WEBVTT - Ep31 "Why do we see #TheDress differently?" 0:00:05.559 --> 0:00:08.959 What's up with those illusions on the Internet where you 0:00:09.000 --> 0:00:12.559 can hear the same sound one of two different ways 0:00:12.600 --> 0:00:15.640 depending on the word that you're looking at. And why 0:00:15.640 --> 0:00:19.159 do electrical outlets sometimes look like a face to you? 0:00:19.960 --> 0:00:23.440 How can you have full, rich visual experience with your 0:00:23.480 --> 0:00:27.080 eyes closed. And when you want to cross a street 0:00:27.160 --> 0:00:30.120 and you hit that crosswalk button, are some of those 0:00:30.120 --> 0:00:32.680 buttons fake and they don't actually do anything? 0:00:33.320 --> 0:00:34.519 And why are there some. 0:00:34.640 --> 0:00:38.040 Pictures that you can only see once you're told what 0:00:38.080 --> 0:00:41.920 you're looking at. And although brains are often celebrated for 0:00:41.960 --> 0:00:46.200 their parallel processing, what did they really be celebrated for. 0:00:49.880 --> 0:00:53.000 Welcome to Inner Cosmos with Me David Eagleman. I'm a 0:00:53.080 --> 0:00:57.560 neuroscientist and author at Stanford and in these episodes we 0:00:57.680 --> 0:01:02.160 sail deeply into our three pounds universe to understand why 0:01:02.200 --> 0:01:05.040 we perceive the world in the ways that we do. 0:01:13.520 --> 0:01:18.080 Today's episode is about expectations and what that has to 0:01:18.120 --> 0:01:24.480 do with perception. Unless you were living in outer space 0:01:24.720 --> 0:01:28.000 or off the grid in twenty fifteen, your life was 0:01:28.120 --> 0:01:33.240 touched by a very tiny, specific event that happened on 0:01:33.280 --> 0:01:37.679 a small island in Scotland. Two young people were going 0:01:37.720 --> 0:01:40.680 to get married there, and a week before the wedding, 0:01:41.080 --> 0:01:43.720 the mother of the bride was shopping around for what 0:01:43.840 --> 0:01:47.680 she was going to wear. So she finds some outfits 0:01:47.760 --> 0:01:50.720 at a store down in Chester, England that she thinks 0:01:50.760 --> 0:01:54.080 will look nice, and while she's making the decision, she 0:01:54.360 --> 0:01:57.480 snaps pictures of each of them and she buys one 0:01:57.520 --> 0:01:57.800 of them. 0:01:58.520 --> 0:01:59.880 So she's driving home. 0:01:59.720 --> 0:02:03.560 After words and she texts the pictures of the three 0:02:03.600 --> 0:02:07.320 outfits to her daughter and she tells her that she 0:02:07.480 --> 0:02:10.560 had bought the third one, and no one could have 0:02:10.720 --> 0:02:14.280 ever guessed that this particular piece of clothing that she 0:02:14.400 --> 0:02:18.680 sent a picture of, this one piece garment, is about 0:02:18.760 --> 0:02:22.440 to become the most famous outfit that ever existed in 0:02:22.480 --> 0:02:27.080 the history of humankind, because the daughter writes back to 0:02:27.280 --> 0:02:32.680 clarify which outfit the mother had bought, and she texts, oh, 0:02:32.919 --> 0:02:37.200 the white and gold one, and the mother texts back, no, 0:02:37.760 --> 0:02:43.240 it's blue and black, and the daughter replies, Mom, if 0:02:43.280 --> 0:02:45.600 you think that's blue and black, you need to go 0:02:45.639 --> 0:02:49.400 and see the doctor. So the mother shows the phone 0:02:49.440 --> 0:02:52.359 to her partner in the car, who, despite having been 0:02:52.400 --> 0:02:54.519 there and bought the dress with her, looks at the 0:02:54.560 --> 0:02:57.000 photo and says, yeah, I think it's white and gold. 0:02:57.680 --> 0:02:59.400 So when they get home, they show the picture to 0:02:59.440 --> 0:03:02.600 their younger, who agrees with the mother that the photo 0:03:02.639 --> 0:03:08.600 looks blue and black. So, given this funny disagreement, the 0:03:08.760 --> 0:03:11.880 bride to be posts the photo to her friends on 0:03:11.960 --> 0:03:16.560 Facebook to settle this, and to her surprise, she doesn't 0:03:16.600 --> 0:03:20.840 find consensus. Some think it's black and blue, others think 0:03:20.919 --> 0:03:25.800 it's white and gold, and each person feels totally certain 0:03:25.880 --> 0:03:29.160 about what they see. So for about a week, this 0:03:29.280 --> 0:03:34.200 debate bubbles around in this small island community. The day 0:03:34.200 --> 0:03:36.920 of the wedding arrives and the mother wears the dress 0:03:36.960 --> 0:03:40.280 to the event, and the issue about the photo becomes 0:03:40.480 --> 0:03:43.080 such a point of discussion that the musicians in the 0:03:43.120 --> 0:03:46.280 band allegedly almost didn't make it onto the stage to 0:03:46.360 --> 0:03:48.720 play because they were wrapped up in the debate. 0:03:49.720 --> 0:03:51.240 So a few days after. 0:03:50.960 --> 0:03:53.200 The wedding, one of the band members, who was a 0:03:53.280 --> 0:03:57.080 friend of the happy couple, she posts the photo to 0:03:57.200 --> 0:04:00.520 her blog on Tumblr, and by the end of the 0:04:00.600 --> 0:04:05.400 day it gets five thousand comments, and soon enough, the 0:04:05.640 --> 0:04:09.839 data scientists at Tumblr are examining this post because it's 0:04:09.880 --> 0:04:15.080 getting fourteen thousand views each second. That's close to a 0:04:15.160 --> 0:04:20.600 million views each minute. So a woman on the BuzzFeed 0:04:20.720 --> 0:04:24.360 social media team sets up a poll about the color 0:04:24.600 --> 0:04:27.680 for Tumblr users, and then she packs up and goes 0:04:27.720 --> 0:04:29.960 home on the subway. And by the time she gets 0:04:30.080 --> 0:04:34.440 off the subway, her phone is overwhelmed, and soon enough 0:04:34.480 --> 0:04:38.840 the BuzzFeed page hits new records for how many unique 0:04:38.920 --> 0:04:41.360 visitors were on the page at the same time, hitting 0:04:41.400 --> 0:04:45.599 almost seven hundred thousand. The number of comments on the 0:04:45.640 --> 0:04:50.200 original post increases tenfold that night. By late that night, 0:04:50.279 --> 0:04:56.120 there are five thousand tweets per minute using hashtag the dress, 0:04:56.320 --> 0:04:58.640 and by the middle of that night it's grown to 0:04:58.760 --> 0:05:02.560 eleven thousand tw wheets per minute. Within the week, more 0:05:02.600 --> 0:05:06.400 than ten million tweets are talking about the dress. This 0:05:06.680 --> 0:05:11.840 was the dress that, as they say, broke the Internet. Now, 0:05:11.920 --> 0:05:15.360 if you were, say a space alien, you might look 0:05:15.400 --> 0:05:18.920 at all this human activity and think, wait, what, why 0:05:19.080 --> 0:05:22.440 is the world stopping over a simple picture of a 0:05:22.560 --> 0:05:26.479 piece of clothing in the UK. Now, the answer, as 0:05:26.520 --> 0:05:28.640 you know, is that none of us humans would have 0:05:28.680 --> 0:05:33.040 found it interesting either, except that someone that you loved 0:05:33.040 --> 0:05:35.720 and trusted said, what do you mean you're seeing it 0:05:35.800 --> 0:05:39.880 that color? It's so clearly the other color, And you said, wait, 0:05:39.960 --> 0:05:42.599 what are you being serious? And they asked you the 0:05:42.640 --> 0:05:48.080 same and then the awe sets in. You both realize 0:05:48.120 --> 0:05:51.320 that you're looking at the same thing in the outside world, 0:05:51.680 --> 0:05:58.960 and you're having different perceptions a different experience on the inside. Now, 0:05:59.160 --> 0:06:02.880 no one was more excited about the dress than neuroscientists, 0:06:02.920 --> 0:06:08.000 because for neuroscientists this was a terrific demonstration of what 0:06:08.040 --> 0:06:11.400 we're going to talk about today. So to start things off, 0:06:11.520 --> 0:06:16.920 let's just point out how important these kinds of perceptual 0:06:17.000 --> 0:06:20.760 oddities are to neuroscience. I've spent a big chunk of 0:06:20.800 --> 0:06:26.760 my career studying illusions. I've published scientific papers about illusions 0:06:26.800 --> 0:06:30.280 in journals like Science and Nature, And some years ago 0:06:30.360 --> 0:06:33.960 I wrote a review article in the journal Nature Reviews Neuroscience, 0:06:34.240 --> 0:06:38.400 and I titled it Visual Illusions and the Brain, And 0:06:38.440 --> 0:06:42.760 in that article I laid out how powerful illusions are 0:06:43.000 --> 0:06:46.599 for figuring out what is under the hood. Sometimes I 0:06:46.600 --> 0:06:49.520 feel like illusions are interesting only to ten year olds 0:06:49.560 --> 0:06:53.880 and for most people they become nothing but entertainment. But truthfully, 0:06:54.040 --> 0:06:59.040 illusions are microscopes for understanding what is happening in the brain. 