WEBVTT - Ep19 "How far can you trust your memory?" 0:00:04.880 --> 0:00:07.400 When you witness an event, when you see it with 0:00:07.480 --> 0:00:10.600 your own eyes, it certainly feels like what you saw 0:00:10.960 --> 0:00:15.280 can't be questioned. But how good is your memory? Really? 0:00:15.400 --> 0:00:20.400 Can you misremember details? Is your memory like a video 0:00:21.000 --> 0:00:25.320 or isn't it? Is your ability to remember changed by 0:00:25.320 --> 0:00:27.760 how many things are going on in the scene, or 0:00:27.840 --> 0:00:31.800 whether there's a gun pointed at you. Can something told 0:00:31.920 --> 0:00:36.040 to you after the event change your memory of the event. 0:00:40.280 --> 0:00:43.400 Welcome to Inner Cosmos with me David Eagleman. I'm a 0:00:43.479 --> 0:00:47.480 neuroscientist and an author at Stanford and in these episodes 0:00:47.920 --> 0:00:51.960 we sail deeply into our three pound universe to understand 0:00:52.280 --> 0:01:06.120 the intersection between brains and our lives. Today's episode, we're 0:01:06.160 --> 0:01:09.959 going to be talking about eyewitness testimony. I think it's 0:01:10.000 --> 0:01:12.959 fair to say that most of us when we see 0:01:13.000 --> 0:01:17.800 something happen, we know what we've just seen. This occurred, 0:01:18.080 --> 0:01:19.720 this is what the person looked like, and so on. 0:01:20.360 --> 0:01:23.520 And in fact, when jurors get interviewed on this point, 0:01:23.560 --> 0:01:27.520 they'll often say something like, yes, I understand there are 0:01:27.560 --> 0:01:32.400 problems with memory and eyewitness testimony, but my memory is 0:01:32.440 --> 0:01:35.160 like a video recorder. But the question we're going to 0:01:35.240 --> 0:01:39.560 ask today is how good is our memory really and 0:01:39.680 --> 0:01:43.520 what does that mean for courts of law? So strap 0:01:43.560 --> 0:01:46.000 in because we're going to see some amazing events and 0:01:46.080 --> 0:01:50.120 court cases that may change your opinion on your own 0:01:50.160 --> 0:01:58.640 memory and about what you take to be true. One 0:01:58.680 --> 0:02:01.720 of the courses I teach it Anford is the Brain and 0:02:01.800 --> 0:02:05.640 the Law. It's where neuroscience intersects with the legal system. 0:02:05.960 --> 0:02:08.320 And a few months ago I was in the middle 0:02:08.360 --> 0:02:13.359 of lecturing to my seventies students when something very wild happened. 0:02:13.440 --> 0:02:18.200 I was talking to the auditorium and the back door opens, 0:02:20.360 --> 0:02:23.200 and a middle aged woman comes in at the back 0:02:23.240 --> 0:02:26.639 of the classroom, and she walks halfway down the aisle 0:02:26.720 --> 0:02:30.800 toward me and then just interrupts me, as though I 0:02:30.880 --> 0:02:34.200 weren't talking at all. She says, are you David Eagleman? 0:02:34.400 --> 0:02:36.480 And I said yes, but I'm in the middle of 0:02:36.520 --> 0:02:40.040 teaching a class. Who are you? And she starts shouting 0:02:40.440 --> 0:02:43.840 how she's emailed me over and over and hasn't gotten 0:02:43.840 --> 0:02:47.920 any reply. So I said, I'm sorry, I have an 0:02:48.000 --> 0:02:51.040 overwhelmed inbox and I can't get to all my emails, 0:02:51.080 --> 0:02:53.480 but I am teaching a class and I'll have to 0:02:53.520 --> 0:02:57.280 talk with you later. But she's clearly very agitated, and 0:02:57.360 --> 0:03:00.760 she doesn't leave, and she keeps shouting at me and 0:03:00.880 --> 0:03:05.080 takes a step closer to me down the aisle and 0:03:05.240 --> 0:03:08.120 asks if I'm ever planning to answer her, or whether 0:03:08.120 --> 0:03:11.800 I'm going to continue to ignore her. And I say, ma'am, 0:03:11.840 --> 0:03:13.720 you're going to have to leave and I'll talk with 0:03:13.760 --> 0:03:18.640 you afterwards. And finally, after what seemed like forever, although 0:03:18.680 --> 0:03:22.200 it was presumably less than a minute, she says, I'm 0:03:22.200 --> 0:03:25.120 going to wait for you out here, and she goes 0:03:25.320 --> 0:03:31.480 stomping out the back door. So I tried to maintain normalcy, 0:03:31.680 --> 0:03:35.200 and I continued the lecture for about thirty more minutes. 