| Summary: | A longstanding question at the intersection of comparative psychology, cognitive ethology, and cognitive neuroscience is what cognitive strategies primates use to tackle complex multi-step decisions, and what are the neural underpinnings of the strategies. Traditionally, cognitive experiments come in two broad flavors. In one flavor, sophisticated tasks thought to invoke high-level cognitive strategies are used, but their complexity precludes them from rigorous quantitative modeling, leading to mixed interpretations. In another flavor, very simple tasks are used, which have afforded detailed characterization of behavior and the underlying neurobiology, but are limited in eliciting high-level cognitive strategies. In this thesis, I capitalize on both traditions. In the first chapter, I present a novel multi-step decision-making task that was sufficiently complex to allow for multiple strategies, ranging from basic heuristics to more optimal strategies, but simple enough to accommodate quantitative modeling. I then use a series of human psychophysical experiments to quantitatively show that humans rely on a heuristic hierarchical strategy to solve the task due to attentional constraints, and when uncertain, flexibly revise their decisions in a computationally rational manner. In chapter two, I train two monkeys on the task and find that monkeys also adopt a hierarchical and revision strategy to solve the task, like humans. Monkeys were also able to readily generalize their strategy to novel scenarios and made eye-movements that were indicative of simple forms of counterfactual reasoning. However, it was difficult from behavior alone to test whether monkeys were actually using multiple different strategies to solve the task. To investigate this possibility and the underlying neurobiology of hierarchical and revision strategies, in chapter three we conducted high-density neural recordings from monkeys while they performed the task. Neural recordings revealed that monkeys were indeed not using one strategy to solve the task, but rather showed the initialization and dynamic progression of two distinct cognitive strategies that monkeys adaptively selected for different scenarios. We find that neural population initial conditions and response dynamics were flexibly modulated to implement these distinct decision-making strategy plans. Finally, we use the neurally inferred strategies to build composite psychophysical models that better capture the monkeys’ behavior. These results point to the importance of detailed neural recordings in combination with quantitative behavioral modeling for understanding primate cognition.
|
|---|