Stories
AI Chatbots Risk 'Knowledge Collapse' by Narrowing Information Diversity, Study Warns
M
Moneycontrol•29-09-2026, 11:53
AI Chatbots Risk 'Knowledge Collapse' by Narrowing Information Diversity, Study Warns
- •A new study reveals AI chatbots like ChatGPT provide a significantly narrower range of information compared to traditional web searches, potentially leading to "knowledge collapse."
- •Researchers from the University of Copenhagen tested 27 large language models across 155 topics, generating 1.7 million AI answers with approximately 70 million claims.
- •Even OpenAI's GPT-5, the most diverse model tested, produced information at least 18.7% less varied than Google search results.
- •The problem stems from how AI models are trained, prioritizing frequently occurring patterns and potentially overlooking less common information.
- •While not yet occurring, the study warns of a future risk if AI models are increasingly trained on less diverse AI-generated content, advocating for AI-generated summaries as one gateway, not the sole source, of information.
Key Questions Answered
Knowledge collapse is a potential long-term risk where AI models trained increasingly on content generated by other AI systems produce a progressively narrower and less diverse range of information over time.
Researchers found that every tested large language model provides significantly less diverse information than traditional web search results, leading users to see more uniform claims and perspectives.
Yes, researchers observed that more recent language models produced slightly more diverse answers than older ones, indicating that knowledge collapse has not actually happened yet.