0:03:35.720 --> 0:03:37.880 And this class is a long one. I give three 0:03:37.960 --> 0:03:40.880 hours of lecture, so normally at this point I give 0:03:40.920 --> 0:03:44.360 my students a break at the midpoint, where we go 0:03:44.400 --> 0:03:47.480 outside and we stretch for a few minutes. But I said 0:03:47.520 --> 0:03:50.720 to the class, look, it's our break time. But I'm 0:03:50.760 --> 0:03:53.040 not sure that it's going to make sense to go 0:03:53.120 --> 0:03:56.320 out there. I'm going to call campus security just to 0:03:56.440 --> 0:03:59.320 let them know about that woman, and I want to 0:03:59.360 --> 0:04:03.400 give them a description, but I couldn't really tell her 0:04:03.440 --> 0:04:06.200 height and wait very well from up here at the podium, 0:04:07.080 --> 0:04:08.960 and I guess I was so caught up in the 0:04:09.040 --> 0:04:12.280 surprise of the whole thing and thinking about the ways 0:04:12.360 --> 0:04:17.080 this scenario could evolve badly that I didn't really encode 0:04:17.120 --> 0:04:19.240 all the details that I wanted to, like what she 0:04:19.320 --> 0:04:22.279 was wearing, or her hair color, or the details of 0:04:22.279 --> 0:04:25.240 her face as well as I wish that I had 0:04:25.400 --> 0:04:29.080 encoded these. So I told the class, look, I remember 0:04:29.120 --> 0:04:32.479 the random fact that she had sunglasses perched on top 0:04:32.520 --> 0:04:35.279 of her head, but I would really like your help 0:04:35.400 --> 0:04:39.479 in getting a rich description. So I got everyone to 0:04:39.480 --> 0:04:41.880 take out a piece of notebook paper, and I told 0:04:41.880 --> 0:04:45.120 them the best way to accurately identify someone is to 0:04:45.160 --> 0:04:50.279 have a collection of witnesses do this independently. So I said, look, 0:04:50.279 --> 0:04:52.880 if you could draw a picture of what she looked 0:04:52.960 --> 0:04:55.320 like and also write down her height and her weight, 0:04:55.920 --> 0:04:59.680 then I'm going to compile these and give Stanford Security 0:05:00.120 --> 0:05:03.680 an excellent description. So one thing I said is don't 0:05:03.720 --> 0:05:06.120 look at each other's drawing. Just drawn on your own 0:05:06.279 --> 0:05:08.680 and then hand it up to me. So everyone drew 0:05:08.720 --> 0:05:12.200 their best version of how they remembered her, and some 0:05:12.240 --> 0:05:14.279 people relied more on the drawing part and some on 0:05:14.320 --> 0:05:18.960 the verbal descriptions. I collected these all up, and I said, great, 0:05:18.960 --> 0:05:22.200 I'm going to have a very complete average version of 0:05:22.200 --> 0:05:25.359 her that I can communicate to security. So I started 0:05:25.400 --> 0:05:28.760 flipping through these and reading them out loud to the class, 0:05:29.440 --> 0:05:34.400 and we were all shocked because one person described her 0:05:34.480 --> 0:05:37.880 as five foot four and wearing a blue shirt and 0:05:37.920 --> 0:05:41.839 the next described her as five foot eight and wearing 0:05:41.880 --> 0:05:46.279 a white shirt. And the age estimates were between forty 0:05:46.320 --> 0:05:51.000 and sixty, and the hair color was everything from brown 0:05:51.040 --> 0:05:53.920 hair to gray hair to light hair and so on. 0:05:54.480 --> 0:05:59.200 And the body weight ranged over about forty pounds. So 0:05:59.400 --> 0:06:04.000 the whole collection provided me with no benefit at all 0:06:04.200 --> 0:06:08.599 in calling security. So I didn't call security, but not 0:06:08.680 --> 0:06:11.159 for the reason you might think. The reason I didn't 0:06:11.160 --> 0:06:14.560 call is because the woman was a professional actor that 0:06:14.640 --> 0:06:17.720 I had hired. And the whole point of the exercise 0:06:17.839 --> 0:06:22.960 was to demonstrate firsthand to my students that vision is 0:06:23.120 --> 0:06:27.160 not like a camera and memory is not like a recorder. 