How early examples might shape AI agents' interactions

The speaker announced “Filling the Multi-Agent Void” for October 19, 2026, at Toronto’s Fields Institute. The abstract argues that current training underspecifies how AI agents behave together. It proposes examples of agents collaborating with humans and one another in ways people endorse, which could shape the next generation’s training prior. The talk was set to outline a mathematical model and describe a small ongoing experiment.