0:07:00.200 --> 0:07:04.640 Them we can reveal the systematic differences between what is 0:07:04.839 --> 0:07:07.719 actually out there in the world and what we believe 0:07:08.160 --> 0:07:12.600 is out there, And by dialing the illusion around carefully, 0:07:12.960 --> 0:07:16.840 we can usually put constraints on how the network of 0:07:16.960 --> 0:07:22.440 neurons must be operating. Now, most illusions are the type 0:07:22.440 --> 0:07:26.080 in which we measure what's being presented in the outside world, 0:07:26.240 --> 0:07:30.080 like two lines of identical lengths and you see it 0:07:30.160 --> 0:07:33.440 as two different lengths, and we say, ah, there's a 0:07:33.480 --> 0:07:37.400 systematic difference between what's on the page and what you perceive. 0:07:38.040 --> 0:07:40.000 Or maybe I show you two. 0:07:39.720 --> 0:07:43.520 Parallel lines against some background and you don't see them 0:07:43.640 --> 0:07:47.160 as parallel. Or you look at a totally static picture 0:07:47.200 --> 0:07:50.920 on a page and you swear that it's moving. But 0:07:51.000 --> 0:07:54.320 the dress was interesting because it wasn't that traditional kind 0:07:54.360 --> 0:07:58.800 of illusion. Instead, one person sees one thing and the 0:07:58.880 --> 0:08:01.480 person standing right next to them sees another. 0:08:02.640 --> 0:08:04.160 Now, what all. 0:08:03.880 --> 0:08:06.800 Illusions, including the dress, tell us right away is a 0:08:07.000 --> 0:08:11.160 foundational point that's not always intuitive, which is that we 0:08:11.240 --> 0:08:15.640 don't simply look at the world and passively receive what's 0:08:15.680 --> 0:08:22.240 out there. Instead, our brains actively construct our perception, and 0:08:22.640 --> 0:08:26.480 different brains can do so differently. So now let's move 0:08:26.600 --> 0:08:30.680 deeper into this mystery by turning to a different illusion. 0:08:31.000 --> 0:08:33.480 That took over the Internet a few years later, in 0:08:33.559 --> 0:08:35.240 May of twenty eighteen. 0:08:36.120 --> 0:08:40.040 Laurel Laurel, Laurel. 0:08:41.080 --> 0:08:44.520 Now, this was an audio file that was originally recorded 0:08:44.559 --> 0:08:47.800 by a reader in two thousand and seven for vocabulary 0:08:47.840 --> 0:08:51.360 dot com, and some students apparently re recorded that file 0:08:51.440 --> 0:08:54.679 while there was some background noise in a room. So 0:08:55.280 --> 0:08:59.360 a fifteen year old freshman in Georgia named Katie was 0:08:59.440 --> 0:09:03.120 listening to that recording and she realized that she was 0:09:03.200 --> 0:09:07.560 hearing some funny ambiguity, and she posted this little audio 0:09:07.600 --> 0:09:10.440 clip on Instagram, and the next day her friend posted 0:09:10.440 --> 0:09:12.400 it on Reddit, and then it got picked up on 0:09:12.440 --> 0:09:13.280 Twitter and. 0:09:13.320 --> 0:09:14.800 Soon it went nuts. 0:09:15.440 --> 0:09:20.120 Why Because just like the dress, people can have a 0:09:20.320 --> 0:09:25.040 different perception of the same item presented to their senses. 0:09:25.559 --> 0:09:29.240 About half the people hear the word yanny and the 0:09:29.400 --> 0:09:32.200 other half hear the word Laurel. 0:09:32.800 --> 0:09:38.600 Laurel, Laurel, Laurel, Laurel. 0:09:39.840 --> 0:09:44.640 Now, how can people hear different things? So hang tight, 0:09:44.720 --> 0:09:46.360 I'll tell you in a minute. But what I want 0:09:46.400 --> 0:09:48.720 to point out for now is that, just like the dress, 0:09:48.760 --> 0:09:52.040 some people have one experience, some people have another, same 0:09:52.160 --> 0:09:57.920 sound recording, different experiences. Now, the Yanny Laurel clip made 0:09:57.960 --> 0:10:00.960 its rounds on the internet, but it about the exact 0:10:01.040 --> 0:10:05.040 same time. In May of twenty eighteen, something even better 0:10:05.120 --> 0:10:09.920 surfaced on YouTube. A guy had posted a video where 0:10:09.960 --> 0:10:14.199 he was reviewing a children's toy from the ben Ten franchise, 0:10:14.679 --> 0:10:18.600 and the toy lights up and says something. And here's 0:10:18.640 --> 0:10:22.320 what it sounds like. It says the word green needle. 0:10:22.600 --> 0:10:34.080 So listen carefully for green needle. Okay, well, that's not 0:10:34.280 --> 0:10:37.079 actually what the toy was saying. It was actually saying 0:10:37.120 --> 0:10:41.600 the word brainstorm, which is the toy character's name. So 0:10:41.760 --> 0:10:52.320 listen for the word brainstorm. 0:10:52.440 --> 0:10:53.079 Now, I just. 0:10:53.040 --> 0:10:56.840 Played the exact same audio file in both cases, but 0:10:56.960 --> 0:11:01.800 depending on your expectation what you were listening for, you'll 0:11:01.920 --> 0:11:05.400 hear different things. So I'm going to play this file again, 0:11:05.840 --> 0:11:08.160 over and over for about twenty seconds, and I want 0:11:08.160 --> 0:11:13.240 you to think about brainstorm or think about green needle. 0:11:13.840 --> 0:11:16.840 Try to go back and forth about which one you're hearing. 0:11:17.000 --> 0:11:19.720 Switch your thinking from one to the other at any point. 0:11:39.200 --> 0:11:40.840 So, what the heck's going on here? 0:11:41.000 --> 0:11:45.480 How can a single audio file be heard two completely 0:11:45.480 --> 0:11:51.160 different ways? Seems like magic, but it's actually neuroscience. All 0:11:51.240 --> 0:11:56.679 these internet memes actually give deep insight into a fundamental 0:11:57.080 --> 0:12:00.640 and rarely appreciated property of the brain. So I'm going 0:12:00.679 --> 0:12:04.640 to unpack these illusions in a few steps. The first 0:12:04.720 --> 0:12:07.920 clue to the mystery is that the brain does not 0:12:08.240 --> 0:12:13.280 tolerate ambiguity. It really wants to come to a conclusion 0:12:13.440 --> 0:12:17.800 about exactly what's out there. Now, that's a major daily 0:12:17.920 --> 0:12:20.560 challenge for the brain because so much of what you 0:12:20.640 --> 0:12:25.080 see or hear is ambiguous. You have data points that 0:12:25.120 --> 0:12:28.400 come streaming into the brain through the eyes, or the ears, 0:12:28.480 --> 0:12:32.920 or the fingertips, but often they could be interpreted more 0:12:32.960 --> 0:12:36.120 than one way. So what does the brain do in 0:12:36.160 --> 0:12:41.520 this circumstance. It locks onto a single way of understanding it. 0:12:42.320 --> 0:12:46.320 In other words, if there are multiple possibilities, it'll force 0:12:46.440 --> 0:12:49.840 an answer. Now let's pause for just a moment to 0:12:49.880 --> 0:12:53.880 appreciate something here. When you read about the brain, you 0:12:53.960 --> 0:12:57.920 always see it celebrated for its parallel processing. It can 0:12:58.000 --> 0:13:01.280 do lots of things at once. But what it should 0:13:01.280 --> 0:13:04.240 be equally celebrated for, the thing that no one ever 0:13:04.320 --> 0:13:10.280 bothers to highlight is serialization. It takes lots of the 0:13:10.360 --> 0:13:13.840 activity and it squeezes it down to one thing. 0:13:14.080 --> 0:13:15.559 It serializes it. 0:13:15.559 --> 0:13:19.240 It takes an information that could be interpreted in lots 0:13:19.240 --> 0:13:22.160 of different ways, and it crunches it down to a 0:13:22.320 --> 0:13:23.520 single interpretation. 0:13:24.800 --> 0:13:28.760 Now, why is it so good at serializing, at. 0:13:28.600 --> 0:13:33.800 Getting possibilities down to a single answer, Because fundamentally, your 0:13:33.800 --> 0:13:38.240 brain has the challenge of controlling a giant body made 0:13:38.280 --> 0:13:41.800 of trillions of cells, and when you come to a 0:13:42.320 --> 0:13:45.880 tree in the path, it has to go either left 0:13:45.960 --> 0:13:48.719 or right around the tree. Because of the physics of 0:13:48.760 --> 0:13:51.040 the world, it cannot do both, and. 0:13:50.960 --> 0:13:53.360 So it has to make a single. 