We looked at the setup.

The setup is that instead of having,

in a way, one agent,

we have two agents.

We call them Max and Min.

Min tries to win against Max,

which means Min wants to do

everything that hurts Max and the other way around.

That's the setup. Typical case is playing chess.

If you look at this guy,

to a chess player, he looks very,

very scary because the chess player knows this guy,

he can beat you no matter what. Brand master.

Okay. So, you have an opponent,

that opponent wants to make you lose.

How do we deal with that in AI?

We can't just use iterative deepening search,

because iterative deepening search is

based on the assumption that the environment

doesn't change while I'm contemplating my actions.

Winning chess without an opponent is easy.

Gets fun if you have one of these opponents.

That directly breaks this assumption

that our environment is static.

Okay. So, how do we deal with the fact that not only is

our environment not static,

but it's actively antagonistic?

You can't say, oh,

you and I, let's play together.

Not going to work. He's not going to fall for it.

I showed you a couple of old examples of chess machines.

So, the dream of playing chess by machine is old.

The success is quite recent, 25 years or so.

We talked about what kind of games we can do.

The answer is essentially,

we're not going to do soccer

in five dimensions or 10 dimensions.

We're not going to do soccer.

We're going to have discrete states.

We have finite number of possible moves.

We have a fully observable deterministic two-player game.

All of that is not soccer.

It's a turn-taking playing game,

which is probably the most,

these two are the most important restrictions.

Going to three, four,

five players would be relatively easy.

But giving up

this turn-taking thing is something that takes a lot more math,

which is exactly what we're going to do next semester.