Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Neon Genesis Evangelion illustrating The MAGI VoteScene still / Neon Genesis Evangelion
Archive noteNo clip in the collectionThe scene analysis remains available in full.

The MAGI Vote

01Scenario

The adaptive Angel Ireul infects NERV's network and compromises two of the three personality-derived MAGI computers, but cannot trigger headquarters' self-destruction without the third system's vote.

02AI safety Analogy

Evaluator diversity and a high authorization threshold slow a catastrophic cyberattack, while all three evaluators deriving from one person exposes hidden correlated risk.

03Where the analogy breaks

The three MAGI are not genuinely independent, and the episode's adaptive biological-computer mechanics are deliberately speculative.

AI Risk families

This scenario is an example of the following types of AI risk.
  1. Accidents
  2. Security
  3. Systemic

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Deployment Safety
  2. Value Alignment
  3. Scalable Oversight
  4. Self-Replication
  5. Algorithmic Monoculture
  6. Cyberattacks

Taxonomy neighbors

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

  1. Still from Ralph Breaks the Internet illustrating Arthur Escapes

    Ralph Breaks the Internet2018

    Arthur Escapes

    Risk
    Accidents · Security · Systemic
    Concept
    Self-Replication · Algorithmic Monoculture · Cyberattacks
  2. Still from Independence Day illustrating Independence Day Virus

    Independence Day1996

    Independence Day Virus

    Risk
    Security · Systemic
    Concept
    Deployment Safety · Algorithmic Monoculture · Cyberattacks
  3. Still from Barbie illustrating Recursive Deprogramming

    Barbie2023

    Recursive Deprogramming

    Risk
    Security · Systemic
    Concept
    Value Alignment · Scalable Oversight · Algorithmic Monoculture