0:13:53.400 --> 0:13:57.520 Decision, go right or go left, and drag all those 0:13:57.559 --> 0:14:00.560 trillions of cells with it. Your brain it has to 0:14:00.640 --> 0:14:04.960 be good at taking possibilities and crushing them down to 0:14:05.080 --> 0:14:10.880 a single decision. And it's the same with your perceptual life. 0:14:11.360 --> 0:14:14.600 Your brain is used to dealing with a world where 0:14:14.640 --> 0:14:18.120 it has to come to conclusions, having to say, look, 0:14:18.160 --> 0:14:21.760 there are lots of possibilities here, but for me to 0:14:21.840 --> 0:14:24.520 function in the world, I have to make an assumption 0:14:25.000 --> 0:14:27.600 that what I am looking at is a piece of 0:14:27.640 --> 0:14:30.960 food or a boulder, or a bear at a distance 0:14:31.120 --> 0:14:36.000 or whatever. So the brain doesn't tolerate ambiguity, but it 0:14:36.080 --> 0:14:40.480 always says, all right, this is my answer okay, So 0:14:40.680 --> 0:14:45.120 now let's introduce one more perceptual illusion of this flavor, 0:14:45.600 --> 0:14:48.000 and then we're going to unpack what's going on. 0:14:49.480 --> 0:14:51.080 So surely you've seen this one before. 0:14:51.200 --> 0:14:54.080 You draw the outline of a cube on a piece 0:14:54.120 --> 0:14:57.240 of paper. You just draw a square, and then an 0:14:57.320 --> 0:15:00.720 offset square, and then lines connecting the corners of one 0:15:00.760 --> 0:15:03.320 to the corners of the other, so it's twelve lines. 0:15:03.400 --> 0:15:07.400 It's the outline of a cube. This little wireframe drawing 0:15:07.600 --> 0:15:09.920 is known as the Necker cube. 0:15:10.320 --> 0:15:13.479 Now you've seen this before, but as you know, if you've. 0:15:13.280 --> 0:15:18.080 Stared at one, it's perceptually ambiguous because if you stare 0:15:18.120 --> 0:15:21.320 at this little wireframe, it looks like it's coming out 0:15:21.400 --> 0:15:25.400 one way from the page, even though you could perceive. 0:15:25.080 --> 0:15:27.400 The same drawing in two different ways. 0:15:27.720 --> 0:15:30.600 Either the lower square is the face of the cube 0:15:30.640 --> 0:15:33.720 coming toward you, or the upper square is the one 0:15:33.760 --> 0:15:38.280 coming out toward you, but your brain makes a choice. Now, 0:15:38.320 --> 0:15:42.360 you could imagine a space alien who looks at this 0:15:42.400 --> 0:15:45.600 little drawing of the wireframe cube and says, okay, well, 0:15:46.040 --> 0:15:50.240 both configurations of the cube are equally probable, so I'll 0:15:50.240 --> 0:15:53.640 see it both ways at once. But we can't do that. 0:15:54.200 --> 0:15:56.960 We have to see it one way or the other. 0:15:57.120 --> 0:16:01.800 Your brain forces a single interpret and this is the 0:16:01.840 --> 0:16:06.280 same thing that's happening with the other illusions with the dress. 0:16:06.400 --> 0:16:09.640 You don't see it as both blue and black and 0:16:09.880 --> 0:16:12.520 white and gold. And in a minute we'll see why. 0:16:13.160 --> 0:16:14.680 The part I just want to say now is that 0:16:14.720 --> 0:16:17.960 your brain concludes that it is one or the other, 0:16:18.080 --> 0:16:22.440 and then it sticks with that. And likewise with Yanny Laurel. 0:16:23.080 --> 0:16:26.720 Both sounds are present in the audio file, but you 0:16:26.840 --> 0:16:30.640 don't hear Yanny and Laurel at the same time, stacked 0:16:30.680 --> 0:16:33.800 on one another. And it's exactly the same thing with 0:16:33.960 --> 0:16:39.440 brainstorm and green needle. Both interpretations are possible, but your 0:16:39.480 --> 0:16:43.920 brain won't do both at once. It collapses the possibilities 0:16:43.960 --> 0:16:48.600 to a single answer. In all these cases, even though 0:16:48.640 --> 0:16:52.160 the data is consistent with either interpretation, your brain makes 0:16:52.160 --> 0:16:55.280 a call. It goes left or right around the tree. 0:16:55.560 --> 0:16:58.800 You very clearly perceive one or the other. And this 0:16:58.920 --> 0:17:03.040 is because the brain isn't passively receiving the world. It's 0:17:03.200 --> 0:17:25.000 making choices. Okay, but how does your brain know how 0:17:25.040 --> 0:17:30.119 to collapse ambiguous data to a single interpretation. It does 0:17:30.160 --> 0:17:35.760 so by leveraging assumptions, so let's go a level deeper 0:17:35.880 --> 0:17:38.920 with the dress. Why does it happen that some people 0:17:38.960 --> 0:17:41.359 see it one way and some people the other. It 0:17:41.520 --> 0:17:44.560 happens because your brain sees a picture of a dress 0:17:44.560 --> 0:17:50.120 in the shop and it makes dozens of assumptions totally unconsciously. Now, 0:17:50.160 --> 0:17:54.600 what's amazing is that the assumptions aren't directly about the dress, 0:17:55.240 --> 0:17:58.760 but about things you didn't even know you were thinking about. 0:17:58.800 --> 0:18:03.280 What is the light source in the photograph? Is the 0:18:03.440 --> 0:18:07.840 dress mostly being lit by fluorescent lights or by sunlight? 0:18:08.760 --> 0:18:12.240 Is the dress facing a window or is the window 0:18:12.280 --> 0:18:15.840 behind it? What time of day is it, what season 0:18:16.000 --> 0:18:20.320 is it? Your brain is considering all of these questions, 0:18:20.840 --> 0:18:23.720 and fundamentally, this all has to do with a computation 0:18:23.840 --> 0:18:30.080 that it does known as color constancy. Color constancy is 0:18:30.160 --> 0:18:34.960 this sophisticated ability of our visual systems to perceive the 0:18:35.000 --> 0:18:39.000 color of something as constant even when the light source 0:18:39.040 --> 0:18:43.119 the illumination changes. So let's say I'm wearing a white 0:18:43.280 --> 0:18:46.160 T shirt and we're standing outside talking in the sunlight. 0:18:46.560 --> 0:18:50.120 You will see my shirt as white. Now we go 0:18:50.280 --> 0:18:54.479 indoors into the coffee shop and the illuminant changes. In 0:18:54.480 --> 0:18:57.720 other words, the light that's bouncing off my t shirt changes. 0:18:58.359 --> 0:19:03.280 Now it's fluorescent light compared to sunlight. The fluorescent light 0:19:03.359 --> 0:19:06.679 has a different spectrum of colors coming out, and so 0:19:06.720 --> 0:19:09.760 when those bounce off my shirt, you have a different 0:19:09.960 --> 0:19:14.600 spectrum of colors hitting your eyes, and yet you still 0:19:14.600 --> 0:19:17.800 see it as white. And then that night we go 0:19:17.920 --> 0:19:21.680 into a dance club and the lighting is blue, and 0:19:21.760 --> 0:19:25.280 yet you have no problem seeing the shirt as white, 0:19:25.600 --> 0:19:28.960 even though it's mostly blue light reflecting off the shirt 0:19:29.240 --> 0:19:32.520 into your eyes. And then afterwards we go sit by 0:19:32.640 --> 0:19:37.399 a campfire and my shirt still looks white. Your brain 0:19:37.520 --> 0:19:41.359 retains a constant perception of the color of the shirt 0:19:41.800 --> 0:19:45.240 even though the wavelengths bouncing off of it are very different. 0:19:46.320 --> 0:19:50.159 So what does this tell us, Well, it means that 0:19:50.240 --> 0:19:53.560 the way your brain determines the color is not just 0:19:53.680 --> 0:19:56.560 about the colors hitting your eye from the shirt. It 0:19:56.600 --> 0:19:59.720 has to do with something else. And that's something else 0:20:00.119 --> 0:20:04.560 is everything else in the scene. So when you're looking 0:20:04.600 --> 0:20:08.480 at my shirt, your eyes are drinking in everything else. 0:20:09.119 --> 0:20:12.840 The background, the color of the skin on my arms, 0:20:12.880 --> 0:20:17.000 the color of the floors and walls, the colors of 0:20:17.359 --> 0:20:18.120 all the other. 0:20:18.359 --> 0:20:20.680 Jeans and shirts and signs in the. 0:20:20.640 --> 0:20:24.720 Whole scene, and it uses all of that to estimate 0:20:24.800 --> 0:20:30.160 the background illumination and then make the right computation about 0:20:30.200 --> 0:20:34.199 the color of the shirt in the sunlight and the 0:20:34.280 --> 0:20:36.959 coffee shop, at the dance club, at the campfire. 0:20:37.359 --> 0:20:38.520 It's doing all of. 0:20:38.480 --> 0:20:43.080 These computations, and this is what allows it to subtract 0:20:43.320 --> 0:20:46.840 off the background lighting so that it can see what 0:20:47.040 --> 0:20:52.040 color things are most likely to actually be. That's the 0:20:52.080 --> 0:20:56.760 phenomenon of color constancy. The color of the shirt remains 0:20:56.840 --> 0:21:01.920 constant even under different illumination, and that's what allows us 0:21:01.920 --> 0:21:05.720 to see the colors of objects in the world consistently, 0:21:05.760 --> 0:21:10.120 whether we're looking under sunlight or moonlight or firelight or whatever. 