Study finds AI assistance for 10 minutes reduces people's persistence and problem-solving ability when the tool is removed

Researchers at Carnegie Mellon, MIT, Oxford, and UCLA conducted three experiments with several hundred people each, asking participants to solve problems including simple fractions and reading comprehension. Some were given access to an AI assistant capable of solving problems autonomously; when the AI was removed, these participants were significantly more likely to give up or provide incorrect answers.

The study suggests that widespread AI use may boost immediate productivity but erode foundational problem-solving skills. Michiel Bakker, an assistant professor at MIT involved with the study, notes that AI can help people perform better in the moment, but proposes that systems should sometimes prioritize a person's learning over solving a problem directly—through scaffolding, coaching, or challenge rather than providing direct answers.

A person's willingness to persist with problem-solving is crucial to acquiring new skills and predicts capacity to learn over time, making the study's findings particularly concerning, according to Bakker. The research suggests that how AI tools are designed—whether they give direct answers versus coaching approaches—may have very different long-term effects on users' abilities.