|
Abstract
|
Computational complexity is ubiquitous and manifested in numerous routine tasks, including scheduling, route planning and economic decision making. Therefore, understanding how the brain manages computational complexity is vital for building computationally plausible models of cognition and behavior. Based on the classic knapsack problem, we created a nonhuman primate (NHP) task that incentivized NHPs to deliberate and find optimal or satisfactory solutions. NHPs invest time to construct and compare multiple combinations and the time invested reflects the number of computational steps efficient algorithms require to achieve good outcomes (Hong and Stauffer, 2023, NatNeuro). Subsequently, we analyzed one animal’s eye movements during deliberation and confirmed that the search trajectories predict the complexity of the behavioral solutions. The animal spent most of the time examining options with the highest value in trials where they produce low-complexity, greedy-like solutions. On the other hand, the animal spent time fixating on a wide variety of options in trials where the animal’s solutions resembled those from high-complexity, combinatorial search. Furthermore, the time spent on these items positively correlate with their importance in achieving good or optimal solutions. To investigate the underlying neural basis, we record single unit activities in the dorsolateral prefrontal cortex (dlPFC) while the animal performs the task. Neurons in dlPFC encode the complexity of the algorithms the animal employed. The proportion of complexity coding neurons increases as the recording locations moved to more rostral part of the prefrontal cortex, consistent with the idea that more rostral part of prefrontal cortex is responsible for higher-level abstraction and cognitive control. Leveraging eye-tracking data, we also uncover how dlPFC neurons can facilitate mental construction of behavioral solutions by coding individual items as well as item combinations. Overall, we discovered the first behavioral and neural evidence for combinatorial reasoning under computational complexity in a nonhuman animal model.
|