JUST IN: New research by Daron Acemoglu, Tianyi Lin, Asuman Ozdaglar, and James Siderius suggests that artificial intelligence doesn't always improve how we learn collectively.

The study argues that when a global system updates too rapidly with data influenced by its own responses, it can reinforce existing biases, reduce informational diversity, and worsen long term social knowledge.
The research introduces a theoretical model where an AI learns from the population's beliefs and then returns signals that modify those same beliefs.