Value-guided attention links what we learn to how much we learn

Our comprehension of learning in uncertain situations is enhanced when we consider the alignment between cues, our choices, and the rewards we receive. The process of learning involves identifying pertinent connections between these elements and determining the extent to which these associations should be revised. Previous research has treated these elements as distinct, but our hypothesis posits that they interact in the brain, influencing the speed at which individuals learn.

To test this hypothesis, we designed a probabilistic learning task involving human participants who made predictions based on multiple cues and received reward feedback. We measured participants' gaze patterns and manipulated the saliency of the cues to influence which cues were processed more during decision-making and feedback.

Using computational modeling, we discovered that learning rates were selectively accelerated for cues that supported the chosen option following a reward and for cues that contradicted it when no reward was obtained. This asymmetric learning rate based on the alignment between cues and choices sharpened the differentiation among predictive cues, increased resilience to noise, and improved overall performance.

Furthermore, we found that individuals with more differential attention toward cues supporting or opposing their chosen options exhibited this learning-rate asymmetry. Manipulating the saliency of the cues directly impacted this asymmetry, indicating that attention plays a unifying role in coordinating what we learn with how much we learn. By preserving distinctions among competing cues and integrating multiple learning asymmetries within a common framework, attention serves as a crucial mechanism in the learning process.