A new multi‑university study finds that after just 10–15 minutes of working with an AI assistant, people are less willing to stick with tough problems and do worse once the tool disappears. Participants who leaned on chatbots cruised through questions while the assistant was available, then answered fewer questions correctly and skipped more items when the tech was removed. The pattern has immediate stakes for teachers, campus tech leads, and employers as chatbots weave into classrooms and office workflows.
Study In a Nutshell
The paper combines three randomized experiments that initially enrolled about 1,222 participants to test mathematical reasoning and reading comprehension, and the authors report that effects kicked in after roughly 10–15 minutes of AI use, according to the arXiv preprint. The research team includes scholars affiliated with Carnegie Mellon, the University of Oxford, MIT and UCLA, and the experiments relied on a GPT‑5–style assistant to help with the tasks.
What the Tests Looked Like
Participants worked through short sets of fraction and reading problems with a chat assistant sitting in a sidebar; partway through the session, researchers cut off access and told people to finish on their own, as reported by Futurism. While the AI group moved faster and was more accurate during the assisted stretches, that same group later solved fewer test problems and was more likely to hit "skip," a behavioral red flag for reduced persistence, once the assistant vanished. "They're also not willing to try without AI," UCLA coauthor Rachit Dubey told Futurism.
Not All AI Use Is Equal
The preprint notes that the downstream cost was concentrated among people who used the assistant to grab direct answers, while participants who leaned on the tool mainly for hints or clarification did not show the same drop‑off in performance or grit. In Experiment 2, the researchers report that about 61% of participants in the AI condition said they used the chatbot to obtain answers directly, with smaller shares using it for hints or not using it at all, per the arXiv.
Local Classrooms Are Paying Attention
The study arrives as campuses and school districts are already piloting AI tools and wrestling with rules of engagement. Cal State’s multi‑million‑dollar ChatGPT rollout and other campus pilots have stirred debate over training and oversight, according to Cal State’s ChatGPT rollout, and regional outlets like FOX5 San Diego have pushed out local coverage of the findings this week.
How Researchers Say to Use AI
The authors urge designers to build assistants that scaffold learning instead of spitting out full answers and warn that cumulative, unreflective reliance on AI could lead to effects that are "profound and difficult‑to‑reverse," as the study and follow‑on coverage note. Education experts are calling for stronger AI literacy and clear classroom norms that steer students toward using chatbots for hints and explanations, not as a one‑click answer machine, per reporting in The Independent.