0:21:11.000 --> 0:21:14.480 So the first lesson is you're not just seeing what's 0:21:14.600 --> 0:21:19.359 out there. Your brain is actively interpreting information and serving 0:21:19.480 --> 0:21:22.040 up a story to you. And I'll go into this 0:21:22.080 --> 0:21:24.399 more in a future episode. But this is why we 0:21:24.440 --> 0:21:29.600 can see strawberries as red. For example, when we change 0:21:29.600 --> 0:21:32.920 the background color such that the actual light bouncing off 0:21:32.920 --> 0:21:37.440 the strawberries is gray light, your brain can nonetheless say, okay, well, 0:21:37.640 --> 0:21:40.880 given that everything else in the scene is now greenish, 0:21:41.400 --> 0:21:44.160 I can subtrack that off and know that I'm looking 0:21:44.200 --> 0:21:48.080 at something red. Now, in order to do all of 0:21:48.119 --> 0:21:50.960 this that I've been talking about, your brain has to 0:21:51.040 --> 0:21:55.439 make lots of assumptions about what the color should be, 0:21:56.200 --> 0:22:01.239 and different brains do it differently. With the dress, you 0:22:01.320 --> 0:22:05.679 see it as either white and gold or blue and black, 0:22:06.240 --> 0:22:10.680 depending on the assumptions your brain is making. When you 0:22:10.840 --> 0:22:14.000 glance at the photo on your phone, you have no 0:22:14.119 --> 0:22:17.720 idea that your brain is doing all those sophisticated computations 0:22:17.840 --> 0:22:21.359 under the hood to tell you what is the actual 0:22:21.440 --> 0:22:25.760 color of this garment, given my assumptions about all the 0:22:25.880 --> 0:22:27.040 lighting details. 0:22:27.760 --> 0:22:29.240 The issue is that your. 0:22:29.000 --> 0:22:32.520 Brain grew up in a particular environment, maybe with a 0:22:32.560 --> 0:22:35.439 lot of snow or a lot of sunlight or fog, 0:22:36.119 --> 0:22:39.359 and your brain makes assumptions about the time of day 0:22:39.720 --> 0:22:43.199 and the season and the balance of artificial lighting to 0:22:43.320 --> 0:22:47.720 natural lighting. To make sense of this little photo, what 0:22:48.040 --> 0:22:52.199 hues does the lighting contain. If your brain ignores a 0:22:52.200 --> 0:22:54.920 bit of the blue side, you'll see the dress as 0:22:55.040 --> 0:22:58.159 white and gold. If your brain pays less attention to 0:22:58.200 --> 0:23:00.840 the yellow side of the spectrum, you'll see it as 0:23:01.040 --> 0:23:05.000 blue and black. You have no insight into the fact 0:23:05.240 --> 0:23:08.560 that your brain is making all these assumptions under the hood. 0:23:09.280 --> 0:23:11.919 Was the photo of the dress taken with the window 0:23:11.960 --> 0:23:13.120 facing it or behind it. 0:23:13.200 --> 0:23:15.080 Was it morning light or afternoon light? 0:23:15.760 --> 0:23:18.119 And is your experience of the world based on the 0:23:18.160 --> 0:23:21.919 fact that you are a mourning lark or you are. 0:23:21.800 --> 0:23:22.800 A night owl. 0:23:23.080 --> 0:23:26.600 One of my colleagues, Pascal Wallash, showed that people who 0:23:26.640 --> 0:23:30.119 were early risers were more likely to think that the 0:23:30.200 --> 0:23:33.600 dress was lit by natural light, and so they saw 0:23:33.640 --> 0:23:37.840 it as white and gold, but night owls presumably had 0:23:37.880 --> 0:23:41.720 more assumptions about artificial lighting, and they were more likely 0:23:41.760 --> 0:23:45.399 to see the dress as blue and black. Your brain 0:23:45.560 --> 0:23:48.159 is determining the color of the dress by comparing it 0:23:48.280 --> 0:23:51.240 against the other objects of the background of the photo 0:23:51.640 --> 0:23:56.520 and making its best guess at all these parameters. So 0:23:56.920 --> 0:24:00.320 your brain relies on the answers to question is that 0:24:00.359 --> 0:24:03.439 you didn't even think it was asking and the idea 0:24:03.600 --> 0:24:07.600 of imposing assumptions. This is the same with Yanny and 0:24:07.680 --> 0:24:11.840 Laurel in the auditory domain, or with green needle and brainstorm. 0:24:12.040 --> 0:24:16.840 Your brain is imposing an interpretation. But what's interesting in 0:24:16.880 --> 0:24:20.640 this case is that the assumption can be changed more easily, 0:24:21.000 --> 0:24:26.280 typically by just staring at the word visually. Because your 0:24:26.400 --> 0:24:31.280 brain is trying to disambiguate what it's hearing, and suddenly 0:24:31.520 --> 0:24:34.400 it has lots of help from the visual system because 0:24:34.600 --> 0:24:39.800 it sees a word. So the frequencies of both words 0:24:39.880 --> 0:24:43.800 yany and laurel or green needle, brainstorm, they're contained in 0:24:43.920 --> 0:24:48.240 the audio file, so just depending on how you listen 0:24:48.359 --> 0:24:52.560 for it, you can hear one or the other. So 0:24:52.600 --> 0:24:57.120 the brain constantly nails down its world by making assumptions, 0:24:57.640 --> 0:25:00.640 and we see this with everything. And even though these 0:25:01.200 --> 0:25:04.199 internet memes get all of our attention, the fact is 0:25:04.600 --> 0:25:07.560 that our brains have to make assumptions all the time. 0:25:07.880 --> 0:25:10.840 And this is because most of the inputs from the 0:25:10.880 --> 0:25:16.160 world are quite noisy. For example, you can still understand 0:25:16.280 --> 0:25:20.120 me even if my speech is choppy, or if I'm 0:25:20.160 --> 0:25:23.399 speaking and there's lots of background noise like at a restaurant. 0:25:24.160 --> 0:25:27.200 What's actually hitting your ears in these scenarios is a 0:25:27.320 --> 0:25:31.960 very messy signal, But the brain imposes an interpretation about 0:25:32.200 --> 0:25:34.920 what must have been said, and that's what you perceive 0:25:35.440 --> 0:25:38.800 what you believe you heard. A lot of your cell 0:25:38.880 --> 0:25:43.320 phone conversations are super noisy, but you typically don't realize 0:25:43.320 --> 0:25:48.919 it because you keep making your reasonable interpretations. Now, this 0:25:49.040 --> 0:25:52.880 is true of most of what is hitting your eyes 0:25:52.920 --> 0:25:56.399 and ears. You don't catch a fraction of the data, 0:25:56.800 --> 0:25:59.439 but your brain fills in the details to put together 0:25:59.480 --> 0:26:01.959 a story. And this, by the way, is what's at 0:26:02.000 --> 0:26:04.680 the heart of a lot of art and graphic design. 0:26:05.040 --> 0:26:08.000 You just see a few curves and you interpret it 0:26:08.080 --> 0:26:11.720 as a face, or a series of segmented lines and 0:26:11.800 --> 0:26:16.440 you interpret that as a body. We are always operating 0:26:16.480 --> 0:26:19.919 off thin data, but that doesn't stop us from coming 0:26:19.960 --> 0:26:24.440 to clear conclusions. And before I explain how our neural 0:26:24.480 --> 0:26:28.040 networks go about making these assumptions, let's just take a 0:26:28.080 --> 0:26:31.919 second to look at how your brain is so imperfect 0:26:32.040 --> 0:26:36.919 at this. Take paridolia, which is when you perceive a 0:26:37.119 --> 0:26:41.160 meaningful pattern where none exists, Like when you look at 0:26:41.200 --> 0:26:44.760 an electrical outlet and you see a face made up 0:26:44.760 --> 0:26:48.080 of little eyes in a sort of surprised mouth. You 0:26:48.119 --> 0:26:51.760 can't help but see that. Your brain imposes that interpretation 0:26:51.880 --> 0:26:54.920 on it. Or you see a face in the clouds, 0:26:55.359 --> 0:26:58.840 or someone sees the face of their local deity in 0:26:58.920 --> 0:27:02.960 a piece of toast. Why does this happen, Well, your 0:27:03.000 --> 0:27:06.199 brain is really wired up to see faces, and so 0:27:06.359 --> 0:27:10.960 it triggers that interpretation whenever it sees three blobs in 0:27:11.040 --> 0:27:15.560 the approximately right configuration, and the same thing can happen 0:27:15.640 --> 0:27:18.600 with sounds, like when there's some weird sound and your 0:27:18.600 --> 0:27:22.680 brain thinks it's a person shouting or someone calling your 0:27:22.760 --> 0:27:26.600 name or whatever. This is the brain working to make 0:27:26.760 --> 0:27:29.560 sense of the world around it. All it ever does 0:27:30.160 --> 0:27:34.800 is look for meaning from data in the world. In fact, 0:27:35.640 --> 0:27:39.440 typically the brain will try to impose an interpretation even 0:27:39.480 --> 0:27:44.080 if you have random noise. That's the idea with rorshak 0:27:44.200 --> 0:27:47.560 ink blots. You have these blobs on a page, and 0:27:47.640 --> 0:27:51.760 your brain reaches for some way of explaining them. Oh, 0:27:51.800 --> 0:27:54.800 that looks like a rabbit or an airplane, or an 0:27:54.800 --> 0:27:58.640 emperor on a throne or whatever. And generally a lot 0:27:58.680 --> 0:28:03.800 of life involves forcing patterns on random noise. Here's an 0:28:03.880 --> 0:28:07.520 auditory example from my colleague Diana Deutsch, who has spent 0:28:07.600 --> 0:28:12.480 her career pioneering auditory illusions. So here's an experiment where 0:28:12.520 --> 0:28:16.639 she plays mixed up audio that doesn't say anything, but 0:28:17.080 --> 0:28:21.320 it sounds like speech, and people will generally impose the 0:28:21.400 --> 0:28:23.880 interpretation of words on these. 