Description:

Abstract:

For good reason, most systems and computational neuroscience studies focus on understanding the brain mechanisms behind a simple cognitive, perceptual, or motor process. This reductionist approach to behavior makes a lot of sense because it makes the problem tractable, and we hope that what we learn from simple situations generalizes to more complex situations. However, natural behaviors often involve many processes interacting at once. And while few disorders affect simple processes in isolation, the inability to flexibly adapt to the changing conditions that characterize real world behavior is a debilitating symptom of disorders ranging from autism to dementia to substance abuse. The complexity of more naturalistic behaviors means that studying them will require bringing together multiple experimental, computational, and theoretical tools. 

I will present our recent and ongoing work focused on the neural basis of perceptual decision-making amid dynamically changing task demands. In natural environments, visually guided decision-making involves two indispensable processes: 1) identifying the relevant problem to solve (updating task-beliefs for appropriate task-switching), and 2) correctly solving the believed-relevant problem (perceptual decision-making). In laboratory studies, those two processes have been extensively studied separately, by requiring subjects to switch between very easy tasks or by studying perceptual decision-making under static task demands. We studied task-switching and perceptual decision-making together by using a dynamic, two-feature discrimination task that requires subjects to infer the visual feature whose correct discrimination will be rewarded. 

Despite the fact that task switching and perceptual decision-making are thought to depend largely on different brain areas, we demonstrated a competitive link between them: when subjects performed well on the perceptual task, they were slower to notice that the task had changed, and vice versa. To understand the neural reason for this behavioral link, we used a combination of techniques: multi-neuron, multi-area recordings in behaving animals, causal manipulations, hypothesis-driven dimensionality reduction techniques, normative and recurrent network modeling, and human psychophysics. 

As our field shifts toward studying more complex and naturalistic behaviors, this work has implications for methodology, basic science, and clinical applications. Methodologically, this work highlights the value of deeply integrating experimental and computational approaches. Scientifically, we demonstrated inextricable links between two disparate processes, highlighting the need to move (cautiously) away from a reductionist behavioral approach. And clinically, the neuronal link between perception and task switching suggests that potential treatments of disorders that affect decision-making under dynamic conditions might do well to target neurotransmitters and other processes that mediate communication between brain areas.