Economics in Artificial Intelligence

Economy (period from 1776 to the present)

• How should decision making be organized to maximize remuneration?
• How to act in such conditions when others can impede the implementation of the planned actions?
• How to act in such conditions when remuneration can be provided only in the distant future?

Economics as a science originated in 1776, when the Scottish philosopher Adam Smith (1723–1790) published his book An Inquiry into The Study of the Nature and Causes of the Wealth of Nations. An important contribution to the economy was made by ancient Greek scientists and other predecessors of Smith, but only Smith was the first to formalize this area of ​​knowledge as a science, using the idea that any economy can be viewed as consisting of individual agents seeking to maximize their own economic well-being.

Most people think that economics is devoted to studying money circulation, but any economist will respond to this, that in reality he is studying how people make choices that lead to their preferred results. The mathematical interpretation of the concept of “preferred results,” or utility, was first formalized by Leon Valras (1834–1910), clarified by Frank Ramsey, and then refined by John von Neumann and Oscar Morgenstern in the book The Theory of Games and Economic Behavior ).

Decision theory , which combines probability theory and utility theory, provides a formal and complete infrastructure for decision-making (in the field of economics or in another field) under conditions of uncertainty, i.e. In those cases when the environment in which the decision-maker operates can be most adequately represented only by means of probabilistic descriptions.

It is well suited for “large” economic entities, where each agent is not obliged to take into account the actions of other agents as individuals. And in “small” economic entities, the situation is more like a game, since the actions of one player can significantly affect the usefulness of the actions of another (either positively or negatively).

Game theory , developed by von Neumann and Morgen Stern, leads us to an unexpected conclusion that in some games a rational agent must act randomly or, at least, in a way that seems random to rivals.

Economists most often do not seek to find an answer to the third question given above, i.e. they do not try to work out a way to make rational decisions in those conditions when remuneration in response to certain actions is not provided immediately, but becomes the result of several actions performed in a certain sequence.

The study of this topic is devoted to the field of operations research , which arose during the Second World War as a result of efforts that were made in Britain to optimize the operation of radar installations, and later found application in civil society in the development of complex management decisions.

In Richard Bellman's work, a certain class of sequential decision-making problems called Markov Decision Process (MDP) was formalized.

Work in the field of economics and operations research had a great influence on the formulated concept of rational agents, but for many years research in the field of artificial intelligence was carried out in completely different directions. One reason for this was the apparent complexity of the task of developing rational solutions. Nevertheless, Herbert Simon (1916–2001) in some of his earlier works showed that models based on satisfaction (making decisions that are “quite acceptable”) give a better description of actual human behavior, rather than models that involve time-consuming optimal solutions, and became one of the first researchers in the field of artificial intelligence, won the Nobel Prize in Economics (this happened in 1978). In the 1990s, there was a revival of interest in using methods of the theory of solutions for agent systems.