AI Is Not a Normal Technology

Summary

This essay argues that the central disagreement about AI is whether it can continue acquiring capabilities until it covers the full range of human work. If it can, the author says, the usual comparison with technologies such as cars or factory automation breaks down: AI could in principle perform not only existing jobs but also any new jobs created in response. The author’s argument rests on three premises: jobs exist, AI will take work it performs better than humans, and AI can eventually do everything better. Human-made goods and services may retain value when authenticity is desired, but the essay treats that demand as a narrow preference rather than a broad economic necessity; anything AI can produce could become abundant and cheap. Examples from AI writing, code, and mathematics illustrate the distinction between producing outputs and preserving the training, understanding, taste, and ability to formulate new questions that traditionally accompanied the work. Quoting mathematician Terence Tao, the author argues that optimizing for solved problems can conflict with mathematics’ deeper goal of conceptual insight and accessibility. The essay also challenges the “AI as normal technology” framework, which emphasizes social institutions’ ability to respond because AI remains under human control and does not suddenly “FOOM.” In the author’s view, corporations, governments, and geopolitical competition create strong incentives to deploy increasingly capable systems, leaving no obvious control mechanism that would prevent widespread adoption. The expected near-term effects are labor shocks, pressure on junior and entry-level workers, and potentially greater wage inequality, while seniority and taste may offer partial protection. The long-term shape of work remains unpredictable, but the author concludes that “being human” may be the only durable distinction, and that this alone offers limited economic reassurance.