We're finally at the place where most AI courses start.

So you don't have to have agents.

You can also just say, oh, when we do AI, let's just do breadth first search.

And that is this kind of an algorithm.

I don't believe that this is a good idea.

What I would like you to do is everything you heard until now, think of that as a context

to the very concrete algorithms we're going to look at next.

Whenever you have an algorithm that is P-POP whole search or whatever, think it as working

inside an agent.

And how that relates to the problem I'm going to hand wave about.

Very often we will look at the mathematics of the algorithm, what it does, what the data

structures are, and all of those kind of things.

And express that in mathematical terms.

And then I'll tell you stories about, ooh, imagine you're playing chess.

Imagine you're playing chess is something that is really agent level.

And then we kind of open the agent out and look at one of these little bubbles saying

how the world will evolve and that will be the algorithm we are actually studying.

So agents will be combinations of these kind of algorithms that we're going to study plus

a lot of algorithms we're not going to study.

For instance, sensor, so vision algorithms, not something we're going to look at.

Say you're a RoboCup football agent, then you want to know where the ball is and you

want to know where the goal is so you can actually kick it in the right direction or

something like this.

For all of that you need to do a lot of processing with the sensors.

And then you will have to have some kind of a deliberative or planning level that tells

you, well, if the ball is here, the opponent is there, the goal is directly behind it,

you better actually take a detour and don't run through your opponent or something like

this.

And we're going to look at some of the algorithms.

And I would like you to, when I forget, I'll try to kind of bind it back, but I would like

you to imagine where those algorithms live in the agents.

Good.

So, we're going to look at a class of algorithms that work quite generally for tasks which

we call problem-solving tasks.

And we've done a couple of assumptions about these tasks, namely that we have a fully observable,

static, discrete, deterministic environment which allows us to state the problem in terms

of something we call states and those will be atomic states.

We cannot look into them.

I think just state one to state one billion something.

And a set of actions, a set of actions that bring the world from one state into another

state.

There are many things we can formulate in this way.

And we can run the algorithms, we're going to search algorithms, we're going to look

at today and in the next weeks.

And then they give us a solution which is usually an action sequence.

Do this, then that, then that, then that, then that, then that, and then you're in Nuremberg.

If you have the problem of planning or the problem of determining a plan for going from here to

Nuremberg-Harpbanoff.

And you have a couple of states of the world.

There's the state Miko in Herzal 10.