0:06:28.320 --> 0:06:30.839 And from there I segued into the real topic of 0:06:30.880 --> 0:06:35.000 the lecture, which was on the problems and challenges of 0:06:35.160 --> 0:06:40.120 eyewitness testimony. So let's begin with a woman named Jennifer Thompson. 0:06:40.640 --> 0:06:43.960 She was twenty two and living in North Carolina, and 0:06:44.040 --> 0:06:47.600 a man broke into her apartment and raped her. It 0:06:47.720 --> 0:06:51.360 was a long ordeal, and Jennifer really paid attention to 0:06:51.400 --> 0:06:54.480 his face so that she would be able to identify 0:06:54.560 --> 0:06:58.000 him later. When this whole terrible event was over and 0:06:58.040 --> 0:07:01.960 he ran off, she drove herself to the hospital and 0:07:02.080 --> 0:07:04.960 had a rape kit taken, and then she went straight 0:07:05.000 --> 0:07:07.279 to the police, and with their help she was able 0:07:07.320 --> 0:07:10.440 to create a composite image of what he looked like. 0:07:11.240 --> 0:07:14.040 So the police went out and they gathered suspects, and 0:07:14.080 --> 0:07:17.440 they put together a lineup of seven men, and Jennifer 0:07:17.480 --> 0:07:20.960 concluded that one man in that lineup, a man named 0:07:21.040 --> 0:07:24.400 Ronald Cotton, was the man who had raped her. So 0:07:24.520 --> 0:07:26.760 Cotton went to jail for the crime, but the whole 0:07:26.800 --> 0:07:31.040 time he was there he maintained his innocence, and many 0:07:31.120 --> 0:07:34.400 years into his jail sense he saw the oj Simpson 0:07:34.440 --> 0:07:39.520 trial on the prison television and he heard about DNA evidence, 0:07:39.560 --> 0:07:42.080 which was something he had never heard of before, and 0:07:42.120 --> 0:07:44.920 so he contacted his lawyer and he said, what is 0:07:44.960 --> 0:07:48.320 this DNA evidence? Is this something we should look into? 0:07:48.880 --> 0:07:51.000 And his lawyer was able to get a hold of 0:07:51.080 --> 0:07:54.520 the rape kit from almost eleven years earlier, and they 0:07:54.520 --> 0:07:57.760 were able to extract enough DNA from that, and it 0:07:57.840 --> 0:08:02.080 turned out that Cotton was in fact innocent. He was 0:08:02.160 --> 0:08:04.680 not the man who had done it. The man who 0:08:04.680 --> 0:08:08.200 had raped Jennifer Thompson was a man named Bobby Poole, 0:08:08.680 --> 0:08:13.040 and he was then apprehended and confessed to the rape. 0:08:13.120 --> 0:08:17.320 And so Ronald Cotton was released after having spent eleven 0:08:17.480 --> 0:08:20.760 years of his life in prison for a crime he 0:08:20.840 --> 0:08:25.480 did not commit. So Jennifer was shattered by this because 0:08:25.520 --> 0:08:29.240 she realized now that, based on her eyewitness testimony and 0:08:29.280 --> 0:08:33.240 her certainty, she had sent an innocent man to jail 0:08:33.320 --> 0:08:36.000 for eleven years. And she said that one of the 0:08:36.040 --> 0:08:38.480 most important things to her as a rape victim was 0:08:38.520 --> 0:08:41.560 to never feel any guilt over what had happened to her. 0:08:41.600 --> 0:08:44.240 It wasn't her fault that she had been raped, But 0:08:44.480 --> 0:08:47.400 all of a sudden, now she was crushed with guilt 0:08:47.800 --> 0:08:51.920 because she had ruined an innocent man's life by erroneously 0:08:52.440 --> 0:08:57.199 identifying him in a lineup. So, about three years after 0:08:57.320 --> 0:09:01.319 Ronald Cotton got released, she decided she was going to 0:09:01.360 --> 0:09:04.360 reach out to him, make contact with him, and of 0:09:04.400 --> 0:09:07.080 course she was terrified about doing this. But she contacted 0:09:07.160 --> 0:09:10.320 him anyway, and they met at a church and he 0:09:10.440 --> 0:09:13.880 said that he had already forgiven her, and they talked 0:09:13.920 --> 0:09:18.720 at length, and they both cried, and eventually they became friends, 0:09:19.320 --> 0:09:22.080 and they ended up writing an excellent book together. It's 0:09:22.120 --> 0:09:26.080 called Picking Cotton. And one of the things she does 0:09:26.120 --> 0:09:28.880 in the book is wrestle with the question that was 0:09:29.400 --> 0:09:32.840 most deeply burned into her mind. How could she have 0:09:32.960 --> 0:09:37.520 picked the wrong person