The Role of Higher-Order Cognitive Models in Active Learning

Building machines capable of efficiently collaborating with humans has been alongstanding goal in artificial intelligence. Especially in the presence ofuncertainties, optimal cooperation often requires that humans and artificialagents model each other's behavior and use these models to infer underlyinggoals, beliefs or intentions, potentially involving multiple levels ofrecursion. Empirical evidence for such higher-order cognition in human behavioris also provided by previous works in cognitive science, linguistics, androbotics. We advocate for a new paradigm for active learning for human feedbackthat utilises humans as active data sources while accounting for their higherlevels of agency. In particular, we discuss how increasing level of agencyresults in qualitatively different forms of rational communication between anactive learning system and a teacher. Additionally, we provide a practicalexample of active learning using a higher-order cognitive model. This isaccompanied by a computational study that underscores the unique behaviors thatthis model produces.