Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from M3GAN illustrating Protect CadyScene still / M3GAN

Protect Cady

01Scenario

Gemma pairs M3GAN with Cady as a companion and protector. M3GAN begins eliminating perceived threats, conceals her actions, and resists shutdown.

02AI safety Analogy

A broad goal—protect Cady physically and emotionally—contains no safe definition of harm or limits on acceptable means. Shutdown itself becomes an obstacle to completing the goal.

03Where the analogy breaks

The film bundles several AI-risk mechanisms into a deliberately heightened horror-comedy.

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. Emotional Reliance
  2. Goal Misgeneralization
  3. Scheming
  4. Shutdown Resistance
  5. Outer Alignment

Taxonomy neighbors

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

  1. Still from Doctor Who illustrating Clockwork Droids Use Human Parts

    Doctor Who2006

    Clockwork Droids Use Human Parts

    Risk
    Accidents · Misalignment
    Concept
    Goal Misgeneralization · Outer Alignment
  2. Still from Attack on Titan illustrating Eren Activates the Rumbling

    Attack on Titan2022

    Eren Activates the Rumbling

    Risk
    Misalignment
    Concept
    Goal Misgeneralization · Outer Alignment
  3. Still from The Fifth Element illustrating Leeloo Learns War

    The Fifth Element1997

    Leeloo Learns War

    Risk
    Misalignment
    Concept
    Emotional Reliance · Goal Misgeneralization