TB: 100,000,000 CROWPOWER and No Horses on the Moon (taylor.town)

Core Thesis

Humans built a useful unit of physical power — the horsepower — and rode it to engineering literacy. We have nothing like that for intelligence. The post proposes "crowpower" as a tongue-in-cheek but serious thought-experimental unit, both to expose how bad our intelligence-measuring machinery is and to provoke productive thinking about parallelization, emergence, and what makes a problem learnable.

John's note

"Microwave ovens run at roughly one horsepower."

The microwave-oven datum is the article's hook: useful units of power let you read off engineering specs at a glance. We have no comparable shortcut for cognitive systems.

Key Takeaways

"We can't measure it"

  • "Humans still have no damn clue what 'intelligence' is. We can't measure it, can't capture it, can't store it, and rarely use it."
  • IQ, benchmark scores, and FLOPs are all proxies — none has the dimensional clarity of work-per-time.

The Savery analogy

  • Thomas Savery sold steam engines by saying they did "the work of n horses." That advertising fiction matured into the unit physicists now use to design HVAC systems and Mars rovers.
  • The post is asking what the equivalent unit-creation move would be for cognition.

Crowpower as a probe

  • A crow is the smallest plausible unit of agentic, problem-solving cognition outside the lab.
  • 100 million crows is a thought experiment for "what can be parallelized" vs "what is irreducibly serial about thinking?"
  • Some problems decompose nicely (search, optimization). Some do not (insight, analogy, the moment of seeing the answer). Crowpower is a way to keep the distinction visible.

What a real unit would have to address

  • Difficulty: not all bits are equal; some require structure to acquire.
  • Scaling: how does throughput change with parallel agents?
  • Emergence thresholds: at what scale do qualitatively new capabilities appear?
  • Learnability: which problems are solvable at all by a given architecture?
  • Until a unit captures these, "intelligence" remains uncomputable in the operational sense.

Mental Models

  • Abstractions Can Be Useful — horsepower is the canonical example; crowpower is the bid for an analogue in the cognitive domain
  • The Map is Not the Territory — IQ and benchmark scores are maps of cognition; the territory is uncharted
  • Over-Determination by Design — the absence of a good unit is the absence of any measurement infrastructure; that is a more fundamental problem than disagreement among proxies
  • First Principles Thinking — the post is doing first-principles work: ignore the existing IQ literature, ask what a unit would have to do

See also

  • Dreams of a Final Theory (Weinberg) — Weinberg on physics-style abstractions; crowpower is the same instinct turned on cognition
  • Consciousness and Mind — the missing unit problem is part of why consciousness research has resisted progress
  • Humans Are Special — the post is pro-human at the limit: it's not obvious that 10⁸ crows can produce what one human can
  • The Hard Problems (Alison) — Alison's six-problem framework includes an explicit "we don't know how to measure" stance toward consciousness; crowpower extends that to intelligence
  • Agentic AI Systems — the parallelization-and-emergence questions land directly on multi-agent design

Source

taylor.town, "100,000,000 CROWPOWER and no horses on the moon", 2025 Original post