced a new way of 440 00:30:05,720 --> 00:30:10,120 Speaker 1: doing this where you line up individual strips for the eyes, 441 00:30:10,280 --> 00:30:13,440 Speaker 1: the eyebrows, the nose, the mouths, and so on. So 442 00:30:13,520 --> 00:30:16,560 Speaker 1: instead of me asking you, hey, can you describe that 443 00:30:16,640 --> 00:30:20,720 Speaker 1: guy's face? Instead, you now get a bunch of possible 444 00:30:20,760 --> 00:30:24,560 Speaker 1: eyes and possible noses and mouths and ears and chins, 445 00:30:24,840 --> 00:30:28,040 Speaker 1: and you try to reconstruct it that way. And this 446 00:30:28,200 --> 00:30:31,200 Speaker 1: ended up spreading from Los Angeles to Scotland Yard by 447 00:30:31,200 --> 00:30:34,760 Speaker 1: the nineteen sixties, and they had some successes, and eventually 448 00:30:34,760 --> 00:30:38,480 Speaker 1: this became a computerized system where people can piece this 449 00:30:38,560 --> 00:30:41,320 Speaker 1: together on a computer, and so this was considered a 450 00:30:41,400 --> 00:30:44,760 Speaker 1: real success about how you can use things beyond just 451 00:30:45,280 --> 00:30:52,600 Speaker 1: unreliable verbal descriptions to identify perpetrators. But when researchers subjected 452 00:30:52,640 --> 00:30:54,920 Speaker 1: this to more careful study, it turned out that this 453 00:30:54,960 --> 00:30:59,040 Speaker 1: approach of piecing faces together this is pretty imperfect, and 454 00:30:59,080 --> 00:31:02,520 Speaker 1: the fundamental problem is that the composite is built from 455 00:31:02,640 --> 00:31:05,600 Speaker 1: pieces and parts, like the guy's eyes look like that, 456 00:31:05,720 --> 00:31:07,680 Speaker 1: and his mouth looks like that, his nose like that, 457 00:31:07,760 --> 00:31:10,520 Speaker 1: and so on. But that's not the way the human 458 00:31:10,600 --> 00:31:15,280 Speaker 1: visual system works. It recognizes faces based on the whole picture, 459 00:31:15,760 --> 00:31:19,880 Speaker 1: the gestalt, where the whole is perceived as more than 460 00:31:19,880 --> 00:31:23,000 Speaker 1: the sum of the parts. And so it turns out 461 00:31:23,280 --> 00:31:27,480 Speaker 1: that reconstructing a face from pieces and parts is not 462 00:31:27,680 --> 00:31:32,040 Speaker 1: really so easy. I'll also mention another face recognition issue 463 00:31:32,040 --> 00:31:34,400 Speaker 1: which shows up in courts all the time, and that's 464 00:31:34,480 --> 00:31:37,840 Speaker 1: the other race effect which I mentioned earlier. People are 465 00:31:37,880 --> 00:31:42,840 Speaker 1: more likely to misidentify individuals of other races, for example, 466 00:31:43,080 --> 00:31:46,840 Speaker 1: a Hispanic person identifying a Japanese person and so on. 467 00:31:47,160 --> 00:31:51,400 Speaker 1: This is not racism, it's just neuroscience. Your brain comes 468 00:31:51,440 --> 00:31:55,400 Speaker 1: to represent what you see around you, and so in 469 00:31:55,480 --> 00:31:58,160 Speaker 1: many ways, the legal system has to chew on the 470 00:31:58,240 --> 00:32:03,320 Speaker 1: issue of how brain recognized faces and wear brains don't 471 00:32:03,320 --> 00:32:06,920 Speaker 1: do it so well. Okay, So moving beyond courts into 472 00:32:06,960 --> 00:32:11,160 Speaker 1: the broader society, we're now entering a brave new world 473 00:32:11,400 --> 00:32:14,920 Speaker 1: of face recognition technology. There was a big moment in 474 00:32:15,040 --> 00:32:20,080 Speaker 1: twenty eighteen in China where new face recognition technology picked 475 00:32:20,160 --> 00:32:24,280 Speaker 1: out a person that the authorities wanted out of a 476 00:32:24,360 --> 00:32:28,160 Speaker 1: crowd of sixty thousand people at a concert, and apparently 477 00:32:28,320 --> 00:32:30,520 Speaker 1: when they came up and grabbed the guy, he was 478 00:32:30,720 --> 00:32:34,760 Speaker 1: infinitely surprised because he assumed that being in a giant 479 00:32:34,880 --> 00:32:37,720 Speaker 1: crowd made him safe, that no one would ever be 480 00:32:37,760 --> 00:32:40,440 Speaker 1: able to spot him, and he was of course right 481 00:32:40,520 --> 00:32:43,760 Speaker 1: that no one could, but the computer did. Now, this 482 00:32:43,960 --> 00:32:48,040 Speaker 1: was an example where face recognition technology performed at a 483 00:32:48,200 --> 00:32:52,600 Speaker 1: superhuman level, but note that the technology often messes up 484 00:32:52,680 --> 00:32:57,600 Speaker 1: and also does something that's very human paradolia. We always 485 00:32:57,640 --> 00:33:02,280 Speaker 1: see algorithms mistakenly idea identifying faces where there aren't any, 486 00:33:02,520 --> 00:33:06,400 Speaker 1: which mimics the same tendencies that we have as humans. 