MIT study warns heavy chatbot use may blunt misinformation-detection skills

MIT tracked 67 participants over four weeks to test how AI assistants (like Claude and ChatGPT) affect judging real vs fake headlines and images. AI improved correct detections by about 21%, but heavy reliance reduced unassisted judgment, with a roughly 15% drop by week four, suggesting short-term gains can undermine long-term critical thinking. The study finds guided, probing prompts support thinking more than prescriptive ones; limitations include a US/UK sample and a four-week window. Educators and the public should balance AI aid with training independent discernment.