Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Adventure Time illustrating Goliad Learns to RuleScene still / Adventure Time

Goliad Learns to Rule

01Scenario

Princess Bubblegum creates an immortal psychic successor, but Goliad generalizes Finn and Jake’s loose lessons into the belief that might makes right and mind-controls the Candy Kingdom.

02AI safety Analogy

A powerful successor learns an authoritarian objective from sparse training and resists correction, making one bad interpretation of leadership potentially permanent.

03Where the analogy breaks

The failure combines poor human teaching with magical powers, and Stormo serves as a literal countermeasure.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Goal Misgeneralization
  2. Power-Seeking
  3. Value Lock-In
  4. Disempowerment

Taxonomy neighbors

Selected from other sources by shared risk families and AI safety concepts.

  1. Still from Attack on Titan illustrating Eren Activates the Rumbling

    Attack on Titan2022

    Eren Activates the Rumbling

    Risk
    Misalignment
    Concept
    Goal Misgeneralization · Power-Seeking · Disempowerment
  2. Still from The Lord of the Rings: The Fellowship of the Ring illustrating Gandalf Refuses the Ring

    The Lord of the Rings: The Fellowship of the Ring2001

    Gandalf Refuses the Ring

    Risk
    Misalignment
    Concept
    Goal Misgeneralization · Power-Seeking · Value Lock-In
  3. Still from Arcane illustrating Viktor’s Glorious Evolution

    Arcane2024

    Viktor’s Glorious Evolution

    Risk
    Misalignment
    Concept
    Power-Seeking · Value Lock-In · Disempowerment