Seminar, Hosted by Relatorium Seminar
Filling the Multi-Agent Void
Cornell University
Abstract
When AI agents meet and interact, their behaviour is currently underspecified by training. There is a window of opportunity now to seed this "multi-agent void" with examples of AI agents collaborating, with humans and with each other, in ways that humans endorse. Such early examples of agent interactions can influence the training prior for the next generation, in the same way that early examples of the "helpful assistant" persona shaped the prior for today's chatbots. The talk outlines a mathematical model of self-reinforcing priors and describes a small ongoing experiment.