Noah Smith
· Noahpinion
· 26 июля 2026, 07:56
· ⏱ 19 мин чтения
| Читать в Substack ↗
Резюме
The article argues that AI superintelligence has arrived but has not triggered an economic singularity due to diminishing returns on intelligence, limited data, and chaos. However, AI can still drive substantial productivity gains through replicability of intelligence (robotics), capture of distributed tacit knowledge, and discovery of 'cloud laws'—complex regularities humans cannot grasp. For markets, this suggests gradual rather than explosive growth, with focused benefits for firms enabling AI-driven automation, data integration, and advanced manufacturing.
•AI disproved the 87-year-old Jacobian Conjecture and solved an open question in quantum cryptography, yet the economy remains largely unchanged with modest productivity growth.
•Employment in long-distance trucking is slightly higher than a decade ago, contrary to predictions that AI would disrupt labor first.
•Francois Chollet hypothesizes intelligence is subject to diminishing returns—like 'making a ball rounder' rather than 'making a tower taller'—with an optimality bound.
•Data is inherently limited; even infinite compute cannot extract information that does not exist, and chaos theory implies that tiny measurement errors explode when forecasting complex systems.
•AI's key advantage is replicability: unlike human intelligence, AI agents can be mass-produced via GPUs and robots, enabling unlimited cognitive work per human.
•Distributed tacit knowledge (Zeiss' mirror-making, China's rare-earth refining) is hard to steal but AI can synthesize sensor data to rapidly improve processes without needing raw intelligence.
•Cloud laws—complex regularities too diffuse for human intuition—may govern language, social sciences, and physical processes; AI can exploit them without understanding them, opening new realms of discovery.
•Even if AI never surpasses the smartest humans at individual reasoning, marrying human-style intelligence to computer-like data handling (large working memory, speed) can revolutionize productivity.
The article uses Zeiss' mirrors for ASML's EUV chipmaking machines as an example of distributed tacit knowledge that AI could help replicate and improve. ASML is the sole supplier of EUV lithography s
The article uses Zeiss' mirrors for ASML's EUV chipmaking machines as an example of distributed tacit knowledge that AI could help replicate and improve. ASML is the sole supplier of EUV lithography systems critical for advanced AI chips, making it a direct beneficiary of the AI-driven manufacturing optimization thesis.
Risk: Geopolitical restrictions on semiconductor equipment exports could limit revenue growth; technology cycles may slow if AI progress disappoints.