AI-1 Lecture

Welcome back to the AI-1 lecture in the new year.

I hope you had a very relaxing time.

I also hope you didn't forget everything.

But just in case you did, let me give you a little bit of an overview.

So we're looking at agents.

Agents are these things which communicate...

...with the environment.

Percepts coming in, actions acting on the environment.

And something interesting happens in here.

That's the frame in which we're working.

And the interesting thing is...

...different agent programs can live in here.

Typically, the agent programs have some kind of a world representation.

The agent remembers what the world is like...

...or at least has a theory about what the world might be like.

Which might be accurate or not.

You can imagine if that theory is accurate...

...then the agent is more likely to survive or to perform better.

We've looked at various algorithms...

...which can help us decide, given a percept sequence...

...how to act on the world.

We've first looked at search algorithms.

And the search algorithms are essentially given by actions...

...that look like that.

You cannot look into them. They're just blobs of something.

And then actions...

The only thing we know is...

...an action on, say, state S gets me into a state S'.

Which makes the world description very easy.

We just have to remember, oh, I'm in state number 3521.

It also...

...makes the algorithms very simple...

...because we can't really do much with the knowledge...

...that we are in state 3531.

So what we've done, we've graduated to better ways...

...of describing the state of the world.

In the constraint satisfaction problems chapter...

...we've looked at the state of the world.

In the problems chapter...

...we essentially had states that, instead of looking like black boxes...

...had something we had a key, some value, another key, another value and so on.

We called that factored world representations.

The keys there are what we call attributes.

Attributes are basically functions...

...that look at the world and give us a value.

The weather attribute is something that looks at the world and says...

...yuck, winter weather with lots of rain.

It gives us values.

Or the attribute over the participant's attribute...