This paper discusses an experimental software project that interactively trains an artificial intelligence program with user input. The project is a commentary on the human-machine relationship inspired by the infamous Turing 'imitation' game. The software is an anthropomorphic play modeled on the different stages of an initiation rite (separation, margin and aggregation) observed by Van Gennep in early African societies. We ask users to interact with the software in order to teach it what is needed to become intelligent. The users reply to a series of questions such as: what kind of things should an intelligent machine have, what kind of things should intelligent machines know about, and so on. By comparing and contrasting the answers given by different participants, we identify the underlying assumptions, desires, and expectations projected onto the attempts to build artificial agency. By pointing out a series of similarities between the birth of agency in human infancy and the processes for building human-like intelligent machinery, we comment on how the perception of intelligent machines and newborns are configured by the established norms and values in society.
Subjects: 5. Common Sense Reasoning; 12. Machine Learning and Discovery
Submitted: May 17, 2006