That was complexity analysis.

We also have some agents to get through.

I don't think anybody of you already knows much about agents, right, as a framework?

Some head shaking.

Well, that doesn't work.

We'll just have to do it like this.

If you ask yourself, what is artificial intelligence?

There's four very popular answers that people tend to give.

The four popular answers are a system that maybe thinks like a human, a system that acts

like a human, a system that thinks rationally for whatever that means.

We'll get to that in a second, and a system that acts rationally.

These are of course systematically linked.

Especially interesting is I think, or what I think is especially interesting is that

there's a deep divide here between what humans do and what is actually rational, because

we tend to be not so rational at times.

Rationality, as we define it here, is performance oriented rather than based on imitation.

This means that a rational agent doesn't necessarily just imitate a human, but is actually measured

by some performance analysis.

It has some idea of what it wants to do, and then we can in a systematic way measure how

well it achieved that.

To each of these four possibilities of defining AI, there's... Oh well.

If you want to test whether a system acts humanly, there's the Turing test.

We'll talk about that in a second.

You can also build, for example, robots that fly so much like little pigeons that they

can fool pigeons and pretend to be part of a flock or something.

That's usually not really reproducible.

Also we like to throw math at things, and throwing math at pigeons is really hard.

We usually try to find something with a mathematical model more.

There's of course cognitive science.

How do humans think?

How does the human brain work?

How do neurons interact with each other?

That is also very, very complex to capture mathematically.

What we usually tend to do is concentrate on thinking and acting rationally.

That includes logic.

That includes knowledge inference, knowledge management, and it's the basics for finding

the right thing to do or making good choices for whatever that means.

I'm pretty sure many of you have heard of the Turing test before.

It was introduced by Alan Turing in, I think, 1950.

It shifted the questions from can machines think, which was then and is now an always

relevant question, to can machines behave intelligently?

The Turing test is basically a game between a human operator.

That is, via text-only interface, talks to either a human or a computer.

The goal of the computer is to fool the human operator into thinking that the human operator

is in fact talking with another human.

The idea would be that if the computer can believably pose as a human, then we can say

that it is acting intelligently or can think, if we so wish.

Turing predicted himself that by the year 2000, the machine might have a 30% chance

of fooling a lay person, so not necessarily an AI researcher who knows all the clever

questions, but just anybody on the street for five minutes.