Clarify the role of optimality in defining probabilistic representations

Clarify and formalize what “optimal” means for a probabilistic representation outside the context of a specific experimental task, and ascertain whether and why optimality should differentiate heuristic representations of uncertainty from probabilistic representations of uncertainty in perceptual inference and decision-making.

Background

The paper critiques the common strategy of inferring probabilistic representations from apparent optimal behavior under an assumed ideal-observer model. The authors argue that such approaches hinge on a notion of optimality that is not well defined beyond narrow task contexts and that may not, in principle, separate probabilistic from heuristic uncertainty encodings.

This conceptual gap motivates their proposed criteria—source invariance and probabilistic transfer—as testable hallmarks of probabilistic representations. Before adopting those, however, the literature still faces an unresolved definitional issue regarding the meaning and diagnostic value of ‘optimality’ for probabilistic representations.