Revisiting Rogers' Paradox in the Context of Human-AI Interaction

Humans learn through individual experimentation and social observation, with different strategies carrying distinct costs and success rates. Rogers' Paradox demonstrated that in simple population simulations, cheap social learning provides no fitness advantage over individual learning alone—a counterintuitive result given centuries of human social learning success. As AI systems increasingly serve as sources of social learning while simultaneously learning from humans, we revisit Rogers' Paradox in the context of human-AI interaction. We extend Rogers' original simulations to examine networks where humans and AI systems learn together about an uncertain world. We propose and evaluate learning strategies across three stakeholder levels: individual humans, AI model builders, and society or regulators. Our analysis examines how these strategies impact the quality of society's collective world model, including potential negative feedback loops where learning from AI may hinder humans' individual learning abilities.