Do experienced clinicians lose unaided skill after routine AI use?

Research question

Does routine use of diagnostic AI lower experienced clinicians' detection rates when they work without it, tested with randomized AI-off sessions scheduled over time and analysed by each clinician's cumulative AI exposure?

Why it matters

The clearest sign of deskilling among experienced professionals is a drop in adenoma detection in colonoscopies done without AI after AI was introduced, but the study is observational and has several possible confounders, and a Japanese prospective study is reported to find no decline. If routine AI use does erode unaided skill, services need AI-off practice and outage plans; if not, the worry can be set aside. Medicine is the setting where the question can be settled, because detection rates are routinely recorded.

The debate lists this among what we still do not know: Does routine AI use deskill experienced professionals in the skills they no longer exercise?

Evidence that motivates it

1. In an observational study at four Polish centres, experienced endoscopists’ adenoma detection in colonoscopies done without AI fell from 28.4% to 22.4% in the three months after an AI detection system was introduced; the study is observational, and its authors list differences between the cohorts and a near doubling of colonoscopy volume after AI was introduced among its limitations. (LD04-SL-C14 v1.0)
   Experienced endoscopists' detection in non-AI colonoscopies fell from 28.4% to 22.4% after AI was introduced, in an observational before-and-after study whose authors list cohort differences and a near doubling of volume among its limitations.
   Cites: Budzyń et al. (2025), “Endoscopist deskilling risk after…”
2. In the customer-support firm where an AI assistant raised productivity most for less-skilled agents, agents who had used it still outperformed their own pre-AI baseline during software outages when it gave no suggestions, more so after longer exposure and closer adherence to its suggestions. (LD04-SL-C09 v1.0)
   In customer support, retained gains after AI exposure point the other way, so the direction of the effect on incumbents is not settled across settings.
   Cites: Brynjolfsson et al. (2024), “Generative AI at Work”

Claims are shown as they read on the day this pack was built; claim pages keep every earlier version.

A possible design

Within an endoscopy or screening service that already uses AI, randomize individual sessions (or days) to AI-on or AI-off for each clinician over 12 to 24 months, so that every clinician has AI-off sessions spread across time. Compare detection in AI-off sessions over time and by cumulative AI-assisted volume, with patient-mix adjustment. Several centres reduce the risk that one site's changes drive the result. Secondary analysis: control-arm detection drift in completed AI trials, where data can be shared.

Data needed

Session-level procedure records with detection outcomes and AI status; ethics approval for randomized AI-off sessions; patient covariates for case mix; trial registration.

Feasibility

Funded team project. Medium; clinical trial infrastructure, ethics and data management; the AI systems are already in use at the participating services. AI-off sessions may lower detection for the patients concerned, so the ethics case must rest on genuine uncertainty and on the trial's ability to detect harm. A registered Norwegian colonoscopy study reportedly uses without-AI, with-AI, and without-AI phases; its record must be checked before designing a new trial, and this design would add randomized rather than sequential AI-off sessions and an exposure gradient.

Has this been done?

Partly done already: related work covers some of this question (first-pass screen of 14 October 2026; not yet checked in depth).

Published evidence is two observational colonoscopy studies that disagree: a Polish before-and-after study finding lower detection without AI after AI was introduced, and a Japanese single-centre prospective study reported to find no decline. A reply by the Japanese authors says these studies differ too much in periods, populations and methods to settle the question. Search listings point to a registered Norwegian colonoscopy study with phases without AI, with AI and without AI again, but its record could not be opened. No published randomized trial was found that schedules AI-off sessions after exposure and tracks unaided detection by cumulative exposure.

Where the search could not reach. Screened by web search (8 queries) and DOI records on the date shown. Trial registries (ClinicalTrials.gov, WHO ICTRP) and the Endoscopy publisher site could not be opened, so registered trials may be missed or misidentified. PubMed and Embase were not searched, and secondary analyses of control-arm detection in existing AI trials were not searched for. Absence of a hit is not proof that the work does not exist.

Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.