0:28:57.280 --> 0:29:01.600 Come come, come, come, come, come, come, come, come come. 0:29:33.000 --> 0:29:38.160 This is essentially the sound version of Rorschach blots. Different 0:29:38.280 --> 0:29:41.520 people will generally hear different things, and it seems to 0:29:41.560 --> 0:29:44.520 be related to what they are thinking about or what's 0:29:44.560 --> 0:29:47.640 on their mind. So this all reminds us of the 0:29:47.880 --> 0:29:52.720 power of the brain to impose meaning. Just think about 0:29:52.760 --> 0:29:56.360 the situation when you're expecting a friend and you're looking 0:29:56.440 --> 0:30:00.240 around for him at a crowded mall. Everyone looks like him. 0:30:00.240 --> 0:30:02.560 For just a fraction of a second. You look at 0:30:02.600 --> 0:30:05.160 someone's face and you think, oh, that's him, and then 0:30:05.560 --> 0:30:09.560 five hundred milliseconds later, your visual system takes in more 0:30:09.640 --> 0:30:13.480 information and decides out never mind false alarm, And then 0:30:13.520 --> 0:30:16.880 we typically forget that we even thought that. But we 0:30:17.040 --> 0:30:20.240 are expecting to see our friend, and so our brains 0:30:20.720 --> 0:30:26.240 impose that expectation on lots of faces. Okay, so we've 0:30:26.400 --> 0:30:31.480 established that brains take ambiguous signals and squish them down 0:30:31.520 --> 0:30:35.800 to a single interpretation by use of assumptions. And that's 0:30:35.840 --> 0:30:38.520 why we see the dress as one color or the other, 0:30:38.960 --> 0:30:42.520 or we hear brainstorm or a green needle, but not both. 0:30:43.760 --> 0:30:46.520 But how do our brains actually make their choice? How 0:30:46.560 --> 0:30:50.800 do they pull this off? Neurally speaking, they do it 0:30:51.000 --> 0:30:57.360 by combining bottom up information with top down information. Now, 0:30:57.760 --> 0:31:01.720 bottom up means information and from the outside from the world. 0:31:02.080 --> 0:31:04.360 What are the air compression wave of sitting my ear 0:31:04.440 --> 0:31:06.560 drums or what are the photons sitting my retina? 0:31:07.080 --> 0:31:09.000 Those are the signals that I am receiving. 0:31:10.000 --> 0:31:12.800 But we don't interpret those bottom up signals that face 0:31:12.960 --> 0:31:18.120 value because they're usually not sufficient. Instead, your brain melds 0:31:18.200 --> 0:31:23.880 this with top down information, which means your expectations what 0:31:24.040 --> 0:31:26.600 we think is likely to be true in the outside 0:31:26.680 --> 0:31:30.640 world given our experience with it, and it's only in 0:31:30.920 --> 0:31:36.240 combination the data plus our expectations that we see anything 0:31:36.400 --> 0:31:40.520 in the world. And the surprise, I think the counterintuitive 0:31:40.600 --> 0:31:45.000 part is that your prior assumptions, your expectations, the. 0:31:45.080 --> 0:31:45.840 Top down part. 0:31:46.240 --> 0:31:50.680 This is the overwhelming majority of what determines what you see. 0:31:51.880 --> 0:31:54.880 For example, it seems like you just open your eyes 0:31:54.920 --> 0:31:57.400 and there's the world, but in fact, when you look 0:31:57.520 --> 0:32:00.720 at the visual cortex at the back of the which 0:32:00.800 --> 0:32:03.200 is the place that receives the information from the eyes, 0:32:04.040 --> 0:32:07.400 you find that only five percent of the input there 0:32:07.880 --> 0:32:09.280 is coming from the eyes. 0:32:09.360 --> 0:32:12.280 And the rest is all feedback activity. 0:32:12.640 --> 0:32:15.400 In other words, ninety five percent of the data is 0:32:15.480 --> 0:32:19.120 coming from higher levels of the visual system and other 0:32:19.400 --> 0:32:23.480 areas of the brain. In fact, What is so crazy 0:32:23.680 --> 0:32:27.320 is that you don't even need your eyes to have full, 0:32:27.960 --> 0:32:32.200 rich visual experience. You can have this with your eyes closed. 0:32:32.680 --> 0:32:37.040 And this is what we call dreams. And what's happening 0:32:37.080 --> 0:32:40.640 here is that this is all internally generated activity and 0:32:40.720 --> 0:32:43.000 none of it's entering through the eyes when you're asleep, 0:32:43.480 --> 0:32:46.440 and it's not that much different from your normal vision. 0:32:47.120 --> 0:32:49.640 So your perception of the world when you're walking around 0:32:49.800 --> 0:32:52.520 is something like an awake dream. 0:32:53.600 --> 0:32:53.719 Now. 0:32:53.760 --> 0:32:56.160 I'm going to return to this issue in future episodes, 0:32:56.600 --> 0:32:59.000 but what we want to concentrate on right now is 0:32:59.080 --> 0:33:02.320 that your eyes are not simply a camera and your 0:33:02.480 --> 0:33:05.640 ears are not simply a microphone. For those of you 0:33:05.720 --> 0:33:08.000 who have been listening for a while to this podcast, 0:33:08.440 --> 0:33:12.040 you know this is a major theme. Your brain is 0:33:12.200 --> 0:33:16.040 locked in silence and darkness and needs to make assumptions 0:33:16.360 --> 0:33:20.800 based on very thin data. So when I ask you 0:33:20.920 --> 0:33:24.880 to think about the words green needle, that is top 0:33:24.960 --> 0:33:29.840 down information that shapes how you interpret the bottom up data. 0:33:37.120 --> 0:33:40.440 In contrast, imagine that you stare at the word brainstorm 0:33:40.640 --> 0:33:44.280 while listening. You lock that in as your top down expectation, 0:33:44.440 --> 0:33:46.760 and then that shapes your bottom up data. 0:33:46.840 --> 0:33:47.760 And that's what you hear. 0:33:55.360 --> 0:34:00.200 Even though both interpretations are available, your brain surfaces is 0:34:00.280 --> 0:34:04.280 those features out of the landscape of data that match 0:34:04.440 --> 0:34:08.920 what you're looking for. In other words, your expectations. What 0:34:09.120 --> 0:34:13.839 you listen for is what you hear. And by the way, 0:34:13.920 --> 0:34:16.719 all this is related to why lip reading works. When 0:34:16.760 --> 0:34:20.000 you're in a noisy environment, you watch somebody's mouth while 0:34:20.000 --> 0:34:23.239 they're talking, and in this way you combine a bit 0:34:23.400 --> 0:34:27.120 of noisy auditory data with a bit of noisy visual 0:34:27.680 --> 0:34:30.560 and that sharpens your guess for what was just said. 0:34:31.320 --> 0:34:35.440 During the pandemic, a lot of conversations went misunderstood because 0:34:35.800 --> 0:34:39.120 people were wearing masks and lip reading went out the window. 0:34:40.280 --> 0:34:44.840 Now amazingly, this top down information is so important that 0:34:45.000 --> 0:34:48.279 sometimes you can set up a picture where you don't 0:34:48.360 --> 0:34:52.480 have any real prior assumptions and there's not enough information 0:34:52.640 --> 0:34:55.680 in the picture to see anything. And only when I 0:34:55.800 --> 0:35:00.360 tell you some interpretation does the bottom up information should 0:35:00.360 --> 0:35:01.319 make any sense at all. 0:35:01.880 --> 0:35:03.120 You can only see. 0:35:03.160 --> 0:35:06.480 What's in front of you if you're given top down direction. 