Behavioral demands organize a decision process into distinct yet coordinated neural representations in parietal cortex

Perceptual decisions are typically modeled as a process where evidence is accumulated until a bound is reached. The lateral intraparietal area (LIP) is believed to implement this computation using a low-dimensional population representation that is focused on a specific action for reporting the choice, which aligns with an intentional framework.

However, this framework suggests that altering the behavioral demands on the report should result in a different representation. This raises doubts about the universality of the low-dimensional decision representation in LIP: is it just a special case of decisions made through a single action, or does it represent a more general computational architecture that can handle multiple behavioral outputs?

To investigate this, researchers trained monkeys to initially report the termination of a motion-discrimination decision by making a saccade to a target that is not relevant to the choice, and then to later report the content of the decision with a saccade to one of two available choice targets. Despite the reports being behaviorally distinct, the timing of the termination decision remained consistently linked to the accumulation of sensory evidence supporting the final choice in both monkeys. This indicates that both reports were still utilizing a shared underlying computation.

However, when using high-density Neuropixels recordings from LIP, the study found that the representation of decision termination and decision content were organized along separate coding directions in the population of neurons, with these groups largely not overlapping. Despite this separation, the two representations were not independent; fluctuations in the population encoding the content of the decision predicted changes in the encoding of the termination decision from trial to trial, with their coupling becoming stronger as the decision progressed.

These findings suggest that a single decision-making process can be adapted into different, action-specific representations, and these representations are coordinated by the selective transfer of information between distinct neural populations.