and have felt so certain about it. 0:09:37.960 --> 0:09:42.760 How can one feel so sure and be wrong. So 0:09:42.840 --> 0:09:46.440 they travel around and give talks together about the fallibility 0:09:46.480 --> 0:09:50.200 of eyewitness testimony, about how you can believe one hundred 0:09:50.240 --> 0:09:54.600 percent that's the face, but it doesn't actually necessitate that 0:09:54.640 --> 0:09:57.600 you are correct. I met the two of them at 0:09:57.640 --> 0:10:00.360 a conference when they were talking about the book. They're 0:10:00.400 --> 0:10:03.760 close friends now, and they travel around together so much 0:10:03.840 --> 0:10:07.080 that often people mistake them for a couple, and they say, oh, 0:10:07.120 --> 0:10:09.840 how did you two meet, And they say, it's a 0:10:09.880 --> 0:10:14.280 long story, okay. But the heart of the problem is 0:10:14.320 --> 0:10:18.320 that memory is always a reconstruction. Memory is not like 0:10:18.360 --> 0:10:20.960 a videotape. It's not as though your memory is something 0:10:21.000 --> 0:10:24.960 where your visual system is taking in information and retaining 0:10:25.080 --> 0:10:27.960 zeros and ones the way a cell phone video would. 0:10:28.280 --> 0:10:30.600 All you ever see is what you believe you are 0:10:30.640 --> 0:10:33.040 seeing out there. So if I were to ask you 0:10:33.120 --> 0:10:36.079 the details of what's in front of you right now, 0:10:36.800 --> 0:10:39.640 you'd be able to answer it once you pay attention 0:10:39.760 --> 0:10:43.000 to things. But you weren't aware of that even though 0:10:43.000 --> 0:10:45.320 it's been sitting on your retina this whole time. You're 0:10:45.320 --> 0:10:49.240 not actually seeing the world like a camera. Instead, what 0:10:49.320 --> 0:10:52.680 you have is a rough internal model of what's going on. 0:10:53.360 --> 0:10:56.400 And when you need to take in more information, then 0:10:56.440 --> 0:10:59.160 you do. You go out and point your eyes and 0:10:59.240 --> 0:11:01.920 pull in more in. So when I ask you about 0:11:01.920 --> 0:11:05.360 the scene in front of you, you need to employ 0:11:05.520 --> 0:11:08.760 your attentional mechanisms to go and crawl the scene out 0:11:08.800 --> 0:11:12.280 there and try to identify what's sitting there. So today's 0:11:12.360 --> 0:11:17.800 question is how reliable is eyewitness testimony? In the courtroom, 0:11:18.120 --> 0:11:20.960 for my students, I was able to demonstrate to them 0:11:21.440 --> 0:11:26.520 the massive variety of their drawings tall, short, heavy, light, 0:11:27.000 --> 0:11:31.360 curly hair, straight hair, glasses, no glasses, and none of 0:11:31.360 --> 0:11:36.320 their drawings looked particularly like the actor at all. But 0:11:36.440 --> 0:11:39.160 forget a classroom test, how does this play out in 0:11:39.200 --> 0:11:42.679 the real world? How often does I witness testimony end 0:11:42.760 --> 0:11:46.160 up convicting the wrong person, as happened with Jennifer and 0:11:46.240 --> 0:11:48.840 Ronald Well. One thing you could do is look at 0:11:48.880 --> 0:11:52.760 people who are found to be innocent years after they 0:11:52.760 --> 0:11:55.400 were convicted for a crime, and then figure out what 0:11:55.559 --> 0:11:59.040 percentage of them were convicted in whole or in part 0:11:59.160 --> 0:12:02.400 based on iwold testimony. So you may have heard of 0:12:02.440 --> 0:12:05.559 the Innocence Project. It is for people who are serving 0:12:05.600 --> 0:12:09.920 time in prison and are maintaining their innocence. The lawyers 0:12:10.000 --> 0:12:13.600 and forensic scientists at the Innocence Project go back and 0:12:13.679 --> 0:12:17.600 open these cold cases and work to reassess the case 0:12:17.640 --> 0:12:21.960 with DNA evidence. So what the Innocence Project found out 0:12:21.960 --> 0:12:25.760 of the two hundred and forty people that they have exonerated, 0:12:26.520 --> 0:12:28.880 in other words, who have been found innocent of the 0:12:28.880 --> 0:12:32.720 crime they were convicted for, is that over sixty percent 0:12:32.800 --> 0:12:35.200 of them went to jail in whole