Policies or knowledge: priors differ between a perceptual and sensorimotor task.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Decision Science

Background:

• The brain's representation of probability distributions is debated: is it abstract knowledge or task-specific policies?
• Bayesian models often assume generalizable knowledge, but task specificity may influence learned representations.

Purpose of the Study:

• To investigate whether learned spatial priors generalize across different tasks (sensorimotor vs. perceptual).
• To determine if learned representations function as abstract knowledge or task-specific policies.

Main Methods:

• Participants performed a sensorimotor estimation task and a two-alternative-forced choice (2-AFC) perceptual comparison task.
• Model and simulation-based analyses were used to measure and compare learned prior distributions across tasks.

Main Results:

• Participants learned a spatial prior distribution in the sensorimotor task.
• Measured priors in the 2-AFC perceptual task were consistently broader than those in the sensorimotor task.
• The lack of full generalization suggests sensorimotor priors function more like policies than abstract knowledge.

Conclusions:

• Learned priors do not fully generalize from sensorimotor to perceptual tasks, indicating task specificity.
• The modality of decision-making (movement vs. perception) significantly influences implied prior distributions.
• Findings challenge the assumption of purely generalizable knowledge in standard Bayesian models of cognition.