Original subtitles

And so you say, what about the brain? I mean we're engineers right. We want to build something.

we want to build something that's intelligent. It'd be a little silly to ignore the fact that we

actually have a working prototype, and we're trying to do this really hard thing. We

should look at the prototype.

And you know human minds are not perfect decision-makers, but by and large they are

actually very good.

And even the cases where humans are famously not rational, there are many cases

there where you can actually say, well they are rational if you think of their

goals in a different way. That's a whole sticky subject, we'lll talk about that

later in the course. But basically human brains are in many ways

better than what we can build.

So let's just reverse engineer them. So it turns out, brains aren't as modular as software, and

they're a whole lot more squishy. And so they're hard to reverse engineer.

And because we can't reverse engineer them we have a very limited understanding of

what brains actually do, we really don't

kind of reverse engineer them and build a kind of

biomimetic brain.

And in particular there is an idea, which, you know, this is one point of view,

but I think it's got a lot of merit to it. People say that brains are to

intelligence as wings are to flight.

That is to say, that

the biggest breakthrough in manned flight--does anybody know what was the

biggest breakthrough in manned flight?

Stop making the wings flap, right. Stop trying to make a really big bird

made out of metal or something, or fabric right. And so this idea of, okay, maybe the

wings can be fixed 'cause we're going to be going faster or we're going to be going higher,

it's going to be rigid. And so when people started thinking, okay, we're going to learn something about

aerodynamics from the proofs of concept, say the birds, but we're not gonna just

slavishly mimic what's in the biology. Same thing with the brains. We're gonna

learn what we can, but we're not gonna just mimic that because we got different constraints

and a limited understanding of what the brain does to begin with.

So, what do we actually get. Well, we do know something from the brain and it boils down

to this, kind of at the very highest level.

We know that in order to make good decisions,

there's really two parts to that.

One way you can make a good decision

is by remembering that in the past you did this thing and it was bad,

so you know not to do that thing again. You know that boils down to memory. That leads into

learning, machine learning, and that's going to be a big part of this course.

The other way you can make a good prediction, is through simulation--having a

model of the world. We think, alright, what would happen if I do this, well then this

would happen, then this would happen--oh, and then that would happen,

that would be bad. And so you can realize that something is a bad decision by thinking

through a chain of consequences. Not actually doing that chain,

but thinking it through in a simulated model of the world

and that's going to occupy the first half of the class, when we think about

what it means to have a model of the world, that's an abstraction of the world,

and what it means to think ahead along a kind of tree of outcomes.