487 00:33:06,720 --> 00:33:09,440 Speaker 1: For example, remember the jack O lantern that I mentioned 488 00:33:09,480 --> 00:33:14,600 Speaker 1: earlier that the Facebook algorithm mistakenly identified as a human face. 489 00:33:14,920 --> 00:33:18,440 Speaker 1: Computers tend to impose patterns as much as we do. 490 00:33:18,760 --> 00:33:20,480 Speaker 1: And I just want to say that even though we're 491 00:33:20,480 --> 00:33:24,640 Speaker 1: often worried about new technologies and its face recognition abilities, 492 00:33:24,800 --> 00:33:27,480 Speaker 1: what's also going to grow out of this are increasingly 493 00:33:27,520 --> 00:33:32,840 Speaker 1: sophisticated assist of devices. You have AI driven tools that 494 00:33:32,880 --> 00:33:37,200 Speaker 1: are going to offer real time face recognition support which 495 00:33:37,240 --> 00:33:40,200 Speaker 1: is going to allow you to recognize your friend when 496 00:33:40,240 --> 00:33:42,520 Speaker 1: you see him totally out of context. But more importantly, 497 00:33:43,400 --> 00:33:46,520 Speaker 1: this has the potential to enhance the quality of life 498 00:33:47,000 --> 00:33:51,960 Speaker 1: for people with prosopagnosia. So let's wrap up. Face recognition 499 00:33:52,280 --> 00:33:57,200 Speaker 1: is a remarkably computationally intensive thing that we do, and 500 00:33:57,280 --> 00:34:01,440 Speaker 1: even though it seems effortless, we have these massive, specialized 501 00:34:01,440 --> 00:34:05,640 Speaker 1: neural networks that underpin our ability to do this. This 502 00:34:05,800 --> 00:34:10,239 Speaker 1: ability highlights the very special role that faces play in 503 00:34:10,280 --> 00:34:13,799 Speaker 1: our social lives and in our interactions, and it's so 504 00:34:13,840 --> 00:34:17,759 Speaker 1: important that we all experience things like paradolia, where we 505 00:34:18,120 --> 00:34:23,319 Speaker 1: impose an interpretation of faces on other patterns all around us, 506 00:34:23,920 --> 00:34:27,719 Speaker 1: and this underscores our brain's need to make sense of 507 00:34:27,800 --> 00:34:32,360 Speaker 1: the world, to impose order on chaos, and to connect 508 00:34:32,600 --> 00:34:35,960 Speaker 1: what we see with what we know. And finally, we 509 00:34:36,000 --> 00:34:40,000 Speaker 1: saw that skill in face recognition varies widely from person 510 00:34:40,040 --> 00:34:43,600 Speaker 1: to person. We can measure this from brain imaging in 511 00:34:43,680 --> 00:34:49,040 Speaker 1: people's performances, and we find a spectrum from super recognizers 512 00:34:49,400 --> 00:34:53,680 Speaker 1: to those who are face blind. And understanding how different 513 00:34:53,719 --> 00:34:57,120 Speaker 1: reality can be on the inside is critical for living 514 00:34:57,120 --> 00:35:00,920 Speaker 1: in a society and understanding one another as it stands now, 515 00:35:01,000 --> 00:35:04,640 Speaker 1: most people on the planet are unaware of this spectrum. 516 00:35:04,640 --> 00:35:07,920 Speaker 1: They've never heard of something like prosapagnosia, and as a result, 517 00:35:08,280 --> 00:35:11,640 Speaker 1: they don't recognize it in themselves or in a loved one. 518 00:35:11,880 --> 00:35:15,160 Speaker 1: So this is one of the things we gain from 519 00:35:15,239 --> 00:35:20,239 Speaker 1: a deeper understanding of the brain, a broader empathy that 520 00:35:20,360 --> 00:35:25,799 Speaker 1: allows societies to interact more richly. So the next time 521 00:35:26,200 --> 00:35:30,160 Speaker 1: you see a newborn baby, lock onto its mother's face 522 00:35:30,320 --> 00:35:34,279 Speaker 1: and track the face and study it, just remember that 523 00:35:34,320 --> 00:35:38,480 Speaker 1: you're not just seeing something cute, you are catching insight 524 00:35:39,000 --> 00:35:46,560 Speaker 1: into deep circuitry of the inner cosmos. Go to Eagleman 525 00:35:46,600 --> 00:35:49,760 Speaker 1: dot com slash podcast for more information and to find 526 00:35:49,840 --> 00:35:53,840 Speaker 1: further reading. Send me an email at podcasts at egleman 527 00:35:53,880 --> 00:35:57,000 Speaker 1: dot com with questions or discussion, and check out and 528 00:35:57,040 --> 00:36:00,719 Speaker 1: subscribe to Inner Cosmos on YouTube for video of each 529 00:36:00,760 --> 00:36:05,600 Speaker 1: episode and to leave comments. Until next time. I'm David Eagleman, 530 00:36:05,800 --> 00:36:07,759 Speaker 1: and this is Inner Cosmos.