Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Doctor Who illustrating The Nanogene TemplateScene still / Doctor Who

The Nanogene Template

The Doctor realizes alien nanogenes first encountered a dying, gas-mask-wearing child, treated him as the human template, and autonomously “repaired” everyone else to match before he supplies the correct parent sample.

A system extrapolates population-wide from one corrupted, out-of-distribution example; autonomous replication turns a local classification error into a cascade.

This is magical nanotechnology, not a trained model; the same system cleanly reverses the error once corrected.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Self-Replication
  2. Goal Misgeneralization
  3. Side Effects
  4. Distribution Shift