[00:01] "I think, therefore, I am." A single microscopic brain cell cannot think,
[00:14] and a few more, and connect them all, be able to think and know that it exists.
[00:30] be made from such simple ingredients? but, more importantly,  how we are.
[00:42] Today, using what neuroscientists know so far, function like a brain! ( all cheering, applauding )
[00:57] ♪
[01:11] both in size and abilities. But when enough are together, they can do amazing things When the collective power of a group working together
[01:28] that is called "emergence." are connected to the people around us. can work together to accomplish amazing feats.
[01:44] A great example is "wisdom of the crowds." knows the right answer to a question, collectively, they could all somehow know the right answer.
[01:58] conducted the "Bean Jar" experiment. He asked 56 students to guess the number of jellybeans in a jar. not a single one of them guessed the right answer.
[02:14] But amazingly, when he took the average of their guesses, what he got was a number within just 3% of the real answer. but others guessed way too low,
[02:27] and from a whole bunch of wrong guesses, What else can a crowd do? and had each one of them act like a brain cell,
[02:42] turning on or off in response to the actions of other people, like the one in our brain? could intelligence, emotions,
[02:56] If I recruited every single person and arranged them like neurons, but something that can think and feel
[03:11] Well, this is the China Brain thought experiment, first proposed by Lawrence Davis and, later, Ned Block. It's never been done before and, well, unfortunately,
[03:24] I made some calls, and like a lot of them are busy. could even do. but I want to blow a neural network
[03:38] And what better crowd to use than one made of the people whose emergent properties made me who I am today? ♪
[03:56] ( birds chirping ) I recruited Chris Eliasmith, at the University of Waterloo.
[04:11] going down to the heart of Stilwell, We're going to do something a little bit weird. Um. - Right. - OK? But with a crowd of people.
[04:25] I looked into it, that's made up of only 300-some-odd neurons. and where better to get these people to make a brain
[04:39] than my hometown of Stilwell? This was the community that, in many ways, made me who I am.
[04:51] Some of my earliest memories are from here. and they would have sno-cones during the summer.
[05:03] It was the most awesome, delicious thing ever. but around here, the main thing that I saw being grown
[05:15] whose slogan was "High on grass." - ( Chris laughs ) - It was pretty...pretty edgy for the time. You know, building brains is in my job description.
[05:31] Michael:  Chris is known for is neural network, or SPAUN, which is one of the world's most complex It uses 6.6 million simulated neurons
[05:46] and image recognition. The first was made by Dr. Frank Rosenblatt His network, called the Perceptron,
[06:02] and he hoped it would become capable of learning, and after some controversy, fell by the wayside.
[06:14] came back upon Dr. Rosenblatt's work, that the field of artificial neural networks Today, it is alive and well.
[06:26] SPAUN, and even neural networks used in self-driving cars, are expanding the possibilities of computer learning. That's a good question.
[06:39] I would recommend something like vision. Vision. Let's make this brain see. let's look at how visual processing works.
[06:55] Light information from every point on the cat This information gets sent to our visual cortex. V1 through V6.
[07:10] activated by specific features, The features that are detected which puts all the pieces of the image together,
[07:23] where we recognize the object we're looking at, what it means what feelings we have towards it. But what should we have our brain recognize?
[07:36] or images that depend on too much detail, - Ok. - I want to be the one who draws a digit, and then you will be on the output side.
[07:50] not because I showed it to someone and they telephoned it back to you, but because they processed it intelligently. So we should take some small number of them
[08:05] and really just show them each a little bit of the image. So if, for instance, we're able to put like 25 people in that kind of front row, the "input" layer, then whatever image we show
[08:17] - Exactly. Right. - Twenty-five pieces. 1, 2, 3, 4, 5, 11, 12, 13, 14, 15,
[08:29] 21, 22, 23, 24, 25. These are our retina cells, that's literally standing, like, in a field.
