AI chatbots may be causing a global 'knowledge collapse' by reducing information diversity

A new study from the University of Copenhagen suggests that AI chatbots could be shrinking human knowledge by providing less diverse information compared to traditional search engines. Researchers tested 27 large language models (LLMs) using 155 topics and 200 prompt variations, resulting in 70 million responses. The findings indicate that even the most advanced models produce answers that are nearly 19% less diverse than those found via Google Search. Experts warn that this trend creates a 'vicious cycle' where popular content becomes dominant, while niche perspectives are pushed to the periphery. This homogenization of information is compared to the downsides of globalization, where local variety is sacrificed for convenience. While researchers note that newer models are showing slight improvements in diversity, they caution that reliance on AI for information could lead to a narrower understanding of the world and potentially lower collective intelligence.