AI Overreliance and Thinking Skills

Summary

Overreliance on AI means leaning on a tool past the point where you would still want the underlying skill.

Automation bias is the older parent concept — the tendency to defer to automated systems even when they are wrong — and has decades of research behind it.

Recent 2024-2026 work on AI assistants extends the same questions to drafting, summarising, calculating, and reasoning at conversational speed.

Overreliance is not the same as AI use — it is the absence of regular practice alongside it.

Senwitt's framing: keep using AI, and keep a daily place where the underlying Skills still get reps.

What is AI overreliance?

AI overreliance is leaning on an AI assistant past the point where you would still want to be able to do the underlying thinking yourself. It is not the same as using AI — it is using AI without keeping any practice of the skill on the side. The result is not catastrophic, but it is real: drafting, calculating, recalling, reading attentively, and reasoning through small problems get fewer and fewer reps over time. Senwitt's answer is a daily Set across six Skills, designed to sit next to AI use, not against it.

The concept has an older name in the human-factors literature: "automation bias." Parasuraman and Manzey's widely cited 2010 review in Human Factors (Parasuraman & Manzey, 2010) defines automation bias as the tendency to over-rely on automated decisions — including failing to detect automation errors and following automated guidance against contrary evidence. The bias predates LLMs by decades; the LLMs just give it more surfaces.

What the source says

Recent research has started to ask, in education and in industry, what happens when AI dialogue systems become the first stop for thinking work. The Springer paper on overreliance on AI dialogue systems in student learning (Springer 2024) documents specific cognitive patterns when AI is the default and practice is optional. The 2024 MDPI Societies study (MDPI) reported an inverse correlation between AI tool usage frequency and self-reported critical-thinking engagement, mediated by cognitive offloading.

Microsoft Research published a survey study in early 2025 (Lee, Sarkar, et al., CHI 2025) of 319 knowledge workers reporting on 936 instances of generative-AI use at work. The headline finding: higher confidence in AI was associated with less critical thinking on the task, while higher confidence in one's own skill was associated with more. The same study found that generative-AI shifted the cognitive work users did from information gathering and analysis toward information verification and AI response integration. The PsyPost summary of the broader 2025 critical- thinking work (PsyPost) and the Phys.org piece (Phys.org) cover the same picture from different angles.

For coding specifically, the Anthropic 2026 study on AI assistance and coding-skill formation (Anthropic Research) — a 52-person controlled study on learning an unfamiliar Python library — reported a ~17% reduction in independent skill mastery for the AI-assisted cohort. Practitioner-side coverage (Addy Osmani's "skill atrophy" essay, Substack; VirtusLab on cognitive debt in code, VirtusLab) describes the same pattern from the field.

What the source does not say

These papers do not claim that AI use causes general cognitive decline. They do not call for banning AI from learning or work. They explore the specific question of what practice patterns survive when AI is always available — and the consistent thread is that deliberate practice matters more, not less, when delegation is easy.

The Microsoft Research and MDPI work is cross-sectional and self-report- based. The Anthropic study is small (n=52), task-specific, and applies to learning rather than applied work. The Springer overreliance paper is one data point in an ongoing literature. None of these establish causation in the strong sense — "AI use causes cognitive harm." What they establish is direction: AI overuse correlates with reduced engagement, reduced ownership, and reduced independent skill formation in the specific settings studied. That is enough signal to act on; it is not enough to declare medical consequences.

What this means for daily practice

For an AI-heavy professional, the healthiest stance is not to use less AI. It is to keep a small, daily, deliberate practice of the underlying Skills so the muscle does not atrophy. Senwitt's daily Set is built exactly for that — short, mixed, finishable, and useful even on days when most of the rest of the work runs through AI.

The practical guidance that has emerged across the 2024-2026 literature is consistent: bound AI use to windows; do at least one thinking task a day without AI; verify AI output against your own understanding rather than the other way around; for any decision that matters, do at least the first pass yourself. None of those require quitting AI. They require keeping enough of your own thinking in active use that the verification step is meaningful.

From Senwitt · advertisement

The text above is editorial. What follows is a promotional message from Senwitt, the maker of this site. Senwitt is a brain-exercise app and is not a medical product. Read the full disclaimer in the footer.

References — canonical order.

1.Kosmyna, N., Hauptmann, E., Yuan, Y.T., Situ, J., Liao, X.-H., Beresnitzky, A.V., Braunstein, I., & Maes, P. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. arXiv:2506.08872. arxiv.org/abs/2506.08872. Primary anchor.

2.Stanković, M., Hirche, E., Kollatzsch, S., & Doetsch, J.N. (2026). Comment on: Your Brain on ChatGPT. arXiv:2601.00856. arxiv.org/abs/2601.00856. The methodological critique — paired with Kosmyna.

3.Risko, E.F. & Gilbert, S.J. (2016). “Cognitive Offloading.” Trends in Cognitive Sciences 20(9):676–688. DOI: 10.1016/j.tics.2016.07.002.

4.Sparrow, B., Liu, J., & Wegner, D.M. (2011). “Google Effects on Memory: Cognitive Consequences of Having Information at Our Fingertips.” Science 333(6043):776–778. DOI: 10.1126/science.1207745.

5.Simons, D.J., Boot, W.R., Charness, N., Gathercole, S.E., Chabris, C.F., Hambrick, D.Z., et al. (2016). “Do ‘Brain-Training’ Programs Work?” Psychological Science in the Public Interest 17(3):103–186. DOI: 10.1177/1529100616661983.

6.FTC v. Lumos Labs, Inc. (2016). “Lumosity to Pay $2 Million to Settle FTC Deceptive Advertising Charges for Its ‘Brain Training’ Program.” Stipulated $50M judgment, suspended on payment of $2M. ftc.gov press release (Jan 5 2016).

7.Max Planck Institute for Human Development & Stanford Center on Longevity (2014). A Consensus on the Brain Training Industry from the Scientific Community. Signed by 70 neuroscientists/psychologists. longevity.stanford.edu.

Not brain training. Brain exercise.

Senwitt is a daily brain exercise app, not a brain training program. We do not claim to improve general cognition, prevent cognitive decline, or treat any condition. Independent scientific consensus — the 2014 Stanford Center on Longevity / Max Planck Institute statement signed by 70 neuroscientists, the 2016 Simons et al. review in Psychological Science in the Public Interest, and the FTC's 2016 settlement with Lumos Labs — has concluded that “brain training” claims are not supported by the evidence. Senwitt is built on a different premise: skills you actively practice get sharper; skills you stop practicing fade.

Why we avoid old brain-training claims

AI overreliance and the thinking skills underneath.