Open-Endedness: Can AI Keep Discovering?

An in-person, English-language Tokyo AI research evening on open-endedness: how an AI system can keep producing new ideas, behaviours or solutions instead of converging on what it already knows. The three talks are pitched to be useful for everyone from newcomers to established researchers, each a 25-minute presentation with five minutes of moderated Q&A. Mizuki Oka (Artificial Life Institute, Chiba Institute of Technology) approaches open-endedness through artificial life and environments that evolve for learning agents. Bert Chan (Artificial Life Institute, creator of Lenia) explores Lenia and Flow Lenia through play, including quality diversity and reinforcement learning to control self-organizing patterns. Kai Arulkumaran (Sakana AI) presents a replication of Picbreeder with vision language models in place of human users and compares the results with the human baseline. An hour of networking follows; the Tokyo venue is still to be announced.