Disentangling Model-Based and Model-Free Moral Learning

To resolve moral dilemmas, people often rely on decision strategies such as cost-benefit reasoning (CBR) or following moral rules. Previous studies show that people learn to increasingly rely on whichever strategy led to better outcomes in the past. Do they learn this by constructing a mental model of what outcomes would result from using either strategy (i.e.,model-based learning) or by assigning value directly to each strategy (i.e., model-free learning)? To answer this question, we adapted the two-step task to a trolley-type dilemma between following moral rules (e.g., obeying authority) versus (e.g., saving a larger group). In each of the 125 trials, participants’ choices led to either a common or a rare transition, which probabilistically led to good versus bad outcomes. Computational modeling and pre-registered analysis of behavioral data provided converging evidence that participants apply both model-based and model-free learning.