Human visual exploration reduces uncertainty about the sensed world.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Psychology

Background:

• Active inference provides a normative framework for scene construction and visual search.
• Previous work established active inference for categorizing visual scenes based on content.
• This study applies active inference to understand human visual search behavior.

Purpose of the Study:

• To investigate evidence for epistemic foraging in human visual search.
• To model how prior expectations change with experience during visual search.
• To link individual differences in visual search to Bayesian belief updating and heuristic use.

Main Methods:

• Bayesian model comparison of Markov decision process (MDP) models for scan-paths.
• Comparing models with and without epistemic imperatives for action selection.
• Analyzing changes in prior expectations across visual search blocks and inter-subject variability using canonical correlation analysis.

Main Results:

• Substantial evidence for epistemic foraging in visual exploration, even in simple scenes.
• Heuristic policies, like left-to-right scanning, are necessary to fully explain observed behavior.
• Implicit prior beliefs about search speed and accuracy systematically change with experience.
• Better scene categorization correlates with reduced reliance on heuristics and increased use of generative scene models.

Conclusions:

• Active inference successfully explains human visual search, including epistemic foraging.
• Visual search behavior is a balance between goal-directed exploration and heuristic strategies.
• Experience dynamically updates prior beliefs, influencing visual search strategies.
• Individual differences in performance are linked to the balance between heuristic use and generative model-based exploration.