Arvind Narayanan argues in Normal Tech that Moravec’s paradox misleads AI forecasts

Arvind Narayanan argues in Normal Tech that Moravec’s paradox misleads AI forecasts

Published January 29, 2026, Arvind Narayanan argues in Normal Tech that Moravec’s paradox—the idea that tasks difficult for people are easy for AI, and vice versa—has not been empirically tested and cannot predict which AI capabilities will come next. He suggests the pattern may reflect which tasks AI researchers choose to study: tasks that are easy for both people and AI, or hard for both, often attract less attention. He also challenges the evolutionary story used to explain the paradox, arguing that reasoning in open-ended settings depends on common-sense knowledge and other limits shared by people and AI. Narayanan says the same simplistic model can create alarm about AI replacing human expertise in fields such as science, while offering false comfort about robotics. A capability breakthrough, he argues, does not immediately reshape society: commercialization and deployment take time, and infrastructure may be missing. He points to self-driving cars as a case where policymakers could have prepared for effects on workers earlier, but instead are reacting as the vehicles reach roads. His conclusion is to focus less on predicting breakthroughs and more on adapting to technologies already known to be coming. How can policymakers prepare for a technology’s effects when its capabilities and deployment timelines remain difficult to predict?

Publication: d74932b3-7d89-5fb4-3d8a-32e27324f3f6

attribution — pass: The essay is bylined to Arvind Narayanan on the normaltech.ai site, whose publisher is Narayanan and Sayash Kapoor. The reading attributes every argument to Narayanan and keeps his claims as his own assertions rather than presenting them as established facts. It is a single argued essay, not a listing.

Fact checking Moravec's paradox The famous aphorism is neither true nor useful Arvind Narayanan

summary — pass: The reading defines Moravec's paradox in plain language and conveys Narayanan's main arguments: the paradox is untested, it may be a selection effect, and its evolutionary story is dubious. It also covers the alarmism and false-comfort consequences, slow diffusion, the self-driving car example, and his adapt-rather-than-predict conclusion. His strong thesis that superhuman AI reasoning is a myth is softened somewhat, but the reading's force stays intact. The question follows from the source.

Instead of relying on prediction, we should get better at shaping and adapting to the tech that we actually know for sure is coming.

date — pass: originalPublished is 2026-01-29T22:35:47Z, which falls on January 29, 2026 in UTC, and the dateline says Jan 29, 2026. The modified date equals the publication date. The page is a publisher page, not an archive copy. The reading uses no hindsight framing and presents no later events.

Jan 29, 2026

support — pass: Each claim excerpt is an exact substring of the source and supports its claim. The summary's statements are grounded in the essay: slow commercialization and missing infrastructure, the self-driving car policy lag, and limits shared by human and AI reasoning. The limitations accurately note that the essay offers no empirical test and no detailed policy plan. The reading adds no unsupported benefits or costs.

Even breakthrough technologies take a long time to be successfully commercialized and deployed.

media — pass: No media was selected, and no verified media candidates were supplied, so selecting none is appropriate.