Quantitative model of action choice based on the concept of flux

Abstract

Despite decades of research into individual human decision-making by researchers across various disciplines, there exists no clear leader among the relevant computational methods. Moreover, the most popular theories often fail to address both explanatory and predictive challenges. This article describes an algorithm for rational action selection based on the concept of flux, designed for implementation in AI agents. The proposed algorithms detail the central elements of the previously published three-system decision-making framework. The selection algorithm mimics the human decision-making process and enables sequential selection of actions, taking into account both external and internal circumstances. Emotions can interfere with rational decision-making through perceptual processes and subjective assessments. Action selection is based on the calculation of current (local) goals, which correspond to the most important global goals (linked to life values). The importance of both local and global goals dynamically changes over time, reflecting the state of the decision-maker. In turn, for a selected local (i.e., essentially instrumental) goal, a person selects the action that brings him closest to the goal, given current constraints and circumstances. The action thus selected is then executed. Thanks to a unified representation, goals and actions are easily correlated, dynamically aligning into meaningful chains. The proposed choice model describes the action selection procedure more fully and accurately than alternative models. One important advantage is the easy interpretability of the results. The choice model described in the article, like the entire three-system decision-making concept, requires implementation as a computer program.