Ten Minutes with AI Impairs Problem-Solving After It's Withdrawn

A multi-institutional study spanning Carnegie Mellon, MIT, Oxford, and UCLA tested several hundred participants across three experiments involving mathematics and reading comprehension. When AI assistance was suddenly removed, participants proved significantly more likely to abandon problems or answer incorrectly. The effect persisted across different task domains, suggesting a consistent cognitive impact rather than domain-specific interference. The research points to a fundamental tension in AI deployment: immediate productivity gains may come at the cost of sustained problem-solving capability.