Generating behavioral acts of predetermined apparent complexity

TL;DR

This paper introduces a new quantitative measure of behavioral act complexity based on Kolmogorov complexity, along with an algorithm to generate acts of specified complexity for evaluating artificial agents.

Contribution

It presents a novel apparent complexity measure for behavioral acts and an algorithm to generate acts with predetermined complexity levels.

Findings

The complexity measure is based on signal readings of percepts and actions.

The algorithm can generate behavioral acts of specific apparent complexity.

This approach aids in evaluating and developing artificial agents' learning abilities.

Abstract

Behavior of natural and artificial agents consists of behavioral episodes or acts. This study introduces a quantitative measure of behavioral acts -- their apparent complexity. The measure is based on the concept of the Kolmogorov complexity. It is an apparent measure because it is determined solely by the readings of the signals that directly encode percepts and actions during behavior. The article describes an algorithm of generating behavioral acts of predetermined apparent complexity. Such acts can be used to evaluate and develop learning abilities of artificial agents.