Abstract
What is a symbol? In this talk I will argue that the classical perspective on symbols in computer science ignores the crucial role of the interpreter in ascribing meaning. Taking this perspective motivates changes in how we assess symbol use, and where it originates. I will therefore outline a set of graded dimensions of behavior through which engagement with symbols can be demonstrated. I will then suggest that symbol interpretation originates in social interaction. Indeed, numerous recent developments in AI attempt to instill symbolic behaviors through imitation learning. These developments highlight broader themes about the differences between passive imitation and active observation, both in training and evaluation. I will close by highlighting how some of these themes may relate to studying whale communication with machine learning; particularly on the need for interactive assessment.