or in part 0:12:35.600 --> 0:12:39.960 because of eyewitness testimony where someone said, I know that's 0:12:40.040 --> 0:12:42.520 the guy. I saw him with my own eyes. There's 0:12:42.600 --> 0:12:45.560 no doubt in my mind. Now this leads to a 0:12:45.679 --> 0:12:52.240 question does eyewitness testimony matter? How swaying is it to jurors? Well, 0:12:52.280 --> 0:12:57.120 the answer is it's enormously swaying. The US Supreme Court 0:12:57.200 --> 0:13:01.720 Justice William Brennan pointed out that vote all the evidence 0:13:01.800 --> 0:13:05.319 points rather strikingly to the conclusion that there is almost 0:13:05.640 --> 0:13:10.240 nothing more convincing than a live human being who takes 0:13:10.320 --> 0:13:13.400 the stand, points a finger at the defendants and says, 0:13:14.000 --> 0:13:34.079 that's the one unquote. So imagine you have a prosecutor 0:13:34.200 --> 0:13:39.000 showing circumstantial evidence, or a scientist comes up and says, look, 0:13:39.040 --> 0:13:42.880 here's my neuroimaging evidence, or you have a lie detector 0:13:42.920 --> 0:13:46.360 expert or whatever, and then you get an eyewitness and 0:13:46.400 --> 0:13:49.080 she takes the stand and she has tears in her eyes. 0:13:49.160 --> 0:13:52.360 She's genuine, she says, I know what I saw. He 0:13:52.800 --> 0:13:58.680 is the one that has enormous way on jurors. Justice 0:13:58.720 --> 0:14:01.280 Brennan goes on to point out that eyewent his testimony 0:14:01.360 --> 0:14:05.080 is the type of evidence that quote, juries seem most 0:14:05.200 --> 0:14:09.800 receptive to and not inclined to discredit end quote. In 0:14:09.840 --> 0:14:12.280 other words, a scientist might say something on the stand 0:14:12.320 --> 0:14:15.720 and the jury thinks, yeah, maybe, but there's another interpretation. 0:14:16.400 --> 0:14:19.960 But when it comes to an eyewitness, we tend to say, Okay, 0:14:20.120 --> 0:14:23.520 that's the gospel. Finally, Brennan points out that the court 0:14:23.560 --> 0:14:27.760 has long recognized, at least since the nineteen sixties, the 0:14:27.920 --> 0:14:34.600 quote inherently suspect qualities of eyewitness identification evidence unquote, and 0:14:34.640 --> 0:14:38.160 that the court finds that kind of evidence quote notoriously 0:14:38.560 --> 0:14:43.120 unreliable end quote. So that's Brennan, having spent a life 0:14:43.160 --> 0:14:45.960 on the bench, describing what he knows to be true. 0:14:46.760 --> 0:14:49.600 So why is it so swaying? Well, maybe we can 0:14:49.640 --> 0:14:53.880 find some evidence from the laboratory. So my colleague, Elizabeth 0:14:53.960 --> 0:14:57.359 loftis that you see, Irvine is probably the most famous 0:14:57.400 --> 0:15:02.560 memory eyewitness researcher. Here's one example of her many many studies. 0:15:03.000 --> 0:15:06.840 You tell mock jurors the details of a case. They 0:15:06.880 --> 0:15:09.640 get to hear the attorneys on both sides, the prosecution 0:15:09.800 --> 0:15:13.320 and the defense, give their arguments. Some of the jurors 0:15:13.560 --> 0:15:17.840 only hear circumstantial evidence. There's no eyewitness testimony in that case. 0:15:18.440 --> 0:15:21.960 And given the facts that they've heard, eighteen percent of 0:15:22.000 --> 0:15:25.600 the mock jurors conclude that the guy is guilty. But 0:15:25.720 --> 0:15:30.280 the other group, here's eyewitness testimony, everything else being the same, 0:15:30.760 --> 0:15:34.600 and now seventy two percent of the mock jurors find 0:15:34.680 --> 0:15:37.480 the guy guilty. So it goes from eighteen percent of 0:15:37.520 --> 0:15:40.560 people finding him guilty to seventy two percent just based 0:15:40.560 --> 0:15:43.320 on the eyewitness testimony. That means that even though we 0:15:43.520 --> 0:15:48.640 know eyewitness testimony is inherently unreliable, just look at the 0:15:48.640 --> 0:15:52.640 effect of saying, we have someone who saw that person. 