Signal

When a machine takes over the routine cases, the practice that kept people sharp goes with them. Doctors at four Polish clinics found fewer precancerous polyps on their own once an AI spotter had become routine, while a Japanese hospital that ran the same kind of AI in half its rooms for three years saw no such decline, so the question is open. Junior engineers who let an assistant write the code learned less than those who wrote it themselves, unless they kept asking it why, and in three randomised trials people who practised arithmetic or reading with AI help then solved fewer problems alone and gave up on more of them. Aviation met this first, and Europe's regulator asks airlines to find moments for pilots to fly by hand. Medicine is now borrowing the rulebook: eye doctors at Moorfields, writing with Lufthansa's flight-safety department, propose a minimum quota of cases without AI and unannounced breaks from it, and Washington State's medical regulator has drafted a policy telling clinicians to practise unassisted reasoning now and then. In a BCG study of 70 senior leaders, half already see skills slipping, and one Indian bank holds an AI-free session on the first Friday of every month. Aviation learned that skill decay under automation is predictable and reversible. Medicine is learning it in real time.

So what?

Skill used to be a by-product of routine work. Now practice has to be planned.

Evidence

The Lancet Gastroenterology & Hepatology, 12 August 2025 (study): at four colonoscopy centres in Poland, 19 experienced endoscopists were compared before and after AI assistance became routine. Their detection rate for precancerous polyps in colonoscopies done without AI fell from 28.4 percent to 22.4 percent. The authors note that an observational study cannot rule out other causes.

Anthropic, 29 January 2026 (study): 52 mostly junior software engineers learned a new programming library with or without an AI assistant. The AI group scored 50 percent on the quiz that followed, against 67 percent for those who coded by hand. People who asked the assistant for explanations kept learning, and those who handed it the whole task did not.

arXiv, 6 April 2026 (study): in three randomised trials with 1,222 people, those who practised fraction arithmetic or reading comprehension with an AI assistant then did worse on their own, solving 57 percent of problems against 74 percent for the unassisted group in the first trial, and they gave up on more of them. A preprint, revised in October 2026.

npj Digital Medicine, 31 January 2026 (essay): eye doctors at Moorfields and UCL, writing with Lufthansa's flight-safety department, propose that clinicians keep a minimum quota of cases without AI as their benchmark and face unannounced breaks from it, the way pilots are expected to keep flying by hand.

Worth asking

Which skill of yours would fade first if a machine did it for you for a year?

Fictional concepts this signal echoes

Today's signals. Tomorrow's possibilities.

New science-fiction stories, emerging signals and essays exploring how life and work could change. Fresh perspectives and questions worth thinking about, delivered to your inbox.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
Unsubscribe any time. Privacy Policy