Okay, so less than one more minute to the quiz.

We've given the process a higher, no, a lower priority.

So we hope that the problems we've been seeing have been fixed.

So another test is in order.

Okay, the twists should start now.

And indeed we have five students.

Okay, are there any questions before we start off?

It looks like I should probably start the review period at 5.50.

Right, okay, so we were talking about three main ways of representing the environment in an agent.

The first and easiest one is the black box representation, which is essentially we cannot look into the states.

They're just, say, numbers.

And the only thing we know is that an action brings us from this state to that state.

For the algorithms, that's the easiest one, but you can already imagine that it's probably the least efficient one.

So we're going to, over the course of this semester, going to graduate from black box states via what we call factored state,

where we have a set of features which are basically values for a certain properties.

Right, for instance, how old are you?

Well, I'm 18.

So age is the feature, 18 is the feature value, and that's usually what databases do.

They represent objects by a set of predefined features.

What the features are is given in the schema, and we have these database records.

Typical factored representation.

It's much better than a black box representation.

You can do quite a lot, but there are always situations, and we looked at the cow being in a driveway and blocking a truck and so on,

which is not what you can actually foresee in a database schema,

and then it might be good to have a more flexible way of representing the environment,

especially if you have more interesting environments where much more can go wrong.

If it's only about keeping grades for students, a database is fine.

If it's about something more interesting, then you might want to have a better representation.

And typically, when things get interesting, we use English or something like that.

If things are relatively boring, we use databases.

And if we have all the times we want for the algorithms, we use atomic states.

And we're going to start with atomic states today.

So that basically concludes our foray into agent land.

We've talked about various types of agents.

And we're going to start with goal-based agents today.

And the area is called general problem solving, which is pretty much a land grab of a name,

almost as much of a land grab as artificial intelligence.

You should probably realize that at the time they were coining these names,

the world computational power, meaning all the supercomputers,

were less than I have on my smartphone today.

So they were kind of hoping that they would find something that is generally useful for problem solving.

And the algorithms, many of those you may have seen, are called search algorithms.

And indeed, they can be used for solving arbitrary problems if you have enough time.

For a large value of if you have enough time.

OK, so remember agents perceive the environment and compute actions.

So agents really always solve the problem of what should I do next, given the percepts that come in.

And so what we are after is algorithms that really help these problems of what to do next.

What to do next might be in chess playing. What's my next move?

What do I do after class? Have dinner or go watch a movie?

All of those kind of things.