Original subtitles

Game playing.

There's been a lot of advances in game playing. I'm going to tell you a lot about

the classic thing because I think actually this is one of these pivotal

points in AI where we kind of

really realize that maybe we have capacities that we didn't before.

Before we could play checkers. Today, checkers is solved.

We'll talk more about checkers when we talk about game-playing later,

but there was this classic moment in May of 1997. Deep Blue, which is shown here--

Deep Blue looks approximately like a refrigerator, is giant, parallel,

special-purpose hardware

that basically exists to crunch through chess positions

and it beat the world champion Kasparov, and the sequence of moves

was described as intelligent and creative, as opposed to kinda maybe alien

and weird. You could imagine humans being, like, well, that's, yeah I guess that those

moves are legal but weird right. You can imagine, you kind of play chess and

all the pieces scatter to the corner and they come back and you've lost and you're like

what the heck just happened.

But no it's just like, yeah, that's a good move, you know. That's better than the human.

And you can interpret it,

and that's amazing!

This thing's crunching two hundred million board positions per second. It was amazing then.

You can now do that with a PC cluster so things have advanced obviously.

But the amazing thing is, it's doing

nothing like with the humans are doing when they play chess. But the resulting

behavior is surprisingly similar, and this is a case of there are a lot of

different computations you can do, but there is good kind of rational

behavior and multiple ways to get there. We don't always know how humans get there.

But we do have ways of making machines do it in some way.

So, there's this open question: How do humans play chess?

There's lots of things you just can't do. When you get into like a multiplication race with the computer,

like here's some five digit numbers, go! Like you know you're going to lose.

There's just things where you know the computer is better.

Why isn't the computer just so much better at chess that we lost this

battle in the eighties? It's an open question how humans are so good. I mean,

they lose, but it's like, they try, its kinda amazing how humans

can compete with computers at all.

Okay, in 1996 actually, Kasparov beat Deep Blue and had some

really nice words to say. Kasparov said, "I could feel--I could smell--a new kind

of intelligence across the table".

I don't really know what that means like some new

intelligence scent you can hang in your car,

but Kasparov is very positive. And then in 1997,

"Deep Blue hasn't proven anything."

So...

yeah, I guess that's more a comment about humans.

There have been huge advances in game playing recently. For example, in Go.

A couple years ago when I would talk about Go I would say, Go is a case where

humans are so much better than machines,

that the human masters will not even play the machines because it's insulting.

Now the machines are pretty good, and able to play extremely competitive people,

and win.

and that's been very recent. We'll talk about how some of those advances work much later.