When your side quest becomes your main one: choice reinstatement revealing model-based credit assignment

Poster Session B - Sunday, March 30, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Lindsay Rondot¹ (ljrondot@ucdavis.edu), Erie Boorman¹; ¹Center for Mind and Brain, Dept. of Psychology, University of California, Davis, USA

Credit-assignment is the attribution of outcomes to the actions that led to them, enabling goal-directed agents to adapt in complex environments. Most research focuses on single-step contingencies between choices and outcomes. However, real-life goals require multi-step planning, where intermediate actions may not directly cause outcomes but are essential for achieving goals. To address this, we modified the two-step task (Daw, 2011), requiring participants (N=28) to make decisions in an MRI scanner. They selected between two options, each leading to one of two destinations 70% of the time. Then, they chose one of two objects to obtain a token. In 10% of trials, tokens were reallocated, disrupting contingencies and requiring strategy adaptation under model-based decision-making. Behavioral analysis confirmed model-based strategies: participants aimed for the same destination in stable trials and adapted following reallocations. To test our hypothesis concerning model-based credit-assignment following reallocations, we used multivoxel pattern analysis to decode the first stage chosen and unchosen stimuli identities when the token is received. Following a common transition, the chosen stimulus was reinstated in the lateral OFC (t=3.44, p<.005), likely updating contingencies. The unchosen stimulus was reinstated in the lateral FP (t=3.16, p<.005), consistent with prospective memory theory as subject are remembering which stimuli to select in the future. Following rare transitions, the chosen stimulus was reinstated in the medial PFC (t=4.57, p<.005), while the unchosen stimulus was reinstated in the anterior (t=3.57, p<.005) and posterior (t=4.53, p<.005) hippocampus, extending into the amygdala (t=5.19, p<.005). These findings reveal reinstatement mechanisms in model-based learning.

Topic Area: THINKING: Decision making