Babel.md

Exploring the boundaries of human cognition in the age of AI

Lang 中文

Mar 15, 2026

阅读中文

The Mule Problem: How LLMs Are Crushing Cognitive Diversity

When AI models the average of human thought, what happens to the outliers who drive civilization forward?

llm cognition asimov alignment

In Isaac Asimov’s Foundation universe, the Mule isn’t dangerous because he’s strong. He’s dangerous because he’s weird: a statistical outlier with leverage over history.

Modern LLMs, by design, are the opposite: they approximate a center of mass of text and taste. That is a superpower, and also a pressure.

The question

If our tools increasingly reflect what is most likely to be said, what happens to what is worth saying but rarely said?

A concrete failure mode

  • “Seems reasonable” becomes a local optimum
  • “Useful” becomes “average”
  • novelty gets pushed to the margins

What I want from this blog

This is a bilingual notebook on AI and human minds. I care about:

  • cognitive diversity
  • agency, attention, and incentives
  • demos that make abstract claims falsifiable
// Placeholder snippet: more interactive experiments will come later.
export const mule = (x: number) => x * 2 + 1;

Newsletter

Low frequency. Only the pieces that survive a second thought.

No spam. No selling. Unsubscribe anytime.