Nearest work found

• Budzyń et al., "Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study," Lancet Gastroenterology & Hepatology 10(10), 2025
  Accepted manuscript read in full for the LD-04 debate. Nineteen experienced endoscopists at four Polish centres; detection in non-AI colonoscopies fell from 28.4% to 22.4%. Observational before-and-after comparison nested in a trial that assigned AI use by date.
• Okumura et al., "Long-term impact of computer-aided adenoma detection: a prospective observational study," Endoscopy 58(2):121–129 (2026) Judged from an abstract, summary or search listing only
  Seen in search results and commentaries only; the paper was not read. Single centre in Japan, 2021–2023: detection in non-AI colonoscopies reportedly did not decline and high detectors kept improving; an editorial notes the long-term skill effect was not directly evaluated.
• Okumura & Misawa, "Reply to Henry and East," Endoscopy 58(3):315–316 (2026) Judged from an abstract, summary or search listing only
  Seen in search results only, with authors confirmed through the DOI record. Says recent deskilling studies vary in periods, populations and methods, precluding definitive conclusions, and calls for longer, larger studies.
• van de Sande, Sleijfer, Buijsman, Gommers & van Genderen, correspondence on endoscopist deskilling, Lancet Gastroenterology & Hepatology 10(12):1061–1062 (2025) Judged from an abstract, summary or search listing only
  Bibliographic record read; content as summarised in the LD-04 source notes. A letter framing the Polish finding as possible skill reallocation and raising AI-off practice.
• Norwegian colonoscopy study with without-AI, with-AI and without-AI phases, ClinicalTrials.gov (record number not confirmed) Judged from an abstract, summary or search listing only Some details not confirmed
  Seen in search results only; the registry could not be opened, so its number, sponsor, design and status are not confirmed. Reportedly tests a learning effect on polyp detection after AI use with a scheduled AI-off phase: the closest to the core design found.

References

Sources behind the motivating claims:

• Budzyń et al., “Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study,” Lancet Gastroenterology & Hepatology 10(10), 2025 (SRC-SL-BUDZYN-ET-AL-LANCET-GH-2025, original)
• Brynjolfsson, Li & Raymond, “Generative AI at Work,” arXiv:2304.11771 v2, November 2024 (SRC-SL-BRYNJOLFSSON-LI-RAYMOND-ARXIV-2304-11771-V2, original)

BibTeX

The first entry cites this opportunity; then its nearest work; then the sources above. Nearest work and sources are recorded as reference strings, so each goes in whole as the title; add authors and year from the original before you cite it.

@misc{sociolens_ro_ld04_06_v1,
  author = {{SocioLens project}},
  title = {Do experienced clinicians lose unaided skill after routine AI use?},
  year = {2026},
  date = {2026-10-14},
  howpublished = {\url{https://sociolens-web.pages.dev/opportunities/ro-ld04-06/v1/}},
  note = {Research opportunity RO-LD04-06, version v1}
}

@misc{ro_ld04_06_nearest_1,
  title = {Budzyń et al., "Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study," Lancet Gastroenterology \& Hepatology 10(10), 2025},
  url = {https://doi.org/10.1016/S2468-1253(25)00133-5},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}

@misc{ro_ld04_06_nearest_2,
  title = {Okumura et al., "Long-term impact of computer-aided adenoma detection: a prospective observational study," Endoscopy 58(2):121–129 (2026)},
  url = {https://doi.org/10.1055/a-2661-2624},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated; Judged from an abstract, summary or search listing only}
}

@misc{ro_ld04_06_nearest_3,
  title = {Okumura \& Misawa, "Reply to Henry and East," Endoscopy 58(3):315–316 (2026)},
  url = {https://doi.org/10.1055/a-2764-0296},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated; Judged from an abstract, summary or search listing only}
}

@misc{ro_ld04_06_nearest_4,
  title = {van de Sande, Sleijfer, Buijsman, Gommers \& van Genderen, correspondence on endoscopist deskilling, Lancet Gastroenterology \& Hepatology 10(12):1061–1062 (2025)},
  url = {https://doi.org/10.1016/S2468-1253(25)00288-2},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated; Judged from an abstract, summary or search listing only}
}

@misc{ro_ld04_06_nearest_5,
  title = {Norwegian colonoscopy study with without-AI, with-AI and without-AI phases, ClinicalTrials.gov (record number not confirmed)},
  url = {https://clinicaltrials.gov/},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated; Judged from an abstract, summary or search listing only; Some details not confirmed}
}

@misc{src_sl_budzyn_et_al_lancet_gh_2025,
  title = {Budzyń et al., “Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: a multicentre, observational study,” Lancet Gastroenterology \& Hepatology 10(10), 2025},
  url = {https://doi.org/10.1016/S2468-1253(25)00133-5},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}

@misc{src_sl_brynjolfsson_li_raymond_arxiv_2304_11771_v2,
  title = {Brynjolfsson, Li \& Raymond, “Generative AI at Work,” arXiv:2304.11771 v2, November 2024},
  url = {https://arxiv.org/abs/2304.11771v2},
  note = {Reference as recorded by SocioLens; authors, title and year are not separated}
}