And we started talking about a class of algorithms

called search algorithms.

And these search algorithms are specifically

designed to solve what we call search problems.

The assumption these algorithms make

is that you can describe the problem at hand, the one you

want the agent to solve, in terms of states.

And determining the set of states

is a design problem that the agent designer has to solve.

We want states that are not overly specific,

because then we get too many of them.

And we want states that are suitable for the agent

to update, to maintain, and all of those things.

And we want to have them fitting to actions,

the second component of a search problem, actions not

overly specific, but specific enough that actually the agent

can actually execute them.

My favorite action is, in Romania,

or one of our main examples, go to Bucharest.

It solves any problem right at once.

It has the little problem that the agent cannot directly

execute it.

So you want to have, and that's also

one thing that, in essence, the agent metaphor gives us,

is it gives us help in designing the search problem.

So search problems consist of states fitting actions,

a transition model that tells us what the actions do

to change the state, a set of initial states,

and a set of goal states.

Five components, and you have to have all of them.

Once you've done that, we can actually

make use of a very simple algorithm, the tree search

algorithm, which is essentially get hold of a node.

That's the recursive step.

Get hold.

You're in a current node.

If that's the goal node, hooray.

If that's not the goal node, you expand it.

You add all the newly seen nodes to the fringe.

Remember, the fringe are all of the unvisited but expanded nodes.

And then you choose a new node according to the strategy.

And the algorithm always stays the same.

The only parameter in this algorithm

is the strategy that governs the choice of the next node

to visit.

And we're talking about strategies.

That's all we do today.

We've already seen the criteria we use to evaluate a strategy.

We're interested whether the strategy is complete

and there are incomplete ones.