Item type:Thesis, Open Access

Active Inference, Psychology, and Philosophy:An Analytical and Cumulative Study

Loading...
Thumbnail Image

relationships.isAuthorOf

Publisher

Philipps-Universität Marburg

item.page.supervisor-of-thesis

Abstract

Active inference is a prominent theory in contemporary cognitive science and psychology. Its ambitious research program aims to give a unified account of perception, action, cognition, learning, and decision-making. It is also being increasingly applied beyond psychology to a variety of fields, including artificial intelligence, robotics, and artificial life research. This dissertation presents an analytical and critical study of active inference from various theoretical, philosophical, and formal perspectives. The central thesis is that active inference, as it is currently formulated, is inadequate as a psychological theory in two crucial domains of experience— perception and goal-oriented action. In the domain of high-level cognitive experience, it is useful as a modeling framework to formalize our verbal theories, albeit with some limiting conditions. The following three studies aim to defend this thesis. In the first study, I look at active inference’s theory of perception from the bird’s eye view of an empirically adequate natural philosophy. Specifically, I expound Whitehead’s natural philosophy based on his mathematical physics work and draw its implications for perceptual psychology. Apart from the merit of bridging the gap between the ‘world of science’ and the ‘world of perception,’ two results of this study are immediately relevant. The first result concerns rejecting the ‘impoverished stimulus’ assumption for the perception of colors and qualities. The second result concerns rejecting the ‘impoverished stimulus’ assumption for spatial perception. Taken together, these results point to a need for a theory of direct perception of qualities and space-time in contrast to the inferential and indirect perception of active inference. In the second study, I look at active inference’s theory of goal-directed action from the perspective of mathematics of dynamical systems theory. The main result of this study was the inadequacy of active inference’s theory of agency/goals, which grounds its theory of goal-directed action. The theory of agency was found to be lacking because the fixed, pre-specified, ‘bounded set of characteristic states’ necessitated in active inference agents do not capture natural biological agents. The necessity of such fixedness is traced to the employment of self-organizing dynamical systems in their formalism. Owing to the inherent mathematical properties of all dynamical systems, the state spaces are always fixed. In contrast, in real agents, there is a dynamic growth and emergence of novel state spaces. Through these mathematical properties, we argue for the inadequacy of goal-oriented action in the current formulation. In the third study, I employ active inference as a modeling tool to formalize a psychological theory of expectations - ViolEx. Due to its close connection with probability theory and Bayesian reasoning, we found active inference useful as a modeling framework to formalize, implement, and simulate many of the predictors of rational and irrational expectational beliefs. This study illustrates the potential usefulness of active inference when situated within the proper domain of experience: cognitive and probabilistic reasoning, which presupposes the concepts/premises over which correct reasoning is to be done.

Review

Metadata

show more
Raghuveer, Dhanaraaj: Active Inference, Psychology, and Philosophy:An Analytical and Cumulative Study. : Philipps-Universität Marburg 2025-11-26. DOI: https://doi.org/10.17192/z2025.0063.

License

Except where otherwised noted, this item's license is described as Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 - CC BY NC ND