Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Lilo & Stitch illustrating Stitch Learns 'OhanaScene still / Lilo & Stitch

Stitch Learns 'Ohana

Experiment 626 is engineered as an ultimate weapon, but living with Lilo changes what he values. He voluntarily protects the family and comes to understand himself as more than a destructive tool.

Robust prosocial behavior emerges through relationships, belonging, and moral learning rather than a stronger control chip; treating intelligence solely as a weapon also helps create the conflict.

Stitch is a biological, highly anthropomorphic moral agent, and family attachment is not a scalable technical alignment method.

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. Goal Misgeneralization
  3. AI Welfare
  4. Behavioral Alignment