#31 What they Don’t Teach You about AI in Business School

But the items, the four key areas that we address in the book in the economics of AI is firstly, we talk about the microeconomics of AI, and this relates to the theory of decision making.

So what economics and AI actually have very much in common is that the artificial intelligence scientists have taken the decision theory from economics.

Because they ask themselves, if you have to build an autonomous agent, an artificial intelligence agent, that agent has to make independent decisions by definition.

That's what an autonomous agent means.

It takes its own decisions.

Without humans watching the AI, the AI will have to take decisions, the system.

So what kind of decision theory do you want it to use? And clearly everybody said we want it to be rational.

quite reasonable.

You don't want an irrational agent.

Who wants to have an irrational agent? Right.

So everybody very quickly said we want a rational decision making.

And then, of course, this is only one theory.

And this is the theory of expected utility theory that economists have been using in microeconomics for a very long time to create what is called the homo economicus.

I struggle over the word sometimes.

It is Latin word.

It's kind of like always there's sometimes this caricature of economists creating this world of these rational decision-making agents who live in this perfect world with perfect information and they make all these perfectly rational, optimized decisions taking the cost and the benefits into account and maximizing their utility, their objective function as a form of decision-making.

But this theory of decision-making is very powerful and it's been taken over by computer scientists and modern AI is built on this basis of expected utility theory and an objective function.

So all AI systems are being taught and trained to maximize some objective function or its converse, minimize some loss function.

So if you want to predict, for instance, whether a particular picture is a picture of a cat or a picture of a lion, the AI system will be trained to minimize the loss of this prediction.

In other words, to make as few mistakes as possible.

And that's the general underlying microeconomics of AI, to put it very simply.

We go into the greater details in our book.

And we also discuss some of the limitations and some of the current shortcomings of the approach.

And we also talk about how this approach is operationalized in practice.

to get artificial agents and systems to learn from what they have done, especially when they are in a multi-agent environment.