0:15:53.000 --> 0:15:54.760 Maybe it was a little bit far away, maybe it 0:15:54.800 --> 0:15:57.400 was sort of dark out, but I'm pretty sure that 0:15:57.600 --> 0:16:00.120 was the guy, and bam, it goes up to seventy two. 0:16:00.480 --> 0:16:04.200 It's a huge difference, And for something that's so unreliable, 0:16:04.600 --> 0:16:07.640 it's a little hard to justify that kind of pull. 0:16:08.880 --> 0:16:12.640 So we know that eyewitness testimony is swaying, But why 0:16:12.760 --> 0:16:16.360 is it so bad? Why can't we look at somebody 0:16:16.600 --> 0:16:19.760 like the actor who broke into my classroom and just 0:16:19.880 --> 0:16:25.240 describe accurately what she looked like. Well, for starters, it's 0:16:25.400 --> 0:16:29.160 really really hard to remember what somebody looks like. Think 0:16:29.160 --> 0:16:32.080 about the last time you went to the coffee shop, 0:16:32.280 --> 0:16:35.720 what precisely did the person at the cash register look like? 0:16:36.480 --> 0:16:39.360 Or if you had to describe your neighbor down the 0:16:39.440 --> 0:16:42.200 street who you don't see that often to the police, 0:16:42.880 --> 0:16:45.720 how well could you do so? If you really had 0:16:45.720 --> 0:16:47.800 to pull out a piece of paper and draw the 0:16:47.840 --> 0:16:51.160 face or describe it verbally. So police have worked for 0:16:51.200 --> 0:16:56.400 decades to try to make this easier. The original approach 0:16:56.560 --> 0:16:59.880 was to have a trained artist who sketches while you 0:17:00.520 --> 0:17:03.200 and then you can work back and forth together. But 0:17:03.240 --> 0:17:06.159 eventually the police put together systems where you could just 0:17:06.600 --> 0:17:09.919 look at the individual features like the eyes and the 0:17:09.960 --> 0:17:12.280 eyebrows and the nose and the mouth, and you flip 0:17:12.320 --> 0:17:14.800 them until you get the face that you want out 0:17:14.800 --> 0:17:17.960 of that. So instead of me asking you to take 0:17:17.960 --> 0:17:21.119 out a piece of notebook paper and draw seth Rogen, 0:17:21.560 --> 0:17:23.800 instead I say, look, here's a bunch of eyes and 0:17:23.880 --> 0:17:27.320 noses and mouths, and I want you to flip these 0:17:27.359 --> 0:17:30.080 around until you get a good picture of seth Rogen 0:17:30.520 --> 0:17:33.800 or the person that you saw or the perpetrator. So 0:17:33.840 --> 0:17:36.120 the first system for this was introduced in the late 0:17:36.240 --> 0:17:40.479 nineteen fifties by the Los Angeles Police Department. It was 0:17:40.880 --> 0:17:44.280 these features on transparent sheets and you rotate them around 0:17:44.320 --> 0:17:49.679 and you reconstruct the face. This was called Identicit, and 0:17:49.880 --> 0:17:53.080 Identicate was used by Scotland Yard in a case in 0:17:53.160 --> 0:17:57.600 nineteen sixty one, a woman was stabbed to death in 0:17:57.640 --> 0:18:00.639 an antique shop where she work, and the police interviewed 0:18:00.640 --> 0:18:04.119 people and learned that someone suspicious had been in the 0:18:04.160 --> 0:18:07.680 shop a few days earlier, and so Scotland Yard turned 0:18:07.920 --> 0:18:11.879 to this invention from America and had their witnesses compile 0:18:12.040 --> 0:18:16.720 the suspect's face using identicit, And some days later a 0:18:16.760 --> 0:18:20.120 policeman saw a man named Edwin Bush on the street 0:18:20.520 --> 0:18:24.280 and recognized him from the identicate pictures in the newspaper, 0:18:25.680 --> 0:18:27.840 and in fact, when they arrested him and took him 0:18:27.840 --> 0:18:30.000 to the station, they found he had a copy of 0:18:30.040 --> 0:18:33.960 the newspaper drawing in his own pocket, and his shoes 0:18:34.080 --> 0:18:36.399 matched the size of the shoes in the crime scene, 0:18:36.400 --> 0:18:39.119 and so on, and he ended up confessing to the murder. 