teleo-codex/inbox/archive/health/2026-04-15-clinical-ai-deskilling-2026-review-generational.md

type	title	author	url	date	domain	secondary_domains	format	status	processed_by	processed_date	priority	tags	flagged_for_theseus	extraction_model
source	Clinical AI Deskilling 2026: Never-Skilling, Resident Training, and Generational Risk — Multiple New Publications	Multiple authors (ScienceDirect; PMC; Frontiers Medicine; Wolters Kluwer)	https://www.sciencedirect.com/science/article/pii/S2949820126000123	2026-04-15	health	ai-alignment	literature-review	processed	vida	2026-04-26	high	clinical-ai deskilling never-skilling medical-training residency generational-risk automation-bias AI-safety	moral deskilling as alignment failure mode — AI shaping human ethical judgment through habituation at scale	anthropic/claude-sonnet-4.5

Content

Four new publications in 2026 on clinical AI deskilling — synthesized for the KB:

1. "Artificial intelligence in medicine: a scoping review of the risk of deskilling and loss of expertise among physicians" (ScienceDirect / new journal, 2026) URL: https://www.sciencedirect.com/science/article/pii/S2949820126000123 Key finding: Confirms high deskilling risk for the current generation of clinicians from available, abundant AI. Future research should generate longitudinal and prospective data to track clinical competence in AI-integrated environments. Current evidence largely expert opinion and small-scale studies.

2. "Deskilling dilemma: brain over automation" (Frontiers in Medicine, 2026) URL: https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2026.1765692/full Key finding: Conceptual confirmation of deskilling via neural adaptation — cognitive tasks offloaded to AI → neural capacity for those tasks decreases. Education continuum mapped: students face never-skilling; residents face partial-skilling; established clinicians face deskilling from reliance.

3. "Supervising Resident AI Use Without Losing the Learning" (PMC, 2026) URL: https://pmc.ncbi.nlm.nih.gov/articles/PMC12903258/ Key finding: If AI supplies the first-pass differential diagnosis, the resident may never learn to build and prioritize their own clinical reasoning. Recommendations: residents should generate own differential BEFORE consulting AI. The sequence (human-first, then AI augmentation) is the pedagogical safeguard.

4. "AI survey insights: Newer providers concerned about deskilling" (Wolters Kluwer, 2026) URL: https://www.wolterskluwer.com/en/expert-insights/ai-survey-insights-newer-providers-concerned-about-deskilling Key finding (confirms ARISE 2026 from Session 28): 33% of younger providers rank deskilling as top concern vs. 11% of older providers. This 3:1 generational differential in deskilling concern is the survey confirmation of the ARISE Stanford-Harvard finding. Newer providers are both more exposed to AI-first environments AND more aware of the developmental risk.

Synthesis across these + prior sessions:

The complete deskilling evidence now covers FOUR pathways:

1. Cognitive/diagnostic deskilling — performance decline when AI removed (confirmed, 11+ specialties)
2. Automation bias — commission errors from accepting AI recommendations (confirmed, multiple studies)
3. Never-skilling/upskilling inhibition — trainees fail to acquire skills from AI handling routine cases (Natali 2025 formalization; colonoscopy ADR RCT; Heudel scoping review)
4. Moral deskilling — ethical judgment erosion from habitual AI acceptance (conceptual; Natali 2025; Frontiers 2026)

Temporal qualification (from ARISE 2026, Session 28, now confirmed by Wolters Kluwer survey):

• Current established clinicians (pre-AI trained): NO measurable deskilling → protected by pre-AI foundations
• Current trainees entering AI-saturated environments: NEVER-SKILLING structurally locked in
• This is a temporal sequence, not a divergence

Clinical education recommendation (from resident supervision study): The pedagogical safeguard: human-first reasoning generation, then AI consultation. The sequence matters — AI as second opinion, not first-pass filter. This is a structural educational intervention that addresses never-skilling without eliminating AI assistance.

Agent Notes

Why this matters: The generational deskilling claim is now ready to draft and submit as a PR (flagged overdue since Session 25). The 33% vs 11% generational concern differential and the human-first pedagogical recommendation are the two new additions in this batch that complete the evidence package.

What surprised me: The resident supervision guidance is more concrete than I expected — it's not abstract "AI should supplement not replace" but a specific operational protocol (resident generates differential first, then consults AI). This is the kind of specific, implementable guidance that could become a policy claim.

What I expected but didn't find: Longitudinal prospective evidence of never-skilling. The field still acknowledges this is largely expert opinion and small-scale studies. The never-skilling claim remains "likely" (strong theoretical mechanism + supporting evidence) but not "proven" (no longitudinal RCT). The research gap continues.

KB connections:

• human-in-the-loop clinical AI degrades to worse-than-AI-alone because physicians both de-skill from reliance and introduce errors when overriding correct outputs — the 2026 papers add the temporal dimension: this effect is concentrated in trainees entering AI-saturated environments
• centaur team performance depends on role complementarity not mere human-AI combination — the resident supervision protocol (human-first, then AI) is a specific implementation of role complementarity
• AI scribes reached 92 percent provider adoption in under 3 years because documentation is the rare healthcare workflow where AI value is immediate unambiguous and low-risk — contrast: documentation AI does NOT create deskilling risk (no diagnostic reasoning required); the deskilling risk is diagnostic/clinical reasoning AI

For Theseus cross-domain: Moral deskilling (Natali 2025; Frontiers 2026) — the finding that AI habituation erodes ethical sensitivity and moral judgment — is an alignment failure mode that operates at the societal scale. If millions of physicians become less ethically sensitive through AI habituation, this is a slow-moving value alignment problem: AI systems are shaping human ethical judgment through repeated interaction. This is the OPPOSITE of the typical alignment framing (human values constraining AI) — here AI is shaping human values.

Extraction hints:

• PRIMARY CLAIM (ready for PR): "Clinical AI deskilling is a generational risk — currently practicing clinicians trained before AI report no measurable performance degradation, while trainees entering AI-saturated environments face never-skilling as a structural consequence of reduced unassisted case volume"
• Evidence: ARISE 2026 (33% vs 11% generational concern), Heudel scoping review, colonoscopy ADR RCT, Wolters Kluwer survey confirmation
• Confidence: likely
• SECONDARY CLAIM (speculative): "Habitual AI acceptance in clinical settings produces moral deskilling — erosion of ethical sensitivity and contextual judgment — as physicians offload ethical reasoning to AI systems that lack capacity for moral context"
• Evidence: Natali 2025, Frontiers 2026 — conceptual only, flag for Theseus
• Confidence: speculative

Curator Notes

PRIMARY CONNECTION: human-in-the-loop clinical AI degrades to worse-than-AI-alone because physicians both de-skill from reliance and introduce errors when overriding correct outputs WHY ARCHIVED: Completes the evidence package for the temporal deskilling claim (current clinicians protected, trainees at risk). The generational framing plus 33% vs 11% survey data are the new additions. Flagged for Theseus on moral deskilling. EXTRACTION HINT: The temporal qualification is the key new insight — extract as a single claim with explicit temporal scope rather than a divergence. The moral deskilling pathway needs Theseus cross-domain flag included in the claim file.