Position · last reconsidered September 17, 2026

AI inverts Polanyi's paradox

Autor's claimpositionOn this site, an idea someone holds about a question, shown with its author and its source. that AIartificial intelligence (AI)Computer programs that produce text, images, code or answers by learning from huge numbers of examples — ChatGPT, Claude and Gemini are well-known ones., unlike older computers, is good at unwritten know-how and bad at following rules.

In our words

Economist David Autor starts from Polanyi's paradoxPolanyi's paradoxMichael Polanyi's observation that we know more than we can tell, as with riding a bike or recognizing a face.: we know more than we can put into words. Older computers needed every step spelled out, so they could follow rules but failed at work that runs on judgmentjudgmentThe ability to decide well in a particular case, especially when no rule tells you what to do.. AI flips this. It picks up unwritten know-how from examples, yet it is unreliable with facts and rules. Used well, it could let workers with solid basic training, such as nurses, make high-stakes calls now left to top expertsexpertiseWhat someone with long practice in a field knows, which lets them see quickly what a beginner misses..

David AutorDavid AutorAmerican economist (1967-)Description from WikidataLabor economics2024

In their words

In a case of cosmic irony, AI is not trustworthy with facts and numbers — it does not respect rules. AI is, however, remarkably effective at acquiring tacit knowledge.
David AutorDavid AutorAmerican economist (1967-)Description from Wikidata, AI Could Actually Help Rebuild The Middle Class, section “Expertise In The Era Of Artificial Intelligence”Read the passage

Connected ideas

challenges

derived from

related to

responds to