You've started by looking at the Wumpus example.

Very simple example.

We'll use it for a variety of purposes and it's been tailored to show all the necessary

problems we'll solve.

We have a cave, we can't see everything, the cave has certain dangers like pits and

wumpuses and you have a reward in the form of gold and there are certain indications

to the dangers, namely the wumpus stinks, the pit has a breeze in some neighboring fields.

Very simple, you can already see it's deterministic, it's partially observable of course and so

on.

It's a very simple world but it gives us a very good basis for looking at descriptions.

If I gave you a phone and you would phone home to your parents or little brother or

so and I told you please describe the wumpus world to your little brother, you wouldn't

have any problem doing that in German or English or whatever language you want to read or write.

And that's what we want to do, description level reasoning aka inference.

So you looked at this problem and to just bind it back to the agent, we're talking about

world models and sensing.

The agent comes into the cave, can only see, because it's dark, their own cell and then

makes a world model that has more information than they have sensed.

The agent is in 1-1, there's no breeze, there's no stench, so the agent knows that this one

must be okay to go and this one must also be okay to go.

Because if the wumpus was here, it would stink, if the pit was here, there would be a breeze

here.

Okay, so that's the important thing about world models.

There's certain things you sense and there's certain things you derive from the information

you sense that is more than you could have sensed.

The agent has no way of sensing there's no pit in here.

The agent can only feel the breeze.

But by the world knowledge, which is part of our description, you can find out that

this is okay.

And you've played through the whole thing, but the important thing is that the world

model has more information than just what you get by sensing.

And you can imagine that a good agent can not only do wumpus, but arbitrarily many other

things as well.

Cook a good meal, study computer science, all those kind of things.

That's what a good agent does.

So you look at the difference between studying computer science and the wumpus world, it's

really only the problem description that differs.

Which is why we're going to study problem description based inference.

It fits very well into the stateful agents.

You have sensing and you have a world model that has a state, in this case the state described

by our description in our description language and we're going to look at at least two.

And with those you can do reasoning about what the world is really about.

And if you want to really write it down in a program, that's what you will get.

You'll basically have the notion of a world state which is a knowledge base.

A long list of things in our description language that we believe to be true about the world.

And then of course thinking is reasoning about the knowledge using a representation language,

a description language which we will call logic.

Now many of you will already know a little bit about logic.

And most of which I'm going to say is going to be consistent with what you already know.

There might be some translation involved.