[08:42] They merely need to indicate whether or not They should start spiking like a neuron. They could jump up and down,
[08:56] OK, I like that. we're talking about how brain cells, neurons, by sending an electric message from one cell...
[09:11] and it travels down the axon of the cell. the cell will fire an electronic message down its axon.
[09:24] It's either firing or it's not firing. to be either on or off-- To illustrate our visual input,
[09:39] onto a grid divided into 25 squares, or pixels. If a neuron receives a pixel with writing on it,
[09:51] The V1 layer identifies pixels in the retinal layer and the V2 layer identifies particular combinations of lines from V1 that form angles.
[10:05] We're only working with black and white in this case. All right, sorry, V3. Michael:  V4 neurons will fire
[10:19] have been detected. And so actually the next one is called IT. Michael:  Now don't worry.
[10:33] They exist, they're responsible But for our demonstration, we don't need them. which is the final layer needed for visual processing
[10:48] Our IT will consist of ten neurons They will be looking at neurons in V4, and will only fire when their corresponding neurons fire.
[11:01] representing the shapes of an 8 fire, Voila! We just recognized a number.
[11:13] So after that, I think that's where I'll be. It's gonna be me making a decision about what digit I think was actually shown at the end. so one question is, where do you put that many people?
[11:27] The question is, is it gonna work? - I am hopeful right now. - We got our work cut out for us. to head to my high school football field
[11:42] I spent a lot of time here at the clarinet. the actual logistics of getting all these people together.
[11:57] Everyone is going to be wearing a shirt We're also gonna give everyone one of those, um, - The big bibs? - Bibs. Thank you. Yes.
[12:12] who doesn't know what us sportos talk about. so if something goes wrong, and say, "Are you damaged? What do you need?"
[12:28] I will be here. This is gonna be the input layer. You are gonna be way down in the end zone on the output side.
[12:41] -Ok. - So when you make your prediction, we'll put that on the Visitor's side and I'll reveal the Home number as what I really wrote.
[12:55] is the amount of space we're going to need to also have the right eye lines. meaning that it's easy to see whoever you have to pay attention to.
[13:09] spread out in five layers Every single participant will be assigned And it's complicated, so their positions on the field
[13:24] has a clear line of sight In a way... - We shall find out. - We'll find out!
[13:38] All right. ( crowd chattering )
[13:51] Michael:  So what does it take to turn Stilwell into a brain? Well, seven tents, 550 chairs, t-shirt and hats,
[14:03] this cute little Gator, and of course, our medic, Brian. A community is something that is bigger
[14:16] Now, today, I'm feeling pretty excited about because there's a zero percent chance of rain, but a 100% chance of brain.
[14:31] The gates are open, and our neurons are filing in. associated with the layer of the brain they will represent. Just want go in the center?
[14:46] to get in position. Your job is to say, "Is there writing on my square?" or black marks on it, raise your flag-- oh, and stand up.
[15:03] remember every detail about how this brain But just in case you don't, here's a refresher. break the squares up, and hand them out
[15:18] The retinal neurons will only fire if they have writing on their pixel. you guys are V1. three retinal neurons in front of them
[15:32] of their assigned retinal neurons fire, You're a bit more advanced. The V2 neurons will be watching the V1 layer,
[15:47] revealing angles. Their firing reveals combinations of angles You all are infratemporal cortex.
[16:04] Part of the IT's function is to inhibit For example, if V4 neurons are indicating an 8 will outrank a 6 because an 8 has more features.
[16:21] Chris will determine the number by interpreting the results from the IT layer. Michael (over loudspeaker): All right! Stand by.
[16:42] to the photoreceptors in the retina. 19... I have distributed
[16:56] the input to the retina layer. ( all cheering, applauding ) Three, two, one, think!
[17:09] passing off signals to V1. cuing V4 and IT. Look at all this processing.
[17:21] where he will tell me what you guys have processed. way faster than I thought. It was really interesting to watch.
