2016: Erik Stayton

Antikythera is a game about an analog AI built by the ancient Greeks, which puts the player in the position of interacting with this mysterious device. The player must solve the device's puzzles by teaching it about its world of interfaces, lists, and dials. The game is built on top of the ANTIK system, an adaptable natural-language system for building artificially-intelligent agents, which makes possible the interaction mode: teaching via reinforcement learning. Together, Antikythera and ANTIK make up a research tool for the experimental anthropology of human-machine interactions. The system provides a context in which users engage with an AI by typing commands, and build a shared language for referring to desired actions. Rather than treating language as a fixed system of statements, each having a particular "meaning" that can be determined unambiguously, we treat language as a negotiated system for producing mutually agreed-upon behaviors. We are interested in examining how this negotiation—as opposed to a more traditional interface in which the meanings of statements are fixed—shapes users' behaviors. How do users try to communicate their intent? What languages do they build? And how do they perceive the capabilities of the AI system? [Project Report]


1 2 3