Bias in the Room, Bias in the Model

Consulting a language model about a case is not consulting a neutral instrument, since the model returns a compressed statistical portrait of how people, clinicians included, have written about minds. This short compendium places the clinician's own cognitive biases (confirmation, availability, halo effect and anchoring) beside the biases of the model, and shows how the two meet in the exchange, where the model tends to take the shape of the hypothesis the clinician already holds. It then describes how the same tool can be used against that pull, through prompts that ask the model to argue against the working hypothesis and to generate differentials, including ones that do not involve pathology. An appendix provides a working bias check for testing how a specific model behaves in practice. The text presents an argument rather than advice.