Overview
- Researchers led by Brian Christian presented a paper at the Conference on Language Modeling on Monday showing that mere minutes of ChatGPT assistance reduced users’ persistence and independent performance when the tool was removed.
- The finding comes from controlled online experiments with 1,222 adult participants who solved fraction problems and SAT-style reading items while researchers removed AI access mid-task to measure changes.
- Participants who used ChatGPT initially solved more problems, but the group’s solve rate and willingness to continue fell sharply after the chatbot was taken away, while the no-AI group improved.
- Authors warn this short-term effect could accumulate into long-term harms to motivation and learning if sustained, and they say more classroom and developmental studies are needed before making broad policy judgments.
- Policymakers are already reacting: New York City restricted AI for roughly 600,000 pre-K–8 students in September, and experts suggest responses such as AI literacy, trust-focused rules, and sandboxed classroom tools rather than default reliance on commercial chatbots.