Learning and controlling dynamic systems
We live and act in a messy world. Scientists’ best models of real-world causal processes typically involve not just stochasticity, but real-valued variables, complex functional forms, delays, dose-dependence, and feedback leading to rich and often non-linear emergent dynamics. How do we learn the causal structure of the world, given these difficulties? In this talk, I’ll present work on how people learn causal relationships between continuous variables as they unfold in time. We find that people segment the continuous flow of information into discretized “events”, and use those simpler representations to infer causal relationships. Complementing this work, we’ll also talk about how people control structured dynamic systems, for example to get the system into a desired state. We find that people learn as little as possible to accomplish their goals, rather than representing the full underlying structure of the system. Together, these findings suggest that people are limited but capable learners, using simple strategies to manage the difficulties of learning structure in complex settings.