Mis-skilling is learning something wrong from AI: a trainee takes an error or a bias in the output on trust and builds it into how they reason. The capability is not lost. It is built, and built incorrectly.
The answer, in one line
Learning wrong reasoning from AI: a trainee accepts errors or biases in the output uncritically and builds them into how they think. The term comes from Ke and colleagues in Nature Medicine on 22 May 2026.
Definition#
Mis-skilling: the point at which “uncritical acceptance of AI errors leads trainees to internalize flawed clinical knowledge as fact”. The term is Ke and colleagues’ (Nature Medicine, 22 May 2026), set beside deskilling and never-skilling in a three-part taxonomy.
The failure that leaves no gap to notice#
Deskilling and never-skilling both leave something missing, and a missing skill can show up in a test. Mis-skilling leaves something present. The trainee has a confident pattern of reasoning, it was learned from a source they trusted, and it produces answers that feel fluent. Nothing in their own experience marks it as wrong.
That changes what a remedy looks like. Restoring practice helps when a skill is absent. When the skill is present and wrong, practice without correction reinforces it. The fix is feedback from someone or something that is checked against the facts, and a habit of testing an AI answer before adopting it.
A risk model, not a measured effect#
Ke and fifteen co-authors wrote a Perspective, which reports no data of its own. Its abstract says direct evidence from medical training is absent and rests the concern on learning theory and early signals from other settings. Graded entry. Mis-skilling should be cited for the name and the distinction. How often it occurs, how long it lasts and whether it can be undone have not been measured.
What the nearest evidence can and cannot show#
No study in this research’s evidence base measures people acquiring wrong reasoning from AI in training. The closest design is Bastani and colleagues’ school experiment, where a version of the tool that gave hints rather than answers largely removed the harm that unrestricted access caused. Graded entry. That result concerns how much learners relied on the tool, not whether what they learned was accurate, so it supports the general direction and no more.
Why verification habits matter more than detection#
This section is interpretation. If the learned error feels like knowledge, the learner cannot be the detector. The control sits outside the learner: a supervisor who checks reasoning, an exercise whose answer is known, or a rule that the first answer from the tool is a claim to test. The same logic runs through the illusion of competence, where confidence is a poor guide to what has been learned.
Key sources
- Ke, Y. et al. (2026). AI-induced never-skilling in medical education. Nature Medicine, 32(6), 1997-2006, 22 May 2026. Graded entry.
- Bastani, H. et al. (2025). Generative AI Without Guardrails Can Harm Learning. PNAS. Graded entry.
Related SuperSkills research#
See never-skilling, deskilling and why AI sounds so confident. The term is Ke and colleagues’, not a SuperSkills coinage.
About this research#
Written by Rahim Hirji, author of SuperSkills (Kogan Page, 2026), keynote speaker on AI and human capability, and founder of The SuperSkills Intelligence Company.
Explainer · SS-2026-421 · Graded against the published rubric
Hirji, R. (2026). What is mis-skilling?. The SuperSkills evidence base, SS-2026-421. https://thesuperskills.com/research/what-is-mis-skilling. Last reviewed 12 October 2026.
An evidence review by Rahim Hirji, not peer-reviewed research. For a material claim, cite the underlying study as well; every study here carries its own permanent link.
How citations and IDs work