Conference on Cognitive Computational Neuroscience

To guide effective behavior in the face of uncertainty and change, animals can efficiently use their limited computational resources by adopting widely generalizable behavioral strategies. To gain a holistic understanding of the space of generalizable strategies, it is important to use a diversity of tasks to probe animal behavior. However, we lack principled methods for designing such tasks. Here, we combine enumeration with task design in order to probe the space of strategies that an animal uses to dynamically select among multiple options that can deliver different rewards. We consider a class of strategies in which an animal’s choices depend locally on the outcomes of previous actions. We then design and enumerate a class of tasks that best discriminate between individual strategies. Through this enumeration, we discover a set of highly discriminative tasks that would not have been devised by typical handcrafted methods. We show that these tasks can evoke and identify nearly all possible groupings of strategies, and can thus be combined to iteratively localize decision strategies from behavioral observations. Moving forward, we plan to use these tasks to uncover the cognitive architectures that guide decisions across different species—from flies to humans.