What have we done?

We've basically looked at uncertainty,

uncertainty of the world, uncertainty of our sensors,

uncertainty about our actions.

And we model the uncertainty by a mechanism called

probabilities.

Actually, not only prior probabilities,

but also conditional probabilities.

Why?

Because our agent learns things as it goes along.

So it has more and more evidence that adjusts

the probabilities further on.

So some initial priors we need to assess.

And some of them we can derive, for instance,

by Bayes' rule or the product rule or other things.

So if we have multiple evidence, then we usually

want to find the probability of the product

and we usually want to exploit conditional independence

because we don't really have a lot of full independence

in our world anyway.

OK.

And so the things we've done in a pedestrian way here,

we're going to do with Bayesian networks next.