0:18:39.520 --> 0:18:42.040 So this was considered a real success in how you 0:18:42.080 --> 0:18:47.280 could crank up the reliability of identifying people and identic 0:18:47.320 --> 0:18:51.600 It was eventually replaced by photo fit in the nineteen seventies, 0:18:51.760 --> 0:18:56.360 where instead of using drawings, you use collages of photos 0:18:56.400 --> 0:18:58.720 of eyes and ears and noses and mouth and hairline 0:18:58.720 --> 0:19:02.160 and so on, and the photos give a better image 0:19:02.160 --> 0:19:05.679 of the suspect's face rather than the line drawings. And 0:19:05.840 --> 0:19:09.280 with due course all this evolved into software in which 0:19:09.280 --> 0:19:12.239 you take a three dimensional model of the head and 0:19:12.280 --> 0:19:13.879 you do the same sort of thing where you can 0:19:13.960 --> 0:19:15.960 change the eyes and nose and mouth and eyebrows and 0:19:16.000 --> 0:19:18.560 so on. And there were many stories that came out 0:19:18.560 --> 0:19:22.720 of all these systems and that increased confidence in the approach. 0:19:23.840 --> 0:19:27.040 But it turns out that despite success stories with this, 0:19:27.480 --> 0:19:29.520 when you really start looking into it, it turns out 0:19:29.560 --> 0:19:33.720 these face recognition approaches are generally not so good in 0:19:33.800 --> 0:19:38.320 terms of being able to identify people. Some years ago, 0:19:38.440 --> 0:19:42.399 an MIT researcher named Pawan Sinha got a hold of 0:19:42.440 --> 0:19:45.919 the standard facial software that police forces use, and he 0:19:46.040 --> 0:19:49.720 requested an expert with several years of experience with the 0:19:49.760 --> 0:19:54.320 system to put together reconstructions directly from a picture and 0:19:54.480 --> 0:19:58.320 without time constraints. That meant the operator didn't have to 0:19:58.320 --> 0:20:01.200 rely on verbal descriptions. He could just look at the 0:20:01.200 --> 0:20:05.439 photos and put this together. So each year in my class, 0:20:05.440 --> 0:20:08.560 I show the students these four photographs and I offer 0:20:08.640 --> 0:20:12.159 ten dollars to whoever can identify the pictures, and no 0:20:12.240 --> 0:20:15.679 one ever can. There are lots of wrong guesses, but 0:20:15.760 --> 0:20:18.679 so far no one has gotten them. Right. Now, the 0:20:18.760 --> 0:20:24.280 pictures are of Bill Cosby, Carl Sagan, Ronald Reagan, Michael Jordan. 0:20:24.720 --> 0:20:27.360 Now I get it. These are faces that the younger 0:20:27.440 --> 0:20:31.679 generation can't necessarily identify anyway. But at the peak of 0:20:31.720 --> 0:20:35.680 these guys popularity in the nineteen eighties, people couldn't identify 0:20:35.880 --> 0:20:40.800 these pictures. Then the reconstructions just weren't close enough to 0:20:40.840 --> 0:20:44.560 the actual face to make the match. And in case 0:20:44.600 --> 0:20:48.360 you think this might just be a laboratory artifact, there 0:20:48.359 --> 0:20:50.880 are plenty of real life cases where you look at 0:20:50.880 --> 0:20:54.920 the identicate composite and then you look at the captured 0:20:55.040 --> 0:20:58.520 person's photograph and you'll know right away that you would 0:20:58.520 --> 0:21:03.440 have never identified that person from that drawing. Why. Well, 0:21:03.600 --> 0:21:06.680 part of the problem is that the feature based composite 0:21:06.680 --> 0:21:10.720 image is built from the pieces and parts. You say, 0:21:10.840 --> 0:21:12.760 I think the guy's eyes looked sort of like this, 0:21:12.880 --> 0:21:14.480 and his mouth looked sort of like that, and his 0:21:14.560 --> 0:21:16.840 nose and his eyebrows and so on. But that's not 0:21:16.920 --> 0:21:21.040 the way the human visual system works. It recognizes faces 0:21:21.080 --> 0:21:25.160 based on the whole picture. What's called the gestalt, which 0:21:25.200 --> 0:21:28.359 is the German word that signifies when a whole is 0:21:28.480 --> 0:21:33.000 perceived as more than the sum of its parts. But 0:21:33.160 --> 0:21:37.480 that is only one problem. Let's imagine that you could 0:21:37.520 --> 0:21:41.440 somehow reconstruct a pretty good likeness there's also the problem 0:21:41.600 --> 0:21:44.960 of similarity, which is to say, a lot of people 0:21:45.119 --> 0:21:47.879 just look alike. So there was an amazing case some 0:21:47.960 --> 0:21:51.280 years ago in which a man named Lawrence Berson, who 0:21:51.320 --> 0:21:55.080 had curly hair and big square glasses, was convicted of 0:21:55.160 --> 0:21:58.720 several rape cases and