This project proposes AI Systems Psychology, an applied sub-discipline that brings the methods psychology developed for characterising and intervening on behaviour to bear on AI systems. It comprises two components. The Psychology of AI systems characterises the AI itself as a behavioural subject by profiling its dispositions, failure modes, and fairness, and treating each evaluation as an instrument whose validity must be defended before its results inform a decision. The Psychology for AI systems protects the people who interact with it by preserving human agency, calibrating trust, and auditing the psychological consequences of deployment. Both are practised across a six-stage lifecycle of AI systems, from Architecting through Retiring, with defined inputs, activities, outputs, and success criteria at each stage. Its theoretical anchor is the AI-IARA framework, which treats human agency as six model-trainable capacities. Its operational outcome is Agency Debt, the population-level shortfall of those capacities against a pre-exposure baseline, measured at discrete waves and reported through three quantities governing bodies can act on. A six-domain competency model specify what a credentialed practitioner brings that a specialist from an adjacent domain cannot. The contribution changes what the field measures, what practitioners do across the lifecycle, and what regulators can enforce.
Zyl et al. (Thu,) studied this question.