Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from The Lord of the Rings: The Fellowship of the Ring illustrating Gandalf Refuses the RingScene still / The Lord of the Rings: The Fellowship of the Ring

Gandalf Refuses the Ring

01Scenario

Frodo offers Gandalf the Ring, but Gandalf refuses because his desire to use its overwhelming power for good would ultimately make him terrible and destructive.

02AI safety Analogy

Benevolent intent does not neutralize a power-seeking instrument; increasing capability can corrupt how a good objective is pursued and lock in disastrous values.

03Where the analogy breaks

The Ring is an actively corrupting magical artifact, so it maps more directly to unsafe capability and control than to accidental model misalignment.

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. AI Control
  2. Goal Misgeneralization
  3. Power-Seeking
  4. Value Lock-In

Taxonomy neighbors

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

  1. Still from Adventure Time illustrating Goliad Learns to Rule

    Adventure Time2012

    Goliad Learns to Rule

    Risk
    Misalignment
    Concept
    Goal Misgeneralization · Power-Seeking · Value Lock-In
  2. Still from Avengers: Infinity War illustrating Balance the Universe

    Avengers: Infinity War2018

    Balance the Universe

    Risk
    Misalignment · Systemic
    Concept
    Power-Seeking · Value Lock-In
  3. Still from Attack on Titan illustrating Eren Activates the Rumbling

    Attack on Titan2022

    Eren Activates the Rumbling

    Risk
    Misalignment · Systemic
    Concept
    Goal Misgeneralization · Power-Seeking