Behavioral direction-selection and information bounds

Extend the behavioral human-AI cascade model to establish direction-selection, information-bound, and spurious-confidence theorems for heterogeneous or non-unit behavioral signal weights.

Background

The paper derives a behavioral threshold for the first-user cascade, showing when a user’s perceived AI weight is at least as large as the perceived weight of her private impression. However, it explicitly limits this result to the first-user threshold and its absorbing consequence.

For downstream behavior, heterogeneous or non-unit weights alter the inference that later users make about earlier judgments. Consequently, the clean Bayesian increment structure and the two-state reduction used for the paper’s direction-selection, information-bound, and spurious-confidence results may fail. Extending those results to a behavioral population is therefore identified as an unresolved theoretical problem.