| Description | Abstract: The growing use of artificial intelligence (AI) raises concerns of knowledge collapse, i.e., a reduction to the most dominant and central set of ideas. Prior work has demonstrated single-model collapse, defined as performance decay in an AI model trained on its own output. Inspired by ecology, we ask whether AI ecosystem diversity, that is, diversity among models, can mitigate such a collapse. We build on the single-model approach but focus on ecosystems of models trained on their collective output. We find that increased epistemic diversity mitigates collapse, but, interestingly, only up to an optimal level. In the context of AI monoculture, our results suggest the need to monitor diversity across AI systems and to develop policies that incentivize more domain- and community-specific models. Jevin West is the co-founder of the new Center for an Informed Public at UW aimed at resisting strategic misinformation, promoting an informed society and strengthening democratic discourse. His research and teaching focus on the impact of data and technology on science and society, with a focus on slowing the spread of misinformation. He is the co-author of the new 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.
washington.zoom.us… |
|---|