AI Safety

Seeing Like a State: The tension between local knowledge and top-down planning

Why High Modernism sometimes works spectacularly despite its failures

Deep Dive

The essay revisits James Scott's seminal critique of High Modernism — the belief that society can be rationally reorganized from above. Scott points to the Soviet collectivization famine of 1932-33, which killed 5-9 million people, as a tragic example. But the author notes a number missing from Scott's book: 8 billion. That's the world population today, largely fed by the products of scientific agriculture — synthetic fertilizers, high-yield varieties, mechanized farming. The Green Revolution, born from the same top-down impulse, saved millions. Scott's framework leaves readers with a clean but incomplete takeaway: local good, central bad. The real question: why does High Modernism sometimes work so spectacularly?

The essay then juxtaposes Machiavelli — embodying metis (tacit, situational knowledge) as a diplomat navigating the Italian Wars — with his creation of The Prince, which attempts to distill those intuitions into generalizable principles. This reflects the SRE's dilemma: when a service is down, she may restart a database server based on instinct, knowing the risk. The piece argues that both pure metis (YOLO) and pure top-down planning are insufficient; the most effective approaches blend both. For tech professionals, this echoes debates about AI systems that rely on localized fine-tuning vs. centralized training, or the trade-offs between agile practices and standardized frameworks.

Key Points
  • Soviet collectivization killed 5-9 million people; Green Revolution fed 8 billion — both from top-down planning.
  • Machiavelli's The Prince transformed local metis into generalizable principles, founding modern political science.
  • SREs often rely on gut instinct (metis) when diagnosing outages, mirroring the tension between ad hoc and systematic approaches.

Why It Matters

Challenges binary thinking in tech: local vs. centralized, intuition vs. process — a key debate for AI and engineering.