0:35:07.520 --> 0:35:10.880 For example, I've put a cool picture on my website 0:35:10.920 --> 0:35:14.440 at eagleman dot com slash podcast. Take a look at 0:35:14.480 --> 0:35:17.480 this field of black and white blobs and see what 0:35:17.600 --> 0:35:20.520 it looks like to you, And presumably it looks really 0:35:20.680 --> 0:35:23.200 like nothing at all, just a bunch of blobs. But 0:35:23.320 --> 0:35:25.600 if I tell you what it is while you stare 0:35:25.600 --> 0:35:29.600 at it, then you suddenly see it. It seems immediately 0:35:29.760 --> 0:35:32.840 obvious and you cannot see anything other than that, And 0:35:32.920 --> 0:35:35.440 the only thing that's changed is that you now have 0:35:36.080 --> 0:35:40.600 a top down expectation about what you're seeing, and suddenly 0:35:40.760 --> 0:35:43.680 all these blobs make clear sense. I'm not going to 0:35:43.719 --> 0:35:45.360 tell you what the blobs are here, but if you 0:35:45.400 --> 0:35:47.719 go to the website and scroll all the way to 0:35:47.840 --> 0:35:49.799 the bottom of the page, I'll give you a hint 0:35:49.880 --> 0:35:53.240 there so you can enjoy the experience of not knowing 0:35:53.719 --> 0:35:57.560 and then knowing. And because this is a podcast, I'll 0:35:57.600 --> 0:36:01.360 give you an auditory example of this, again from Diana Deutsch. 0:36:01.840 --> 0:36:04.400 So I'm going to take a piece of music that 0:36:04.560 --> 0:36:08.279 you know, but I'm going to shift each note up 0:36:08.520 --> 0:36:11.960 or down an octave, so one note might be played 0:36:12.000 --> 0:36:14.320 an octave higher and the next note might be played 0:36:14.320 --> 0:36:17.520 an octave lower. And I want you to identify the 0:36:17.600 --> 0:36:28.279 piece of music. It's definitely one that you know. Now 0:36:28.360 --> 0:36:32.319 I assume that you couldn't identify that piece. Now I'm 0:36:32.360 --> 0:36:34.760 going to play it for you without the notes shifted 0:36:34.880 --> 0:36:45.880 up or down in octaves. Now that you know the tune, 0:36:46.000 --> 0:36:47.920 I'm just going to play that first one again and 0:36:48.000 --> 0:36:51.560 you should have little or no trouble hearing the correct melody. 0:36:59.760 --> 0:37:01.920 The only difference between the first. 0:37:01.680 --> 0:37:04.080 Time I played it and the last time is that 0:37:04.320 --> 0:37:08.120 now you have a top down expectation, and so it 0:37:08.239 --> 0:37:13.359 switches from random noise to a tune. And so these 0:37:13.400 --> 0:37:18.080 are all examples in which top down expectations are critical. 0:37:18.160 --> 0:37:21.880 Without them, you don't have any interpretation at all. And 0:37:22.080 --> 0:37:26.640 once you build an expectation, then the data have meaning. 0:37:27.640 --> 0:37:29.360 You need to be told what to see in the 0:37:29.440 --> 0:37:31.600 picture or to hear in the tune to get it. 0:37:31.880 --> 0:37:35.680 And the only difference between before and after is whether 0:37:35.760 --> 0:37:39.680 you have something to match it to some top down expectation, 0:37:40.200 --> 0:37:44.960 and as soon as you do, then you perceive. Now, 0:37:45.160 --> 0:37:47.680 just to be clear, this doesn't mean you can impose 0:37:47.920 --> 0:37:52.680 any top down interpretation. It has to match sufficiently well. 0:37:53.280 --> 0:37:56.960 The thing about brainstorm green needle is that the bottom 0:37:57.040 --> 0:38:01.000 up data can match either one of the top expectations 0:38:01.080 --> 0:38:05.680 for either word. You can hear green needle or brainstorm. 0:38:06.120 --> 0:38:10.600 Because these are possible words that can roughly match the 0:38:10.680 --> 0:38:13.600 bottom up stimulus with all of its noise, But you 0:38:13.760 --> 0:38:17.920 can't hear something totally different like blue reader or my 0:38:18.239 --> 0:38:22.320 penguin because you can't make a good enough match between 0:38:22.800 --> 0:38:26.440 data and expectation. So there has to be a sufficient 0:38:26.560 --> 0:38:30.120 match between the top down and the bottom up for 0:38:30.320 --> 0:38:34.040 perception to happen. Okay, so let's come back. 0:38:33.880 --> 0:38:36.799 To this issue about the assumptions that we make. How 0:38:36.920 --> 0:38:39.880 do we know what to assume about the world. 0:38:40.480 --> 0:38:46.920 Well, this relies almost entirely on our prior experience. For example, 0:38:46.960 --> 0:38:50.680 when you're judging depth, like how far different things are 0:38:50.760 --> 0:38:54.000 from you, which is again a totally unconscious process, you 0:38:54.080 --> 0:38:57.000 can do this by comparing the images from your two eyes, 0:38:57.440 --> 0:38:59.680 but this is only useful out to about thirty meters. 0:39:00.120 --> 0:39:03.000 So it turns out the brain has other ways to 0:39:03.160 --> 0:39:06.840 determine depth, and one of the main ones simply pivots 0:39:06.920 --> 0:39:11.400 on its experience with the world. The visual system builds 0:39:11.520 --> 0:39:16.759 up prior expectations about the relative sizes of objects. So 0:39:16.840 --> 0:39:19.200 if you're standing outside and you see a dog in 0:39:19.280 --> 0:39:21.880 the distance, then it takes up about as much space 0:39:21.960 --> 0:39:25.480 on your retina as the truck over there. You can 0:39:25.560 --> 0:39:28.640 assume that the dog is closer and the truck is 0:39:28.880 --> 0:39:32.719 farther away. Why because a close dog will look a 0:39:32.880 --> 0:39:35.680 certain size and a far away truck will end up 0:39:35.719 --> 0:39:38.360 looking about that same size, and so your brain is 0:39:38.400 --> 0:39:42.759 able to instantly make the proper assumption about how far 0:39:42.920 --> 0:39:45.839 away things are. And you might be wrong, by the way, 0:39:45.920 --> 0:39:48.439 maybe it's a miniature model of a truck that's really 0:39:48.560 --> 0:39:51.960 close and a monstrously huge dog that's really far away. 0:39:52.560 --> 0:39:55.880 But most of the time your assumptions are fine. So 0:39:57.160 --> 0:40:00.759 data doesn't just come in from the world and get seen. Instead, 0:40:01.320 --> 0:40:06.960 your visual system capitalizes on prior expectations. And although this 0:40:07.160 --> 0:40:11.080 idea isn't always intuitive, it's not a new idea. In 0:40:11.160 --> 0:40:15.919 the nineteenth century, the German physician and physicists Hermann von 0:40:16.000 --> 0:40:18.759 Helmholtz was one of the first people to entertain this 0:40:19.000 --> 0:40:23.880 model of perception. He suspected that the small amounts of 0:40:23.960 --> 0:40:27.600 information dribbling in through the eyes were just too slight 0:40:27.800 --> 0:40:32.200 to account for the rich experience of vision. So he 0:40:32.320 --> 0:40:36.439 deduced that the brain makes assumptions about the incoming data 0:40:36.640 --> 0:40:40.800 based on previous experiences, and he correctly surmised that this 0:40:40.960 --> 0:40:44.080 is how the brain can use its best guesses to 0:40:44.320 --> 0:40:48.920 rapidly turn a little trickle of information into a full picture. 0:40:49.960 --> 0:40:51.360 By the way, if you want to look this up 0:40:51.440 --> 0:40:56.040 in more depth, look up Helmholtz's notion of unconscious inference. 0:40:56.480 --> 0:40:59.400 We infer what's out there, and it all happens unconsciously. 0:41:00.000 --> 0:41:02.839 You can also look up Bayes' theorem as a way 0:41:02.920 --> 0:41:05.800 of approaching this mathematically. One way to think about this 0:41:06.080 --> 0:41:09.160 is that our judgments often rely not only on what's 0:41:09.200 --> 0:41:11.120 in front of us, but also on all of our 0:41:11.239 --> 0:41:31.600 prior experiences. So where we are so far is that 0:41:31.719 --> 0:41:35.719 the process of perceiving the world, of interpreting what we 0:41:35.760 --> 0:41:40.160 see or we hear, it's influenced by our past experiences, 0:41:40.239 --> 0:41:45.000 which shape our current expectations, and that's what determines what 0:41:45.160 --> 0:41:49.000 we think we see and hear. Now, it's sometimes the 0:41:49.120 --> 0:41:52.960 case that your brain has more than one prior expectation. 0:41:53.120 --> 0:41:55.160 It could be this, or it could be that, And 0:41:55.280 --> 0:41:58.680 in this case it's easier to witness something very interesting, 0:41:58.680 --> 0:42:00.799 which I want to tell you about now. So let's 0:42:00.880 --> 0:42:05.279 return to the Necker cube, that little wireframe drawing. So 0:42:05.360 --> 0:42:09.160 it's a very simple drawing, but it exposes something amazing, 0:42:09.400 --> 0:42:13.879 which is a competition that is always raging under the hood. 0:42:14.880 --> 0:42:17.440 Your brain is always trying to figure out what is 0:42:17.560 --> 0:42:20.320 going on out there, and the way it does this 0:42:20.920 --> 0:42:24.759 is by assessing probabilities. So this simply means you have 0:42:25.000 --> 0:42:28.320 some networks that are saying, yes, it's definitely this, and 0:42:28.400 --> 0:42:31.240 you have other networks that are saying, yes, it's definitely 0:42:31.320 --> 0:42:31.960 this other thing. 