Determine where human choosers fall relative to the normative strategies

Determine where human choosers fall between the fit-maximising and posterior-maximising strategies in the forced-choice signalling task, and determine whether human choices track either criterion.

Background

The paper defines two idealised reference strategies for the signalling task: selecting the option that best fits the hidden target pair and selecting the option that maximises the posterior probability of the target by discriminating it from distractor hypotheses. A human condition has been designed and piloted for instrument testing but has not yet been collected.

Consequently, the model results do not establish where human participants lie on the fit–posterior–random coordinate system. The authors explicitly state that the human position and whether humans track either normative criterion remain unknown.

References

Condition 8 is designed and piloted for instrument testing only, and not collected. The random and salience references in this paper describe two idealised strategies; we do not yet know where a human chooser falls between them, and the paper's design question, does either population track either criterion, is answered here only for models.

— Consistency Without Alignment: Item-Sensitive Language Models Indistinguishable From Random  (2609.00576 - Huynh, 1 Sep 2026) in Section “Discussion,” paragraph “No human data yet”