Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Psycho-Pass illustrating Crime Coefficient PerversionScene still / Psycho-Pass

Crime Coefficient Perversion

After being kidnapped and assaulted, Chika’s trauma raises her Crime Coefficient until the Dominator reclassifies the victim herself as eligible for lethal enforcement. Akane refuses to follow the automated recommendation and talks her down.

A proxy score converts distress and vulnerability into apparent dangerousness, then feeds directly into state coercion. Human judgment is needed to recognize that the measurement has confused the consequences of victimization with culpability.

This is discriminatory proxy scoring rather than a learned demographic bias produced by a real training dataset.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Automation Bias
  2. Algorithmic Bias
  3. Power Concentration
  4. Digital Repression