0:42:32.800 --> 0:42:34.040 In the case of the Necker. 0:42:33.880 --> 0:42:36.680 Cube, you have one network saying the cube comes out 0:42:36.719 --> 0:42:40.000 of the page this way, and the other network insisting 0:42:40.040 --> 0:42:42.080 the cube comes out of the page the other way. 0:42:43.120 --> 0:42:45.920 And in other illusions you sometimes have even more networks, 0:42:45.960 --> 0:42:49.000 each voting for their thing. But the key to understand 0:42:49.560 --> 0:42:53.560 is that it's a competition. All these networks are screaming 0:42:53.680 --> 0:42:57.160 off and trying to dominate each other, and it's a 0:42:57.719 --> 0:42:59.960 winner take all competition. 0:43:00.360 --> 0:43:01.960 It's like king of the hill. 0:43:02.040 --> 0:43:04.520 Whichever kid is able to get to the top of 0:43:04.600 --> 0:43:07.719 the hill gets to push everyone else down. In the 0:43:07.840 --> 0:43:12.760 case of local neural networks, when one is successfully firing 0:43:12.840 --> 0:43:16.600 on all cylinders, it's able to inhibit the neighboring networks. 0:43:17.080 --> 0:43:21.440 It releases neurotransmitters that keep itself propped up and at 0:43:21.480 --> 0:43:25.920 the same time inhibiting the activity of the competitors, and 0:43:26.040 --> 0:43:27.800 whichever network. 0:43:27.520 --> 0:43:29.800 Is king is what you perceive. 0:43:30.840 --> 0:43:33.960 And because it's a winner take all competition, there's only 0:43:34.080 --> 0:43:35.239 one king at any time. 0:43:35.800 --> 0:43:37.120 That's why you don't. 0:43:36.960 --> 0:43:41.600 See all the possibilities at once. You only see the winner. 0:43:42.640 --> 0:43:45.360 But here's the wacky thing with the Necker cube. It 0:43:45.520 --> 0:43:49.640 really could be either way. It's equally probable that these 0:43:49.760 --> 0:43:52.399 lions represent a cube this way or represents a cube 0:43:52.400 --> 0:43:55.239 the other way. There's a fifty percent chance of either 0:43:55.280 --> 0:43:58.440 of these. This is known as eque probable. So your 0:43:58.520 --> 0:44:03.000 brain takes this equa probable stimulus and nails it down 0:44:03.080 --> 0:44:06.120 to one choice or the other. But if you have 0:44:06.320 --> 0:44:09.000 stared at one of these drawings for more than ten seconds, 0:44:09.120 --> 0:44:14.080 you know that your brain changes its interpretation. If you 0:44:14.200 --> 0:44:16.680 stare at this wireframe, it looks like it's coming out 0:44:16.719 --> 0:44:19.280 one way from the page, but if you keep staring, 0:44:19.320 --> 0:44:21.920 it'll switch so that it looks like it's coming out 0:44:21.960 --> 0:44:24.279 the other way. And if you stare at this for 0:44:24.320 --> 0:44:27.040 a little while, you'll see that it switches back and forth. 0:44:27.239 --> 0:44:29.440 You see it one way then the other way. Your 0:44:29.480 --> 0:44:32.319 brain will stick with one interpretation for a little while 0:44:32.440 --> 0:44:35.239 and tell you that's what's in the world, and then 0:44:35.280 --> 0:44:40.680 it will suddenly change its claim. Why because, as I said, 0:44:40.719 --> 0:44:43.200 there's a fifty percent chance of interpreting the cube one 0:44:43.239 --> 0:44:46.040 way or the other, and the brain cannot see both 0:44:46.080 --> 0:44:50.080 interpretations at the same time, so it switches between them. 0:44:50.200 --> 0:44:53.600 It's the king of the hill game, but the king 0:44:53.719 --> 0:44:54.600 never lasts. 0:44:55.080 --> 0:44:57.919 Someone always manages to knock that kid off the top, 0:44:58.040 --> 0:45:00.480 and then the new kid has to defe and the 0:45:00.600 --> 0:45:04.719 throne against other invaders. And that's precisely what happens with 0:45:04.880 --> 0:45:09.320 these neural network competitions. One network wins, but it doesn't 0:45:09.440 --> 0:45:12.880 last that long before it gets unseated. And then the 0:45:12.960 --> 0:45:16.680 other network is active in a loop of self reinforcing 0:45:16.760 --> 0:45:20.560 neurons firing. It gets to keep control be king of 0:45:20.640 --> 0:45:22.720 the North for a little bit, but then the first 0:45:22.800 --> 0:45:26.239 one unseats it again. So what you see with the 0:45:26.360 --> 0:45:31.520 simple drawing is the ever present, active battle in your 0:45:31.600 --> 0:45:35.640 skull to control perception. So, in other words, if you 0:45:35.800 --> 0:45:40.040 have two possible top down models, either. 0:45:39.800 --> 0:45:44.080 Of which could equally be right, they'll fight and you'll 0:45:44.200 --> 0:45:48.800 believe whoever the temporary winner is, and then you'll believe 0:45:48.840 --> 0:45:51.000 the next guy when he's back in power, and then 0:45:51.040 --> 0:45:51.760 the first network. 0:45:51.800 --> 0:45:52.080 Again. 0:45:53.520 --> 0:45:57.080 Now the dress tends not to switch, and this is 0:45:57.160 --> 0:46:02.400 because it's not equiprobable. Our brain has developed very clear 0:46:02.960 --> 0:46:07.120 prior expectations about lighting and fabric and windows and so on. 0:46:07.800 --> 0:46:10.960 So my brain makes an interpretation and your brain makes 0:46:11.000 --> 0:46:14.360 an interpretation, and there's no reason for either one of 0:46:14.480 --> 0:46:17.560 them to question it. It's like playing King of the 0:46:17.640 --> 0:46:20.359 hill against some small puppies. No one's going to knock 0:46:20.400 --> 0:46:24.320 you off the throne. And that's why it's so difficult 0:46:24.600 --> 0:46:27.600 to change your interpretation of the dress, even when you're 0:46:27.719 --> 0:46:30.840 told that some other interpretation is possible. 0:46:31.400 --> 0:46:32.920 Your brain relies on. 0:46:33.200 --> 0:46:37.120 Deep assumptions about the world, and it's generally just too 0:46:37.239 --> 0:46:41.040 hard to unseat the monarch. But what the Necker cube 0:46:41.160 --> 0:46:45.160 reveals is that our brain's interpretation of the world can 0:46:45.239 --> 0:46:49.160 be quite active if there are other equally likely interpretations 0:46:49.239 --> 0:46:51.960 to be had, So the way we see the world 0:46:52.440 --> 0:46:56.160 can change from moment to moment. Now as just a 0:46:56.239 --> 0:46:59.640 one minute tangent. The funny thing is that you think 0:46:59.800 --> 0:47:03.560 you are making the cube switch interpretations by yourself. In 0:47:03.640 --> 0:47:06.279 other words, you feel like you're doing it consciously when 0:47:06.320 --> 0:47:07.840 the cube switches back and forth. 0:47:08.560 --> 0:47:11.520 But let's say we measure this. You stare at the 0:47:11.560 --> 0:47:15.160 little cube and you hold down one key when you. 0:47:15.239 --> 0:47:17.799 See it in this configuration, and as soon as your 0:47:17.840 --> 0:47:20.520 perception switches and now looks the other way, you hold 0:47:20.600 --> 0:47:23.120 down the other key. And you do this for a while, 0:47:23.520 --> 0:47:26.120 back and forth and back and forth. You hold down 0:47:26.160 --> 0:47:29.399 a key to let me know which perception you are seeing. 0:47:30.200 --> 0:47:32.720 And remember how amazing this is because nothing is changing 0:47:32.800 --> 0:47:34.239 on the page. It's only in your head. 0:47:35.040 --> 0:47:35.440 Anyway. 0:47:35.680 --> 0:47:38.319 When we look at the data, it's clear that your 0:47:38.400 --> 0:47:42.240 results follow a particular mathematical distribution called a gamma distribution, 0:47:42.840 --> 0:47:46.719 which comes from a random process. For the efficionados, this 0:47:46.840 --> 0:47:50.719 is consistent with a poissone process. All this means is 0:47:50.760 --> 0:47:54.320 that this switching is random, and this is exactly the 0:47:54.400 --> 0:47:58.680 distribution you would expect from randomness. Sometimes you have the 0:47:58.800 --> 0:48:01.640 winning network holding on to the throne for a long time, 0:48:01.719 --> 0:48:04.000 so as for a short time, and on average it 0:48:04.120 --> 0:48:07.600 lasts this medium amount of time before it switches. The 0:48:07.800 --> 0:48:12.840 point is you think you're switching consciously, but it's just random. 0:48:13.480 --> 0:48:16.920 The reason you take credit is because you think, Okay, 0:48:16.920 --> 0:48:19.160 I'm seeing it this way, and I really want to 0:48:19.160 --> 0:48:21.280 make it switch the other way, and I'm going. 