arXiv paper warns AI trained on reasoning traces harms thinking

AI models are trained on the finished products or traces of human thought rather than the messy processes of trial and error that create them. The authors argue that this focus on traces hobbles both AI systems and human users whose own critical thinking and manual skills risk eroding through automation dependency. To address this, the authors suggest designing AI systems and educational tools that incorporate genuine processes of friction, revision, and struggle. Open article →