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

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

The effects of reassurance calibration on evidence sampling and commitment in LLM simulations of jumping-to-conclusions bias

Reema Abdulaziz¹; ¹Independent Reseacher

Presenter: Reema Abdulaziz

Jumping-to-conclusions (JTC) is a data-gathering bias characterised by premature decision making, and is implicated in both clinical and everyday decision-making. Although JTC is well studied, interventions modifying JTC behaviour remain limited and are often embedded within broader studies of delusion. The study investigates whether reassurance calibration influences evidence sampling and commitment in JTC-like decision processes. Using a controlled large language model (LLM) simulation, reassurance is manipulated (calibrated vs. miscalibrated) within a sequential evidence-gathering task to measure its effects on commitment timing and evidence requests. Across conditions, calibrated reassurance delays commitment, increases evidence seeking and improves accuracy, while miscalibrated reassurance shows inverse effects. Reassurance calibration provides a potential mechanism for modulating these behaviours central to the bias, offering a testable direction for future reasoning-bias research.

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