Uploaded June 2026 | Updated September 2026, 3 hours ago
A growing percentage of online content is AI-generated. This has consequences for science and society, but also for future versions of the machines trained on this content. There is growing concern, for example, of knowledge collapse—a degradation in both diversity and accuracy of information, a kind of “data inbreeding." Prior work has demonstrated single-model collapse when a model is trained on its own output. I will extend beyond a single model in this talk. Inspired by ecology, my co-author and I ask whether AI ecosystem diversity, that is, diversity among models, can mitigate such a collapse. Through a set of experiments, I will show that increased epistemic diversity mitigates collapse, but, interestingly, only up to an optimal level. In the context of AI monoculture and narrowing feedback loops, the results highlight the important role of data and model diversity in these co-evolving, human-AI systems.
Jevin D. West is a professor and associate dean for research in the Information School at the University of Washington. In both his teaching and research, he draws inspiration and examples from across the disciplinary landscape—aligning with the themes of this Absolutely Interdisciplinary conference. He is the co-founder and inaugural director of the Center for an Informed Public. His research applies network methods to a wide range of systems, including the movement of water molecules in stomatal networks, the spread of infectious disease through human contact networks, and passenger flows in airline transportation networks. More recently, he applies computational methods to study the sociology of science, the spread of misinformation, and the impact of generative AI on collective discourse online. He has published more than 100 scholarly articles from computer science to biology, philosophy, law, and sociology. His work has been featured in The New Yorker, The Economist, Washington Post, NPR, BBC, CBC, Nature, Science, and others. He believes strongly in the teaching mission of the university and has co-created open courses, games, and programs devoted to improving critical thinking and data reasoning. He is the co-author of the book, “Calling Bullshit: The Art of Skepticism in a Data-Driven World,” which helps non-experts question numbers, data, and statistics without an advanced degree in data science.
A growing percentage of online content is AI-generated. This has consequences for science and society, but also for future versions of the machines trained on this content. There is growing concern, for example, of knowledge collapse—a degradation in both diversity and accuracy of information, a kind of “data inbreeding." Prior work has demonstrated single-model collapse when a model is trained on its own output. I will extend beyond a single model in this talk. Inspired by ecology, my co-author and I ask whether AI ecosystem diversity, that is, diversity among models, can mitigate such a collapse. Through a set of experiments, I will show that increased epistemic diversity mitigates collapse, but, interestingly, only up to an optimal level. In the context of AI monoculture and narrowing feedback loops, the results highlight the important role of data and model diversity in these co-evolving, human-AI systems.
Jevin D. West is a professor and associate dean for research in the Information School at the University of Washington. In both his teaching and research, he draws inspiration and examples from across the disciplinary landscape—aligning with the themes of this Absolutely Interdisciplinary conference. He is the co-founder and inaugural director of the Center for an Informed Public. His research applies network methods to a wide range of systems, including the movement of water molecules in stomatal networks, the spread of infectious disease through human contact networks, and passenger flows in airline transportation networks. More recently, he applies computational methods to study the sociology of science, the spread of misinformation, and the impact of generative AI on collective discourse online. He has published more than 100 scholarly articles from computer science to biology, philosophy, law, and sociology. His work has been featured in The New Yorker, The Economist, Washington Post, NPR, BBC, CBC, Nature, Science, and others. He believes strongly in the teaching mission of the university and has co-created open courses, games, and programs devoted to improving critical thinking and data reasoning. He is the co-author of the book, “Calling Bullshit: The Art of Skepticism in a Data-Driven World,” which helps non-experts question numbers, data, and statistics without an advanced degree in data science.
