Reduced Pavlovian value updating alters decision-making in sign-trackers

Successful reward-guided behavior relies not only on learning actions to obtain rewards but also on learning cues that predict the reward, which motivate and prepare the animal to pursue and consume it. Although these two types of learning-instrumental learning and Pavlovian conditioning-have been extensively studied, it remains unclear how the brain updates and arbitrates between these systems, especially when Pavlovian signals are irrelevant to decision making. To address this, we used eye-tracking, pupillometry, and computational modeling in a Pavlovian-to-Instrumental Transfer task with 60 humans (30 females), consisting of three phases: the Pavlovian phase (learning conditioned stimulus-outcome associations), the instrumental phase (learning response-outcome associations), and the transfer phase (testing Pavlovian bias on instrumental responses). Using this approach, we aimed to identify different types of learners and their strategies, especially how individual differences in sign-trackers versus goal-trackers influence Pavlovian bias. To that end, we used eye gaze data to categorize participants as sign- or goal-trackers, and found that although both groups learned the task, sign-trackers’ performance was lower when exposed to Pavlovian cues, as they favored options based on their cue-outcome associations. Fitting data with multiple computational models revealed that participants dynamically arbitrated between values estimated through Pavlovian and instrumental systems. Importantly, lower performance in sign-trackers was due to slower updating of Pavlovian cue values during the transfer phase, not overweighting of Pavlovian cue values relative to instrumental action values. Overall, our study offers a computational framework for understanding inflexible decision making and potential interventions for disorders marked by maladaptive cue reactivity. Significance Statement Everyday decisions are shaped by the feedbacks received after making choices (instrumental learning) or predicted by the presence of cues —like images or sounds - around us (Pavlovian learning). Sometimes, these cues can lead us to make poor choices. Our study examines how these two learning systems interact, and why some individuals are more influenced by such cues than others. Using eye-tracking, pupillometry, and computational modeling, we show that people more drawn to reward-predicting cues update their beliefs more slowly, leading to biased decisions. These findings may offer new insight into why some individuals struggle with inflexible or maladaptive behavior, as seen in conditions like addiction and compulsive disorders.