Ten minutes with an AI helper made people give up faster once it was gone
A UC Berkeley-led study found people who briefly used ChatGPT gave up sooner on hard problems once it was removed. Here is what it showed and what it did not.
The short version
- Researchers led by Brian Christian at UC Berkeley ran randomized experiments with 1,222 participants.
- Two fraction-math experiments had 354 and 667 people; a third, SAT reading test had 201.
- ChatGPT was switched off around question 12, and the AI group's accuracy dropped and many quit.
- Those who never had AI persisted longer and did better on the same later questions.
- The work was presented at the Conference on Language Modeling.
Why it matters
Many of us now reach for a chatbot first, yet exams, interviews and live meetings usually take it away. The study suggests that skipping the struggle can cost persistence when you face hard problems alone. It does not show that AI makes people less intelligent, only that unused practice does not build skill.
What the researchers did
As technews.tw reported, a team led by Brian Christian of UC Berkeley's Center for Human-Compatible AI tested how quickly people lean on an AI helper. Researchers from Carnegie Mellon, MIT, Oxford and UCLA also took part. The results were presented at the Conference on Language Modeling.
The team recruited 1,222 people for randomized controlled experiments. That means participants were assigned by chance to different conditions, which makes the comparison fairer.
Help, then sudden removal
In the first two experiments, with 354 and 667 people, participants solved basic fraction problems. One group worked alone. The other could keep ChatGPT open beside them.
The helped group began with high accuracy. Around question 12, the researchers switched the AI off. Their accuracy fell sharply and many stopped answering, while the group that never had AI kept going and succeeded more often.
A third experiment gave 201 people an SAT reading test, to check that this was not just maths anxiety. The pattern repeated: persistence and accuracy both fell once the AI disappeared.
Why the authors think it happens
The researchers argue that people adjust fast to a machine that answers instantly. After that, slow thinking starts to feel inefficient. Using AI also removes what they call productive struggle: the slow, frustrating work that helps the brain grow.
Christian worries about the long-term effect on expertise, including habits good scientists need, such as staying close to the data and working through problems as a team.
Tutor, not oracle
The team's proposed fix is about design. They want AI to act like a tutor that helps learners understand a problem, rather than an oracle that hands over the answer. Christian said the aim should be to augment human ability instead of supplanting it.
We have only a news summary, not the paper. How long the effect lasts and how well it carries beyond these tasks remain unclear from what was reported.
What to watch
- Whether the full paper shows how long the drop in persistence lasts after the AI is removed.
- Whether AI makers build tutor-style modes that guide users instead of giving answers.
Try it
Pick one recurring task this week, such as a reply or a calculation, and do a first pass without a chatbot. Then use the chatbot to check your work.
Sources and links
- Reportingtechnews.tw: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by technews.tw · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.