Why it’s still open
Letting some skills go is rational — few people do long division. But debugging, verifying, noticing that an answer is wrong and knowing what a good answer looks like may be exactly the skills needed to supervise AI. Studies find the largest losses on debugging and self-assessment. There is no map of which skills are load-bearing.
What would count as an answer
Evidence linking the loss of specific sub-skills to worse detection of AI errors in real work.
Papers that tackle it
- Shen Tamkin 2026 - How AI impacts skill formation — The biggest learning gap was in debugging.
- Fernandes et al 2025 - Smarter but none the wiser — Better scores, worse self-assessment.
- Shaw Nave 2026 - Cognitive surrender — Following wrong advice.
- Lee et al 2025 - Generative AI and critical thinking — From doing the work to overseeing it.
- Budzyn et al 2025 - Endoscopist deskilling — Detection skill in medicine.
Methods in play
Who’s closest
Placed by the editor, v1 — corrections welcome.
Related questions
Know a paper that belongs here, or think this is framed wrongly? Suggest a paper or a correction (forms on the Atlas home).