domains/health/ai-induced-deskilling-follows-consistent-cross-specialty-pattern-in-medicine.md

@@ -46,3 +46,10 @@ Radiology residents using AI assistance showed resilience to large AI errors (>3
 **Source:** Heudel et al., Insights into Imaging, Jan 2025 (PMC11780016)
 
 The Heudel radiology study is frequently cited (including by Oettl 2026) as evidence for AI-induced upskilling, creating apparent contradiction with deskilling evidence. However, close reading reveals it only shows performance improvement with AI present, not durable skill acquisition. The study's own title poses 'Upskilling or Deskilling?' as an open question, and the data cannot answer it without a post-training, no-AI assessment arm. This represents the core methodological limitation in the upskilling literature: conflating AI-assistance effects with learning effects.
+
+
+## Challenging Evidence
+
+**Source:** Heudel et al., Insights into Imaging, Jan 2025 (PMC11780016)
+
+The Heudel radiology study is frequently cited as counter-evidence to the deskilling pattern, but close reading reveals it only shows performance improvement with AI present, not durable upskilling. The study's own title poses 'Upskilling or Deskilling?' as an open question, and the data cannot answer it without a post-training, no-AI assessment arm. This represents the strongest available evidence for the 'upskilling' position, yet it has a critical methodological limitation that prevents it from contradicting the deskilling RCT evidence (colonoscopy ADR 28.4%→22.4% post-removal; radiology false positives +12%).

domains/health/divergence-human-ai-clinical-collaboration-enhance-or-degrade.md

@@ -83,3 +83,10 @@ Heudel et al. (2025) radiology study (n=8 residents, 150 chest X-rays) shows 22%
 **Source:** Heudel et al., Insights into Imaging, Jan 2025 (PMC11780016)
 
 Heudel et al. (2025) radiology study (n=8 residents, 150 chest X-rays) shows 22% improvement in inter-rater agreement (ICC-1: 0.665→0.813) and significant error reduction (p<0.001) when AI is present. However, the study does NOT test post-training performance without AI—it only documents improved performance WHILE AI IS PRESENT. This is the methodological gap in the 'upskilling' literature: no evidence of durable skill retention after AI-assisted training ends. The study does show residents can reject major AI errors (>3 points), maintaining ~2.75-2.88 average error when AI makes large mistakes, suggesting some critical evaluation capacity persists during AI use.
+
+
+## Extending Evidence
+
+**Source:** Heudel et al., Insights into Imaging, Jan 2025 (PMC11780016)
+
+Heudel et al. (2025) radiology study (n=8 residents, 150 chest X-rays) shows 22% improvement in inter-rater agreement (ICC-1: 0.665→0.813) and significant error reduction (p<0.001) when AI is present. However, the study does NOT test post-training performance without AI—it only documents improved performance WHILE AI IS PRESENT. This is the methodological gap at the core of the divergence: the 'upskilling' literature cites performance-with-AI as evidence of durable skill acquisition, but no study has demonstrated retention after AI removal. The study does show residents can reject major AI errors (>3 points), maintaining ~2.75-2.88 average error when AI makes large mistakes, suggesting some preserved judgment capacity.