Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster B in Poster Session B: Tuesday, August 4, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Perceived partner complexity affects agency attribution and asymmetries in reinforcement learning

Sepehr Razavi¹, Hayley M. Dorfman², Reuth Mirsky³, Joseph M Barnby⁴; ¹University of Oxford, ²Harvard University, ³Tufts University, ⁴King's College London

Presenter: Sepehr Razavi

Successful behaviour depends not only on learning which actions yield favourable outcomes, but also on inferring who or what caused those outcomes. While prior work has shown that agency attribution modulates reinforcement learning, these accounts have treated the intervening agent as a generic, undifferentiated entity. Here we examine how two fundamental dimensions of agent representation, intent and complexity, shape agency attribution and learning asymmetries across two preregistered studies. In Study 1 (N = 400), participants complete a two-armed bandit paradigm in which a hidden agent (presented as human or AI; sophisticated or unsophisticated; adversarial, benevolent, or neutral) occasionally modifies outcomes, while providing trial-by-trial attribution judgments. In Study 2 (N = 400), a transfer design dissociates whether learned attributions are anchored to partner representations or environmental context. We use hierarchical Bayesian and computational modelling to test whether perceived sophistication amplifies learning asymmetries and whether adversarial contexts produce stronger attribution to capable agents.

Topic Area: Decision-Making, Cognitive Control & Event Cognition