MIT Economist Warns of 'Knowledge Collapse' Risk from Agentic AI Systems

Summary

Economist Daren Acemoglu has published research examining how generative AI, particularly agentic AI systems that make autonomous decisions, could fundamentally reshape human learning incentives and society's long-term information ecosystem. The research highlights a critical tension: while AI agents can improve decision-making in the near term, they may simultaneously erode the human learning and knowledge-building activities that sustain collective knowledge over time. According to the study, human learning generates both private signals (personal knowledge) and 'thin' public signals that accumulate into society's general knowledge base—a learning externality that benefits the broader community. Acemoglu's work suggests that if AI systems reduce incentives for humans to engage in costly learning and knowledge production, society risks experiencing what he describes as a 'knowledge collapse,' where the foundations of long-run innovation and societal progress deteriorate despite short-term efficiency gains.

• The erosion of learning incentives poses a systemic risk to humanity's information ecosystem and knowledge-building capacity

Editorial Opinion

Acemoglu's research addresses a crucial but often overlooked consequence of AI adoption: the structural incentives that shape how societies learn and innovate. While much AI discourse focuses on near-term productivity gains, this work raises a profound question about whether we're optimizing for quarters while undermining centuries of accumulated human knowledge-building. Policymakers should seriously consider whether current AI deployment trajectories adequately protect the human learning activities that sustain progress.