Okay, welcome to the AI lecture.

I apologize for being so late.

I had to come by car, which

is always a bad idea in around eight.

Really crazy today.

So we are currently talking about

agents for game playing

where game means two player turn taking

fully deterministic

fully observable, what else, discrete, finite state games.

So everything that, for instance,

soccer isn't.

Okay?

Think about chess or tic-tac-toe or whatever.

And this is something where a

goal-based agent is clearly what we need.

Right?

Very simple situation.

The goal is to win the game.

We have a state space.

Unfortunately, state spaces are huge.

So we can't really keep up this pretense

of not being able to look into the states.

Okay?

So we're going to kind of keep that as a mental

way of thinking about it.

But we're going to allow ourselves to generate states and actions

as we need them.

Just as a practicality, not as a program, if you want.

And so we've defined

game problems

game search problems

as being kind of two copies of a search problem

that have the same states

but the states you see in front of

both players see in front of them.

But we have kind of max actions and min actions.

Max and min being the two players and max actions

take a max state and give a min state and the other way around.

The classical thing you would do when you are modeling a two-agent turn-taking game situations.

And so

we want to kind of look at the baseline algorithm.

Baseline in the sense that it gives us

a handle on how to let agents, goal-based agents, choose states.

So the idea is we're going to find

an algorithm for max, min being the obvious dual to that.

And we're going to

basically use the utility as something that guides our search.

That's kind of the idea.

And remember, we have an evaluation function.