Science & Tech · AI · 1 hr ago
Brief AI use was linked to lower persistence on later tasks in study
A study of 1,222 online participants tested how using AI affected people’s ability to work on difficult tasks.
In several experiments, participants used an AI tool for part of a task, then had to continue without it.
The tests included fraction problems and a reading comprehension question.
After AI access was removed, those who had used it were less accurate or more likely to give up than participants who had not used it.
The researchers say quick answers from AI may make independent thinking feel slower and reduce the useful effort involved in working through a challenge.
The findings raise questions about whether relying on AI can weaken people’s persistence when they have to work without it.
The study paper has been published online and was presented at a conference last week.
A study of 1,222 participants found that brief AI use was followed by less persistence and weaker performance when people had to work without AI.
The participants took math and reading-comprehension tasks in three randomized experiments.
After AI access was removed, participants who had used it skipped 20% of remaining problems, compared with 11% in the control group.
Their solve rate was 57%, compared with 73% for the control group.
Researchers said the study did not establish how long the effect lasts or whether repeated AI use makes it worse.
- Who
- Researchers studied 1,222 participants. Brian Christian of the University of California, Berkeley, worked on the study.
- What
- The study found that brief AI use was followed by reduced persistence and performance on tasks completed without AI.
- When
- The study was presented at a conference last week, according to the Independent, published on 12 October 2026.
- Where
- The experiments were conducted online.
- Why
- Researchers suggested that fast AI answers may make unaided work feel slower and more demanding, and may reduce “productive struggle.
This story does not have two clearly opposing sides.
We’ve built these systems to be helpful, but are they being helpful to us in a way that’s actually helping us?
We showed, ironically, that they are often helping us in ways that are kind of unhelpful.
Researchers split 354 people into groups, gave them fraction problems, and removed ChatGPT access from the AI group after 12 problems.
Researchers tested another 667 participants and then 201 participants on a reading-comprehension task, again removing AI assistance.
The study was presented at a conference, according to the Independent.
- Participants
- 1,222
- Experiments
- Three randomized experiments
- First experiment
- 354 participants worked on 15 fraction problems
- After AI removal
- AI users skipped 20% of remaining problems; the control group skipped 11%
- Solve rate
- 57% for AI users after removal; 73% for the control group









