Placeholder Image

字幕表 動画を再生する

  • "I think, therefore, I am."

  • But am I?

  • I think. Ha.

  • A single microscopic brain cell cannot think,

  • is not conscious,

  • but if you bring in a few more brain cells,

  • and a few more, and connect them all,

  • at a certain point, the group itself will

  • be able to think

  • and experience emotions

  • and have opinions and a personality

  • and know that it exists.

  • How can such astonishing things

  • be made from such simple ingredients?

  • Well, answering that question means learning not only who we are,

  • but, more importantly, how we are.

  • Today, using what neuroscientists know so far,

  • I am going to make my hometown

  • function like a brain!

  • ( all cheering, applauding )

  • A single brain cell is tiny,

  • both in size and abilities.

  • But when enough are together, they can do amazing things

  • like be aware of themselves.

  • When the collective power of a group working together

  • is greater than the sum of their individual parts,

  • that is called "emergence."

  • In a similar fashion, we as individuals

  • are connected to the people around us.

  • Those connections form communities that, when functioning properly,

  • can work together to accomplish amazing feats.

  • A great example is "wisdom of the crowds."

  • Even if not a single person in a crowd

  • knows the right answer to a question,

  • collectively, they could all somehow know the right answer.

  • In 1987, economist Jack Treynor

  • conducted the "Bean Jar" experiment.

  • He asked 56 students to guess the number of jellybeans in a jar.

  • Now, as you can probably guess,

  • not a single one of them guessed the right answer.

  • But amazingly, when he took the average of their guesses,

  • what he got was a number within just 3% of the real answer.

  • Now, some people guessed way too high,

  • but others guessed way too low,

  • so all together, their errors balanced out,

  • and from a whole bunch of wrong guesses,

  • the true answer emerged.

  • What else can a crowd do?

  • If I got a bunch of humans together

  • and had each one of them act like a brain cell,

  • turning on or off in response to the actions of other people,

  • could I make a neural network

  • like the one in our brain?

  • And if I had enough people,

  • could intelligence, emotions,

  • a mind, emerge?

  • If I recruited every single person

  • in the country of China

  • and arranged them like neurons,

  • would the result not only be a simple brain,

  • but something that can think and feel

  • and be aware of its own existence?

  • 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,

  • I don't have access to everyone in China.

  • I made some calls, and like a lot of them are busy.

  • But the first step is to see what a crowd in real life

  • could even do.

  • This hasn't been done successfully before,

  • but I want to blow a neural network

  • up to the scale of a crowd.

  • And what better crowd to use than one made of the people

  • whose emergent properties made me who I am today?

  • That's right, I am going home to Stilwell, Kansas.

  • ( birds chirping )

  • Michael: For help designing the brain

  • we would make out of people,

  • I recruited Chris Eliasmith,

  • director of the Center for Theoretical Neuroscience

  • at the University of Waterloo.

  • So Chris, we're headed south,

  • going down to the heart of Stilwell,

  • - where I grew up. - Nice.

  • We're going to do something a little bit weird. Um.

  • I want to create a brain.

  • - Right. - OK? But with a crowd of people.

  • It sounds like a challenge, for sure.

  • I looked into it,

  • and I found that the roundworm has a brain

  • that's made up of only 300-some-odd neurons.

  • - That's right. - We can get 300 people,

  • and where better to get these people to make a brain

  • than my hometown of Stilwell?

  • This was the community that, in many ways, made me who I am.

  • Michael: This is all downtown Stilwell.

  • Some of my earliest memories are from here.

  • This used to be, and maybe still is, a feed store,

  • and they would have sno-cones during the summer.

  • It was the most awesome, delicious thing ever.

  • But as you can see,

  • a lot of corn is grown in Kansas,

  • but around here, the main thing that I saw being grown

  • - was just sod. - Oh, really?

  • Yeah, There's a famous sod farm around here

  • whose slogan was "High on grass."

  • - ( Chris laughs ) - It was pretty...pretty edgy for the time.

  • OK, so back to the brain that we're gonna make.

  • You know, building brains is in my job description.

  • I wrote a book called How to Build a Brain.

  • Michael: Chris is known for is neural network,

  • the Semantic Pointer Architecture Unified Network,

  • or SPAUN, which is one of the world's most complex

  • computer simulations of the brain.

  • It uses 6.6 million simulated neurons

  • to perform functions like counting, reasoning,

  • and image recognition.

  • SPAUN is cutting-edge,

  • but neural networks are nothing new.

  • The first was made by Dr. Frank Rosenblatt

  • of Cornell University in 1957.

  • His network, called the Perceptron,

  • was designed for image recognition,

  • and he hoped it would become capable of learning,

  • just like a brain.

