Skip to main content
Preprint Credible — Major labs, established NGOs, reputable named-author preprints

AI Assistance Reduces Persistence and Hurts Independent Performance

A series of randomized controlled trials on human-AI interaction (N = 1,222) providing causal evidence for two consequences of AI assistance: reduced persistence and worse unassisted performance. Across tasks including mathematical reasoning and reading comprehension, AI assistance improved short-term performance, but people then performed significantly worse without AI and were more likely to give up, with effects emerging after roughly ten minutes of interaction. The authors attribute the effect to AI conditioning people to expect immediate answers and argue model development should prioritise scaffolding long-term competence.

Publisher

arXiv (Carnegie Mellon University; University of Oxford; Massachusetts Institute of Technology; University of California, Los Angeles); published at COLM 2026

Published

6 Apr 2026

Added

today

Key Findings

  • Randomized controlled trials with 1,222 participants found that AI assistance improved short-term task performance but left people performing significantly worse without AI afterwards.
  • Assisted participants were more likely to give up on subsequent unassisted tasks, a persistence effect the authors call foundational to skill acquisition.
  • The effects appeared after only brief interactions with AI, approximately ten minutes.
  • Tasks included mathematical reasoning and reading comprehension; the proposed mechanism is that AI conditions people to expect immediate answers and denies them the experience of working through challenges.

Methodology Notes

Series of preregistered online RCTs (N = 1,222 total) with assisted versus unassisted conditions and subsequent independent-performance measurement; the AI systems used are not named in the abstract. arXiv v1 2026-04-06, v4 2026-08-05, published as a conference paper at COLM 2026 per the PDF header. Affiliations from the PDF title block. Educational tasks rather than health or companion settings; the payload is the deskilling mechanism.

Authors

Grace Liu, Brian Christian, Tsvetomira Dumbalska, Michiel A. Bakker, Rachit Dubey

Tags

rctpersistencedeskillingcolm-2026cmuoxford

Cite This

APA

Grace Liu et al. (2026). AI Assistance Reduces Persistence and Hurts Independent Performance. arXiv (Carnegie Mellon University; University of Oxford; Massachusetts Institute of Technology; University of California, Los Angeles); published at COLM 2026. https://arxiv.org/abs/2604.04721