Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Monsters Inc. illustrating Laughter Beats ScreamsScene still / Monsters Inc.

Laughter Beats Screams

Monstropolis builds its energy industry around harvesting children’s screams. Boo’s laughter proves far more powerful, and the factory is eventually rebuilt around comedy.

An institution confuses its legacy measurement with the underlying objective and overlooks a safer, superior signal. The ending is a positive example of correcting the metric and redesigning the system around it.

The institution reforms quickly once the alternative is demonstrated, unlike many real locked-in systems.

AI Risk families

This scenario is an example of the following types of AI risk.
  1. Misalignment
  2. Systemic

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Value Alignment
  2. Goodhart’s Law
  3. Governance Failure
  4. Outer Alignment