"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,

and will only fire when their corresponding neurons fire.

For example, if one or multiple V4 neurons

representing the shapes of an 8 fire,

the IT neuron representing the 8 will also fire.

Voila! We just recognized a number.

So what happens after the infratemporal layer?