Two Distinct Attentional Priorities Guide Exploratory and Exploitative Gaze in Parallel
Gaze is directed to visual objects that are informative, reward-predictive, or novel. These gaze preferences may reflect the parallel influence of two separable attention systems: Exploratory attention prioritizing uncertainty and exploitative attention prioritizing learned information about reward. We tested this hypothesis in nonhuman primates learning feature-based attention to objects that had either previously learned reward associations or were novel. The reward history of features slowed down learning by attracting fixations of non-rewarded distractors that were previously targets. This reward history bias persisted in fixations used to choose objects even after choice accuracy stabilized. In contrast, fixational sampling that preceded a choice showed negligible history biases that were overcome quickly in favor of wider exploratory sampling. Quantifying the exploratory value object features with a Parallel Belief States model of attention confirmed that exploratory fixational sampling was unaffected by reward history, while exploitative fixations that committed to a decision showed persistent target history biases. These findings suggest that gaze is guided by two separable attentional priorities in parallel. Exploratory attention prioritizes uncertain items and instantiates information sampling, while exploitative attentional priority guides gaze to current and previously goal-relevant features.