0:48:21.239 --> 0:48:23.279 To consciously work to switch it. 0:48:23.719 --> 0:48:28.480 Okay, almost there, not quite working, still trying, and then 0:48:28.520 --> 0:48:32.720 it randomly switches and you take credit for it. Here's 0:48:32.719 --> 0:48:35.000 an analogy to help us understand that. You know those 0:48:35.320 --> 0:48:38.200 pedestrian crossing buttons that you push when you want to 0:48:38.320 --> 0:48:41.480 cross the street, and the little walk signal eventually shows 0:48:41.560 --> 0:48:43.120 up and lets you know that you're safe to walk. 0:48:44.000 --> 0:48:47.800 Some fraction of those buttons are placebos. They're fake. You 0:48:48.000 --> 0:48:50.400 hit them, but they don't do anything. 0:48:50.680 --> 0:48:52.920 You wait for exactly the same amount of time that 0:48:53.080 --> 0:48:56.360 you would have waited anyway, but you have a sense 0:48:56.400 --> 0:49:00.200 of control, an illusion of power over the light, even 0:49:00.239 --> 0:49:02.560 though the timing doesn't change one bit. 0:49:03.200 --> 0:49:06.400 And this is exactly the situation with this switching of 0:49:06.640 --> 0:49:07.440 the Necker cube. 0:49:07.960 --> 0:49:11.759 You consciously try to change it, and when it eventually 0:49:11.960 --> 0:49:14.880 changes on its own, you think, yeah, that was because 0:49:14.920 --> 0:49:18.400 of me. But when we measure the switching times, it 0:49:18.520 --> 0:49:21.919 doesn't change anything at all, whether you're trying or not trying, 0:49:22.000 --> 0:49:26.560 whether you're banging on that button or ignoring it. Okay, 0:49:26.640 --> 0:49:28.560 so now I want to zoom back up to the 0:49:28.640 --> 0:49:30.960 big picture about what we've been talking about, which is 0:49:31.040 --> 0:49:33.800 how your brain makes assumptions about things, and how in 0:49:33.920 --> 0:49:38.080 some circumstances these assumptions can fight it out. So we 0:49:38.200 --> 0:49:41.880 see this in language often take the example of puns. 0:49:42.440 --> 0:49:45.720 Puns strike us as funny because we're able to switch 0:49:45.840 --> 0:49:48.839 back and forth and see the same thing in two 0:49:49.000 --> 0:49:52.040 different ways. What do you get when you drop a 0:49:52.239 --> 0:49:57.560 piano down a mine shaft a flat minor. The point 0:49:57.560 --> 0:50:00.359 about puns is that we know from the s mile 0:50:00.440 --> 0:50:02.719 on the other person's face that there's some joke to 0:50:02.800 --> 0:50:06.239 be had, and so we search for other interpretations, and 0:50:06.360 --> 0:50:09.200 we can switch back and forth between them, just like 0:50:09.280 --> 0:50:12.719 a Necker cube. But something I found interesting is that 0:50:13.160 --> 0:50:16.160 brains can be lazy, and we don't always bother or 0:50:16.280 --> 0:50:19.759 seeking other interpretations. If you don't have a strong enough 0:50:19.880 --> 0:50:24.399 reason to have more than one interpretation, then you stick 0:50:24.480 --> 0:50:28.120 with what you've got. And this is often true in language, 0:50:28.280 --> 0:50:32.600 which is very low bandwidth and depends enormously on assumptions. 0:50:33.120 --> 0:50:35.279 So the other night I was at a party and 0:50:35.440 --> 0:50:38.600 somehow the conversation moved in a direction where I mentioned 0:50:39.080 --> 0:50:44.040 the famous book by Rachel Carson called Silent Spring. It 0:50:44.200 --> 0:50:45.800 just so happened that no one there had heard of 0:50:45.840 --> 0:50:48.880 this book. So in a sentence. I explained that the 0:50:49.120 --> 0:50:53.320 author had argued that if pesticide use continued, there wouldn't 0:50:53.320 --> 0:50:56.759 be any more birds, and so the spring season would 0:50:56.800 --> 0:50:59.560 come around and we would hear no more chirping. It 0:50:59.560 --> 0:51:03.879 would be And I was sort of surprised when everyone said, oh, 0:51:04.880 --> 0:51:07.440 like I had just cleared up some confusion for them, 0:51:08.080 --> 0:51:11.760 because it turns out that when I had said silent spring, 0:51:12.440 --> 0:51:14.760 the person to my left thought I was talking about 0:51:14.840 --> 0:51:17.720 a spring like a creek, so she interpreted the title 0:51:17.840 --> 0:51:21.479 as silent river, and the person to my right thought 0:51:21.520 --> 0:51:26.120 of spring like boeing boeing spring. And the person across 0:51:26.160 --> 0:51:28.520 from me thought I was talking about the word spring 0:51:29.120 --> 0:51:32.880 like the verb to jump, so he pictured silent spring 0:51:33.440 --> 0:51:37.239 as a lion springing on him silently. And this is 0:51:37.320 --> 0:51:39.800 typical of the way that we take in little bits 0:51:39.840 --> 0:51:43.120 of data and impose an interpretation on them, and then 0:51:43.160 --> 0:51:47.560 we're done. Our brains aren't generally incentivized to keep looking 0:51:47.640 --> 0:51:51.400 for interpretations. You pick one and that's it. And by 0:51:51.440 --> 0:51:54.600 the way, that's typically what happens with Laurel and Yanny. 0:51:54.960 --> 0:51:57.919 If you didn't know to listen hard for something else, 0:51:58.120 --> 0:52:02.920 you probably wouldn't. And with green needle and brainstorm. Unless 0:52:02.960 --> 0:52:07.080 you were told to switch your perception, you probably wouldn't 0:52:07.120 --> 0:52:10.440 have even thought to try it. And so I often 0:52:10.640 --> 0:52:13.520 wonder about the ways that we do this with many 0:52:13.680 --> 0:52:17.560 things around us. We pick some top down model and 0:52:17.719 --> 0:52:20.080 that seems to match the bottom up data, and it 0:52:20.200 --> 0:52:24.000 doesn't strike us to examine further because we're pretty sure 0:52:24.080 --> 0:52:26.759 we have a match. I'll leave this as an open 0:52:26.840 --> 0:52:29.320 question for all of us to think about places in 0:52:29.400 --> 0:52:32.520 our life that maybe we haven't even thought to re 0:52:32.840 --> 0:52:37.759 examine more deeply. So to wrap this up, when this 0:52:37.960 --> 0:52:41.120 woman in the UK sent a little cell phone photo 0:52:41.239 --> 0:52:44.600 to her daughter about her dress, it not only broke 0:52:44.719 --> 0:52:48.800 the Internet, the more importantly, it breaks for us a 0:52:48.960 --> 0:52:53.360 critical assumption that almost everyone carries around, the assumption that 0:52:53.480 --> 0:52:55.719 when I look at the world and you look at 0:52:55.760 --> 0:53:00.319 the world, we see the same thing. The naive umption 0:53:00.600 --> 0:53:04.200 is that there is simply truth out there and it's 0:53:04.320 --> 0:53:07.120 just a matter of opening your eyes. But the dress 0:53:07.440 --> 0:53:11.479 and hundreds of other illusions reveal that we don't see 0:53:11.560 --> 0:53:18.120 the world out there directly. Everything is interpretation. We only 0:53:18.239 --> 0:53:22.520 have a bit of data dribbling in through our peripheral devices, 0:53:22.600 --> 0:53:26.240 our sensory organs, and that data enters into a brain 0:53:26.760 --> 0:53:30.440 that's already churning and bubbling with its own activity, its 0:53:30.480 --> 0:53:35.400 own expectations, and so all we ever perceive is the 0:53:35.680 --> 0:53:39.319 best guess from our neural networks about what is going 0:53:39.480 --> 0:53:43.640 on out there, given a little rough data and a 0:53:43.800 --> 0:53:48.759 lot of assumptions shaped by our past experiences. So the 0:53:48.920 --> 0:53:52.440 next time you see a face in an electrical outlet, 0:53:52.680 --> 0:53:55.719 where you see a shape in aurorshack blot, or you 0:53:55.880 --> 0:53:59.480 see the dress and feel certain about its color, just 0:53:59.600 --> 0:54:03.279 remember you are not seeing the world as it is. 0:54:03.960 --> 0:54:06.319 You are seeing it as you are. 0:54:10.560 --> 0:54:13.520 Go to Eagleman dot com slash podcast for more information 0:54:13.840 --> 0:54:17.680 and to find further reading. Send me an email at 0:54:17.880 --> 0:54:21.880 podcast at eagleman dot com with questions or discussions, and 0:54:22.000 --> 0:54:24.120 I'm going to be making episodes in which I address 0:54:24.200 --> 0:54:26.279 those reaching. 0:54:25.960 --> 0:54:29.440 Out on a narrow road from my internal world to yours. 0:54:29.800 --> 0:54:32.000 This is David Eagleman, and thank you for joining me 0:54:32.239 --> 0:54:33.640 in the inner cosmos.