Active inference and artificial reasoning

2025-12-24

AI Summary

This study addresses model uncertainty in structure learning: how to efficiently identify the underlying world model with minimal observational data. We propose an active inference framework that guides agents to select actions maximizing information gain by minimizing expected free energy. Crucially, we integrate Bayesian model reduction into the active inference pipeline for the first time, enabling rapid posterior updates over models and facilitating “artificial insight”—abrupt, discontinuous cognitive transitions akin to human顿悟. Information gain is quantified via KL divergence between prior and posterior model distributions. We validate the approach in a partially observable, discrete-state setting—the three-ball paradigm—demonstrating substantial improvements in sample efficiency and successfully replicating synthetic insight dynamics. Our core contribution is the first artificial reasoning mechanism that supports both sample-efficient and cognitively plausible structure learning, bridging computational principles with human-like heuristic inference.

Technology Category

• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Reasoning under Uncertainty: Probabilistic Inference
• Machine Learning: Active Learning

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• User Modeling, Personalization and Recommendation: Psychology-informed user models and recommender systems

Abstract

This technical note considers the sampling of outcomes that provide the greatest amount of information about the structure of underlying world models. This generalisation furnishes a principled approach to structure learning under a plausible set of generative models or hypotheses. In active inference, policies - i.e., combinations of actions - are selected based on their expected free energy, which comprises expected information gain and value. Information gain corresponds to the KL divergence between predictive posteriors with, and without, the consequences of action. Posteriors over models can be evaluated quickly and efficiently using Bayesian Model Reduction, based upon accumulated posterior beliefs about model parameters. The ensuing information gain can then be used to select actions that disambiguate among alternative models, in the spirit of optimal experimental design. We illustrate this kind of active selection or reasoning using partially observed discrete models; namely, a 'three-ball' paradigm used previously to describe artificial insight and 'aha moments' via (synthetic) introspection or sleep. We focus on the sample efficiency afforded by seeking outcomes that resolve the greatest uncertainty about the world model, under which outcomes are generated.

Problem

• Optimizes sampling to maximize information about world models
• Selects actions based on expected free energy for structure learning
• Uses Bayesian Model Reduction to efficiently disambiguate alternative models

Innovation

• Active inference selects policies via expected free energy
• Bayesian Model Reduction efficiently computes posterior beliefs
• Optimal experimental design disambiguates models using information gain