Is AI Making Physicians Dumber? What the Evidence Says

No robust evidence supports the claim that AI uniformly “makes physicians dumber”. The literature describes a more nuanced risk: overreliance on AI can erode clinical reasoning in specific circumstances. This is usually split into deskilling (loss of established skills), mis-skilling (uncritical adoption of flawed AI outputs) and never-skilling (failure to develop independent reasoning in the first place, most relevant to trainees).[1][2][3] What matters is how AI is used, as a substitute for reasoning or as a tool that informs it, rather than whether it is used at all.

What is the evidence for a deskilling risk?

Older data on computer-aided detection show the mechanism. In a mammography study, experienced readers had a 14% drop in diagnostic sensitivity when challenging images carried computer-aided-detection prompts. Internal medicine residents’ ECG interpretation accuracy fell from 57% to 48% when tracings were pre-annotated with inaccurate computer diagnoses.[4]

A more recent study found that frequent AI tool use correlated with lower critical-thinking scores, an effect mediated by increased “cognitive offloading” and more pronounced in younger users.[2] A British Journal of Psychiatry analysis similarly warns that younger, less experienced physicians are the most susceptible to over-trusting AI recommendations, with potential effects on clinical reasoning and patient communication.[3]

What does family medicine data show?

A mixed-methods study of 114 clinicians using AI-assisted diagnosis in family medicine simulations found overall sentiment toward AI was positive. However, clinicians did not feel current AI assistance improved their diagnostic capability, and on a quantitative measure it was associated with a negative effect on performance (β = -0.421, P = .02). The authors attributed this to AI reasoning patterns being poorly aligned with how clinicians actually think through a diagnosis.[5]

What is the evidence against a blanket “dumbing down” effect?

• Fewer errors with decision support: a cluster-randomised trial of a generative AI decision-support tool across nearly 40,000 primary care visits found fewer diagnostic and treatment errors with AI support, and a Pakistani trial found large gains in diagnostic reasoning among physicians trained to use large language models effectively.[6]
• Clinicians kept their own judgement: in a primary care cluster-RCT, LLM-assisted consultations improved diagnosis appropriateness, treatment-plan quality and documentation completeness. Clinicians fully adhered to AI advice in fewer than 20% of flagged safety alerts, so they were not passively following the system.[7]
• AI as a second opinion: a collaborative-workflow trial found AI raised the floor of diagnostic performance (compressing the worst-performing cases) without clinicians becoming dependent or less accurate than AI alone.[8]
• Colonoscopy: AI-assisted detection improves adenoma detection rates similarly regardless of endoscopist experience, which argues against experience-based deskilling in that setting.[9][10]

What is the practical takeaway for family medicine?

The technology itself is not the determinant. Overreliance, meaning using AI to replace rather than inform reasoning, or accepting outputs uncritically, is what correlates with skill erosion. Structured, critically engaged use is associated with improved diagnostic accuracy and documentation quality.[1][2][7][8] Training programmes are increasingly emphasising supervised, “informing, not offloading” AI use to reduce the risk of deskilling and never-skilling in residents and early-career physicians.[1][2]

This article summarises published evidence for healthcare professionals. It is not a substitute for clinical judgement. Written by Dr Seah, a locum doctor who built Locum Radar for other locum healthcare professionals.

References

1. Ke Y, Jin L, Ong JCL, et al. AI-induced never-skilling in medical education. Nat Med. 2026;32(6):1997-2006. doi:10.1038/s41591-026-04438-y
2. Abdulnour RE, Gin B, Boscardin CK. Educational strategies for clinical supervision of artificial intelligence use. N Engl J Med. 2025;393(8):786-797. doi:10.1056/NEJMra2503232
3. Monteith S, Glenn T, Geddes JR, et al. Artificial intelligence and deskilling in medicine. Br J Psychiatry. 2026:1-3. doi:10.1192/bjp.2025.10496
4. Cabitza F, Rasoini R, Gensini GF. Unintended consequences of machine learning in medicine. JAMA. 2017;318(6):517-518. doi:10.1001/jama.2017.7797
5. Hah H, Goldin DS. How clinicians perceive artificial intelligence-assisted technologies in diagnostic decision making: mixed methods approach. J Med Internet Res. 2021;23(12):e33540. doi:10.2196/33540
6. Bergman A, Wachter RM, Emanuel EJ. A licensure framework for autonomous clinical AI. JAMA. 2026;335(20):1751-1754. doi:10.1001/jama.2026.5483
7. Agweyu A, Mwaniki P, Menon V, et al. Generative AI-enabled clinical decision support system in primary care: a pragmatic, cluster-randomized trial. Nat Med. 2026;32(8):3032-3039. doi:10.1038/s41591-026-04503-6
8. Everett SS, Bunning BJ, Jain P, et al. From tool to teammate in a randomized controlled trial of clinician-AI collaborative workflows for diagnosis. NPJ Digit Med. 2026;9(1):409. doi:10.1038/s41746-026-02545-1
9. Repici A, Spadaccini M, Antonelli G, et al. Artificial intelligence and colonoscopy experience: lessons from two randomised trials. Gut. 2022;71(4):757-765. doi:10.1136/gutjnl-2021-324471
10. Makar J, Abdelmalak J, Con D, Hafeez B, Garg M. Use of artificial intelligence improves colonoscopy performance in adenoma detection: a systematic review and meta-analysis. Gastrointest Endosc. 2025;101(1):68-81.e8. doi:10.1016/j.gie.2024.08.033