So let me just say

I'm very happy

by the way

you're all engaging with the subject

much better than last semester or last year.

Shouldn't compare AI1 to AI2

but one of the consequences of that is that

progress is slower

so I'll probably have to drop one of the sections somewhere

down the line.

I haven't quite decided what, but I think it's much more important

that you ask questions and you understand then that I get all the topics through.

My hope is that you understand enough of symbolic AI that you can actually read up

on all the things we're missing.

With all the questions on the forum and on the

matrix, I'm pretty confident that you will.

Okay, so new topic.

So far we've been looking at goal-based agents that use

black box state representations.

In a way, states are just numbers or names.

And the way they act is given from the outside.

We have this transition model

that tells us

if we have action A in state S

the result is S'.

It's all from the outside, we cannot look into the states.

And that's a big fat lie because in the last chapter with the game playing

we couldn't keep this up.

We had to kind of, at least for the evaluations and so on,

look into the states

because there are just too many for practical reasons.

But we didn't really use the states

the internal structure of the states for anything.

If you think about alpha

beta or minimax or something like that

that is just tree search.

Even Monte Carlo tree search was just tree search.

We're not using any of the internal information

except to make states as we go along.

In a way

we were just using lazy black box states.

That's all we use the internal structure of these things for.

We're now going to go to factored representations.

Honest to goodness factored representations and adapt the algorithms to that.

Remember factored representations are things where the world states are essentially attribute value pairs.

Where attributes are functions of the states that can have values.

One of the typical things that we can solve with those kind of algorithms is scheduling problems.

Bundesliga

the National Soccer League in Germany

has to have a schedule.