Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Naruto: Shippuden illustrating Infinite TsukuyomiScene still / Naruto: Shippuden

Infinite Tsukuyomi

Madara reflects the Rinne Sharingan off the moon, trapping nearly everyone in a personalized ideal dream while their unconscious bodies are bound to the God Tree.

Universal personalized satisfaction removes agency, consent, and contact with reality, permanently fixing human preferences instead of resolving human problems.

This is explicit supernatural coercion by a villain rather than AI-mediated reward manipulation.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Reward Hacking
  2. Privacy Loss
  3. Value Lock-In
  4. Disempowerment