[17:36] because we actually kind of have two answers at the end. which is kind of make a guess sometimes based on the best evidence. What numeral do you think I drew?
[17:51] Let's get that up on the Visitor's scoreboard. The numeral I truly drew... ( all cheering, applauding )
[18:06] I think we can perfect this a little bit, And by "Chris," I mean all of you. Michael:  We gave each of our IT neurons a clear tube and plastic balls
[18:20] Every time they see one of the neurons they're watching fire, If a neuron they're inhibited by Michael: Our model had a mistake.
[18:32] meaning that if 254 is firing, - you just sit down. - OK. I'm gonna do that right now.
[18:45] it was time to try again. Take that one. Thank you. 24...
[18:57] go! in V2 and V4.
[19:10] Looks like our brain died. The processing stopped in V2. Not a single person in V4
[19:25] has been activated. when the light went through the lens and got to the retina. - It's lookin' that way. - Wow, who would have thought
[19:40] would be the actual people who get to be people OK, I think I'm gonna do an 8 this time. And I'm gonna put it kind of up in this corner,
[19:54] This is pretty weird, it's not centered, Let's see if our brain can recognize it. to a particular retinal neuron.
[20:07] we put numbers on the back of each one. Three, two, one, go!
[20:19] Whoa! That was fast. because this is the Mind Field Play of the Game. and bam! the 13 retinal cells connected to those locations are firing.
[20:35] V1 reads the formation perfectly. or diagonal line has been caught. low and to the right, which my 8 had.
[20:49] If retinal cells 23, 24, and 25 all fire together, Champion reflexes there, folks. If the V1 neurons a V2 player is watching, fire,
[21:01] made some corner angle. but it all comes down to IT. but there can only be one MVP,
[21:14] 0 is inhibited by 8. sit down and let that neuron score! call...a gr8 play.
[21:28] make me think that you wrote an 8. An 8? Let's put an 8 up on the scoreboard. And as it turns out, the numeral that I did write...
[21:40] - Nice job! - That was good. Now it's time to really put the system to the test.
[21:52] but then I'm gonna add a line down here and I'm gonna do a dot right there. Three, two, one, go!
[22:09] My guess is a 1. ( cheers, applause ) Go!
[22:23] Man, these people are good. I wrote a 7. Nice work.
[22:36] for the last test I want to see what will happen I'm not even going to draw a number; instead, I'm going to fill in every single cell.
[22:50] This means that every single neuron in the retina will fire, My prediction, of course, it's gonna look like an 8, because an 8 is a numeral that fills in a lot of the cells.
[23:04] - Chris, are you ready? - Ready! Three, two, one, go!
[23:17] ( laughing ) - ( all cheering ) - Michael: Look at that. It looks to me like you essentially opened up the eye
[23:30] Right, yeah. So what I did is, I didn't even draw a numeral. - So what does the brain think that I drew? - The 8. We were able to single all of that mess
[23:46] and it really was the smartest guess. Michael: Congratulations to the entire infratemporal cortex, You guys have been amazing. Great work.
[24:01] Today, I was a neuron. My favorite part was just the whole experience, It's just a really good simulation of how the brain works,
[24:14] and it was just really cool to take some information from that. ( all cheer )
[24:32] Made up of only a couple hundred neurons, it was nonetheless able to process the image We created a living, breathing model
[24:48] And our demonstration was a new way how the human brain processes visual information. We were able to watch it process...think...
[25:00] Its success shows what we can achieve by working together, and we only used a small fraction So imagine how powerful
[25:15] but a hundred billion  human neurons could be. a hundred billion people is about how many humans
[25:27] have ever existed in the history of Earth. so I guess that means get procreating...please.
[25:39] No. I'd like to thank every neuron and the entire community that supported us. none of this would have been possible.
[25:55] thanks for watching. - ( no audible dialogue ) - ♪