  • But the project was only partially successful,

  • and after some controversy, fell by the wayside.

  • It was only when researchers in the 1980s

  • came back upon Dr. Rosenblatt's work,

  • and as computing power increased,

  • that the field of artificial neural networks

  • came back to the mainstream.

  • Today, it is alive and well.

  • SPAUN, and even neural networks used in self-driving cars,

  • are expanding the possibilities of computer learning.

  • If I want to make a brain out of people, where do I start?

  • That's a good question.

  • I think the first thing we want to do is figure out

  • what we want our brain to do.

  • I would recommend something like vision.

  • Vision. Let's make this brain see.

  • Michael: Before we can design the intricacies

  • of the brain we're making,

  • let's look at how visual processing works.

  • Let's say we look at a cat.

  • Light information from every point on the cat

  • lands on the retina.

  • This information gets sent to our visual cortex.

  • The visual cortex is structured in layers--

  • V1 through V6.

  • Each of these layers are made up of neurons

  • activated by specific features,

  • like lines, angles, and shapes.

  • The features that are detected

  • are sent to the infratemporal cortex

  • which puts all the pieces of the image together,

  • and we get our Eureka! moment

  • where we recognize the object we're looking at,

  • what it means what feelings we have towards it.

  • I love cats.

  • But what should we have our brain recognize?

  • We don't want really high resolution images

  • or images that depend on too much detail,

  • - Ok. - so things like letters and digits.

  • Let's say we use digits. Ok?

  • - Ok. - I want to be the one who draws a digit,

  • and then you will be on the output side.

  • You should be able to determine what I've drawn;

  • not because I showed it to someone and they telephoned it back to you,

  • but because they processed it intelligently.

  • That's what we need to figure out,

  • how we're gonna show an input to our people.

  • So we should take some small number of them

  • and put them at the front, as the retina,

  • 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

  • should be made up of 25 pixels.

  • - Exactly. Right. - Twenty-five pieces.

  • I'm gonna draw 25 people.

  • 1, 2, 3, 4, 5,

  • 6, 7, 8, 9, 10,

  • 11, 12, 13, 14, 15,

  • 16, 17, 18, 19, 20,

  • 21, 22, 23, 24, 25.

  • See? I can count.

  • These are our retina cells,

  • and each one is an individual person

  • that's literally standing, like, in a field.

  • What do they then do next?

  • They merely need to indicate whether or not

  • their cell is on or off.

  • - All right. - So, they should start firing.

  • They should start spiking like a neuron.

  • What could they do to indicate

  • that they're firing or not?

  • They could jump up and down,

  • they could wave a flag...

  • OK, I like that.

  • When Chris and I use words like "firing" and "spiking,"

  • we're talking about how brain cells, neurons,

  • talk to one another

  • by sending an electric message from one cell...

  • to another.

  • It's called an "action potential,"

  • and it travels down the axon of the cell.

  • When the ionic flow into a brain cell

  • reaches a certain threshold,

  • the cell will fire an electronic message down its axon.

  • So a neuron can either be on or off.

  • It's either firing or it's not firing.

  • What we need to find is a way for a person

  • to be either on or off--

  • raising a flag or their hand should do the trick.

  • To illustrate our visual input,

  • I will be drawing a number from 0 to 9

  • onto a grid divided into 25 squares, or pixels.

  • Now each person, or neuron, will receive one pixel.

  • If a neuron receives a pixel with writing on it,

  • it will fire.

  • The V1 layer identifies pixels in the retinal layer

  • that form particular lines in the number,

  • and the V2 layer identifies particular combinations

  • of lines from V1 that form angles.

  • - What does V3 do? - V3 is more sensitive to color.

  • We're only working with black and white in this case.

  • So you're saying we won't even need to have a V3 in our brain?

  • We're skipping V3 altogether.

  • All right, sorry, V3.

  • - So we're gonna go straight to V4? - Yeah.

  • Michael: V4 neurons will fire

  • when their assigned combinations of angles

  • have been detected.

  • At this point the basic shape of a number

  • is beginning to take form.

  • And so actually the next one is called IT.

  • - Ooh! - And that stands for infratemporal cortex.

  • Michael: Now don't worry.

  • We haven't forgotten V5 and V6.

  • They exist, they're responsible

  • for higher-level image processing in our brain.

  • But for our demonstration, we don't need them.

  • We do, however, need the infratemporal cortex,

  • which is the final layer needed for visual processing

  • in the brain we're designing.

  • Our IT will consist of ten neurons

  • representing the numbers 0 through 9.

  • They will be looking at neurons in V4,