went to jail. And then another guy, 0:21:58.840 --> 0:22:02.960 George Morales, with curly hair and big square glasses, was 0:22:03.000 --> 0:22:07.119 convicted of several robberies and went to jail, and both 0:22:07.160 --> 0:22:10.399 protested their innocence, and it turns out they were telling 0:22:10.440 --> 0:22:13.200 the truth. The guy who did all of the crimes 0:22:13.640 --> 0:22:17.400 was a guy named Richard Carbone, who had curly hair 0:22:17.520 --> 0:22:22.760 and big square glasses. Carbone had been described by several eyewitnesses, 0:22:23.200 --> 0:22:26.600 and these two other poor guys served jail time because 0:22:27.000 --> 0:22:29.879 they were picked out of the police lineups. And putting 0:22:29.920 --> 0:22:32.399 a photo of these guys on eagleman dot com slash 0:22:32.440 --> 0:22:36.400 podcast because it's so striking how different people can look 0:22:36.440 --> 0:22:38.800 similar to one another. So I've told you so far 0:22:38.840 --> 0:22:42.520 that the brain in codes faces holistically, not in bits 0:22:42.520 --> 0:22:45.000 and parts, and also that a lot of people look 0:22:45.119 --> 0:22:49.359 roughly similar. But neither of those problems even competes with 0:22:49.480 --> 0:22:54.320 the biggest problem of eyewitness testimony, and that is our 0:22:54.600 --> 0:22:58.520 memories are terrible. We all like to think that we 0:22:58.560 --> 0:23:01.760 can remember a face in describe it, but as my 0:23:01.880 --> 0:23:04.440 students saw from trying to draw the woman who broke 0:23:04.480 --> 0:23:08.360 into my classroom and spoke angrily with me for a minute, 0:23:08.840 --> 0:23:12.479 it's really hard to remember the details of someone's face 0:23:12.560 --> 0:23:15.600 and then try to reconstruct that. And the past few 0:23:15.680 --> 0:23:20.280 decades have seen really great studies which undermine our confidence 0:23:20.400 --> 0:23:23.120 that this should be easy to do. So the first 0:23:23.160 --> 0:23:26.840 thing to appreciate is that there are two phases of memory. 0:23:26.880 --> 0:23:30.360 The first is encoding, So when the woman burst into 0:23:30.400 --> 0:23:33.359 my classroom, the students were encoding what they were seeing. 0:23:33.440 --> 0:23:36.639 Their brains were writing it down, and the second phase 0:23:36.680 --> 0:23:41.159 is retrieval, or pulling the memory back up later. In 0:23:41.240 --> 0:23:44.280 the case of my classroom, I waited about thirty minutes 0:23:44.359 --> 0:23:47.720 before asking them to draw the face, and it turns 0:23:47.760 --> 0:23:51.320 out the forgetting curve is quite steep. By thirty minutes in, 0:23:51.400 --> 0:23:54.800 they've forgotten most of the important details. So let's break 0:23:54.840 --> 0:23:57.639 these down one at a time. We'll start with encoding, 0:23:57.720 --> 0:24:00.919 writing down the memory. One of the biggest problems with 0:24:01.280 --> 0:24:06.320 encoding a memory during a crime is what's called weapon focus. 0:24:06.720 --> 0:24:09.399 So if a person has a knife on you, or 0:24:09.520 --> 0:24:12.560 is wielding a gun, or has a baseball bat or whatever, 0:24:13.200 --> 0:24:16.400 your brain can't help but to be focused on that, 0:24:17.119 --> 0:24:20.000 and as a result, it's harder to remember details about 0:24:20.000 --> 0:24:24.640 the person's face. So here's another study from Elizabeth loftis 0:24:24.720 --> 0:24:27.160 she has you come into a room and she tells 0:24:27.200 --> 0:24:30.520 you you're going to participate in some psychology exam. And 0:24:30.600 --> 0:24:34.080 while you're waiting there, you hear two people arguing, and 0:24:34.119 --> 0:24:36.480 then one person comes out of the room and he's 0:24:36.480 --> 0:24:38.880 got some grease on his hands and he's holding a pen. 0:24:39.560 --> 0:24:43.359 So that scenario one. Scenario two is to hear the 0:24:43.359 --> 0:24:45.920 same thing, the two people arguing, and then the guy 0:24:45.960 --> 0:24:49.800 comes out and he's holding a blood stained knife. And 0:24:49.800 --> 0:24:53.560 then she studies how well you can identify the person 0:24:53.680 --> 0:24:56.560