Right, good.

So let's restart.

We are progressing into the technical part of the course.

And we're looking at a class of algorithms

called tree search algorithms that

work on any kind of a problem.

Any kind of a problem is really good,

because we want to have one algorithm that

does all of intelligence, really.

It works on all kinds of problems

as long as we can formulate them in something like this.

The basic tenets of this is that we are in a very simple

environment.

It's deterministic.

It's fully observable.

It's static and all of those easy things.

And that we have a goal-based agent.

We know where we want to go.

And that really directly translates into we have black

box states.

We have actions that get us from one state to the other.

We have goal states.

That's what we need for a goal-based agent.

And we have initial states, which is something we can

deserve.

And we do offline problem solving here.

All of that directly condenses in an architecture like this.

It really condenses into a mathematical structure.

And while the ideas are primal here, the implementation into

mathematics gives you the necessary level of detail so

that you can actually start implementing.

And that's really what I tried to convince you of.

That there's a very natural progression of these things.

We start out with this idea of having a class of tree search

algorithms that we can then mathematicize that and then

talk about implementations.

We talked about tree search algorithms.

We had a very simple first idea of how that might go.

And then using the vocab we fixed in the mathematical

treatment of this, we could make clear some differences

between the state space as a graph and the search space as a

tree.

Those are different animals.

It's basically because graphs are weird and wonderful and

complicated and trees are weird and wonderful and simpler.

Trees are kind of the natural structures.

They have the disadvantages that as we have it here, that

certain nodes, no, certain states might

appear more than once.

That's almost always a bad idea.