All research
Technology & tools / Exploratory

AI changes where critical thinking happens

Knowledge workers reported that AI shifts critical thinking toward verification, integration, and task stewardship; greater confidence in AI predicted less reported critical scrutiny.

2025Peer-reviewed survey of knowledge workers319 knowledge workers · 936 work examples5 min read
Read this with care

This is self-reported survey evidence, not a causal test of whether AI use reduces cognitive ability.

Three takeaways

What the paper gives us.

  1. 01

    The work of thinking moves

    Participants described less effort in some traditional forms of critique and more work in verification, response integration, and deciding what the task requires.

  2. 02

    Confidence changes scrutiny

    When users were more confident in the AI, they reported doing less critical thinking. The useful design question is how to preserve appropriate skepticism without making tools unusable.

  3. 03

    Do not claim brain decline

    The study cannot show that AI causes a lasting decline in critical-thinking ability. It describes reported work practices and points to a design challenge worth monitoring.

Study snapshot

What was actually studied.

Design
Peer-reviewed survey of knowledge workers
Sample
319 knowledge workers · 936 work examples
Population
Knowledge workers who used generative AI at work at least weekly.
Comparison
Task- and user-level differences in confidence and critical-thinking effort.
Outcomes
Reported critical-thinking practices in AI-assisted work.
Effect
Higher confidence in AI was associated with less reported critical thinking; higher self-confidence with more.
VSX interpretation

What this may mean for work.

Use AI as a partner in a visible thinking loop: ask for options, verify important claims, state your own decision criteria, and retain responsibility for the final judgment. The goal is not to avoid offloading; it is to offload without becoming passive.

This is VSX's practical inference, not a claim directly established by the study.

Try it carefully

Keep one human checkpoint

Before accepting an AI output, write one criterion it must satisfy and perform one independent check appropriate to the stakes. Save a one-sentence rationale for the final choice. Notice whether the checkpoint catches errors, clarifies the decision, or becomes needless ritual.

Duration: Five AI-assisted tasksObserve: Verification actions and decision rationale
Limitations

Where the claim stops.

  • Participants self-reported effort and examples rather than completing controlled performance tasks.
  • The design is observational and cannot establish causality.
  • Findings may not transfer to people who rarely use AI or to every kind of knowledge work.
Source

Read the original research.

Hao-Ping Lee, Advait Sarkar, Lev Tankelevitch, Ian Drosos, Sean Rintel, Richard Banks, and Nicholas Wilson. “The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers.” Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 2025.

Open source paper

Reviewed by VSX on July 18, 2026.

Continue exploring

Related research

Technology & tools/Synthesis

A nearby smartphone may not cause brain drain

Across available experiments, a smartphone’s mere presence had no meaningful average effect on cognitive outcomes—challenging a popular productivity claim.

Caveat. This addresses physical proximity, not notifications, active use, compulsive checking, or sleep disruption.

202433 studies · 4,368 participants
Read brief
4 min read
Focus & cognition/Synthesis

Thinking hard is supposed to feel unpleasant

Across tasks and populations, people experienced greater mental effort as more unpleasant—so deep work discomfort is not automatically a sign of personal failure.

Caveat. Feeling effortful does not prove that a task is valuable, sustainable, or worth pushing through indefinitely.

2024170 studies · 4,670 participants
Read brief
4 min read