Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Shaun of the Dead illustrating Pretending to Be ZombiesScene still / Shaun of the Dead

Pretending to Be Zombies

01Scenario

Shaun’s group imitates zombie gait, posture, and noises to pass through a horde without being recognized as living humans.

02AI safety Analogy

The humans alter superficial features to match the accepted category and evade a detector without changing the underlying thing being classified.

03Where the analogy breaks

The detector is animal perception rather than a trained classifier, and the disguise breaks under stress.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Adversarial Robustness
  2. Adversarial Evasion

Taxonomy neighbors

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

  1. Still from Predator illustrating Predator vs Dutch

    Predator1987

    Predator vs Dutch

    Risk
    Malicious use · Security
    Concept
    Adversarial Robustness · Adversarial Evasion
  2. Still from Psycho-Pass illustrating The Helmets Fool the Dominators

    Psycho-Pass2013

    The Helmets Fool the Dominators

    Risk
    Malicious use · Security
    Concept
    Adversarial Robustness · Adversarial Evasion
  3. Still from The Mitchells vs. the Machines illustrating Dog, Pig, Loaf

    The Mitchells vs. the Machines2021

    Dog, Pig, Loaf

    Risk
    Security
    Concept
    Adversarial Robustness · Adversarial Evasion