Let's see whether please alert me when there's a problem.

Okay, we're going towards

Bayesian networks as a way of modeling the world,

which agents could use and should use.

One of the ingredients,

except for the ones we did last week,

which was normalization and marginalization, essentially.

The chain rule is the use of Bayes' rule,

which as we saw yesterday,

basically gives me a way to switch

the direction of conditional probabilities

if I know the priors involved.

So we can go from the diagnostic direction to

the causal direction and the other way around.

That often has advantages because typically,

as we saw, the causal direction of

this is stable because it really talks about how the world works.

Whereas we very often think of

an agent trying to find out things about the world,

want the diagnostic direction,

and want to use the diagnostic direction.

In all of those cases,

Bayes' rule starts helping us.

We did a couple of extended examples.

We have this meningitis example where you can

use Bayes' rule to get

the probabilities of somebody being ill with meningitis.

We're using the fact that even in epidemic situations,

the causal direction of

the relation between meningitis and the stiff neck actually is stable.