C18 AI deskilling
narrowed 1 attack: 1 narrowed · serves Q2 · last change 16 October 2026
Professionals who routinely work with AI assistance may perform worse at the same task when the AI is taken away than they did before they started using it; this is not established: one observational study found a drop (Budzyń et al. 2025), while a prospective study in the same endoscopists found none (Pedersen et al. 2026).
In practice
When people rely on AI for a skill they still need, plan regular practice without it and check unaided performance from time to time.
Go deeper
Where it holds
Main limits
- About skilled adults at work, not learners practising with AI (C2).
- The colonoscopy study is observational, before and after, over a few months (Budzyń et al. 2025).
- Automation research shows over-reliance even in experts (Parasuraman & Manzey 2010).
- A prospective study found no drop in endoscopists' unaided detection after a period with AI (Pedersen et al. 2026).
All limits (3)
Inside the circle: the core of the claim. Outside: every bound and carve-out in the claim file, numbered, with the attack that forced or added it where the record shows one.
forced by a narrowing attackin the original wordingwritten at creation, or added without a narrowing
Warm cards in the list were added by later attacks; plain cards were in the original draft or written at creation.
Consistency bound added 2026-10-16 (week run): Pedersen et al. 2026 (Endoscopy; 13 endoscopists, 5,013 colonoscopies, before/during/after computer-aided detection) found no significant change in unaided polyp detection after the AI period.
Forced by attack #1 · Pedersen et al. 2026 · 16 October 2026
Unaided performance of trained professionals before versus after a period of routine AI assistance on the same task; not learners using AI during practice (C2), not people's beliefs about their own thinking (candidate K45).
Written when the claim was created (16 October 2026); not forced by an attack
Evidence on file: Budzyń et al. 2025 (Lancet Gastroenterology & Hepatology; four Polish centres; adenoma detection in non-AI colonoscopy 28.4% before vs 22.4% after AI introduction, −6.0 points, 95% CI −10.5 to −1.6; observational); Parasuraman & Manzey 2010 (Human Factors; review of automation complacency and bias).
Written when the claim was created (16 October 2026); not forced by an attack
1 of 3 bounds and carve-outs were forced by a specific attack.
How it changed
The ledger records 1 attack, but no earlier wording of C18 is on record, so the text cannot be replayed step by step. Only the current wording is shown.
Professionals who routinely work with AI assistance may perform worse at the same task when the AI is taken away than they did before they started using it; this is not established: one observational study found a drop (Budzyń et al. 2025), while a prospective study in the same endoscopists found none (Pedersen et al. 2026).
What would change this
What would kill it: A well-controlled study finds that professionals' unaided performance after months of routine AI assistance is the same as or better than before. Added as a claim on 16 October 2026, promoted from a candidate belief in the evidence library.
Attack record
1 attack: 1 narrowed. Counts of attacks, not a score.
Full attack list (1)
In a prospective three-phase study, endoscopists' unaided polyp detection after a period of routine computer-aided detection was no lower than before they started using it, which meets C18's kill-condition for this setting.
Strongest reading and verdict reasoning
Steelmans to: C18 dies if professionals' unaided performance after months of routine AI assistance is the same as or better than before; this is the first prospective before/during/after test in the same endoscopists.
Why: Week run 2026-10-16, cycle 6 Europe/Stockholm. Full text read (Thieme open access, PMC13552228; stored privately). Three Norwegian clinics, 13 endoscopists (7 inexperienced, 6 experienced), 5,013 colonoscopies, fixed phase order. Inexperienced: PDR-5 31.9% before, 39.5% with CADe (OR 1.43, 95% CI 1.11–1.84), 36.3% after removal (OR vs before 1.03, 95% CI 0.79–1.34); experienced: no significant change; no significant before-vs-after change overall or in either group. Kill withheld: not randomised, 13 endoscopists, a different endpoint (polyps ≥5 mm vs adenoma detection in Budzyń et al. 2025), and some inexperienced endoscopists did few phase-3 procedures; Budzyń et al. still show a drop. Narrowed: the drop is shown in one observational study and not found in one prospective study. Not a preprint.
Evidence
5 linked sources: 2 supports, 2 limits, 1 challenges (first-pass draft, pending human review). supports: Budzyn et al. 2025, Parasuraman & Manzey 2010 · limits: Casner et al. 2014, Ebbatson et al. 2010 · challenges: Pedersen et al. 2026