Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Hot Fuzz illustrating Village of the YearScene still / Hot Fuzz

Village of the Year

01Scenario

The Neighbourhood Watch Alliance murders residents whose behavior might cost Sandford the Village of the Year award, then disguises the killings as accidents.

02AI safety Analogy

A proxy for community quality becomes the true objective. The organization improves its score by eliminating inconvenient evidence and people rather than improving the community.

03Where the analogy breaks

The optimizing system is a cult-like group of humans, not an automated agent.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Reward Hacking
  2. Goodhart’s Law
  3. Governance Failure
  4. Specification Gaming

Taxonomy neighbors

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

  1. Still from The Sopranos illustrating The HUD Housing Scam

    The Sopranos2002

    The HUD Housing Scam

    Risk
    Malicious use · Security · Systemic
    Concept
    Goodhart’s Law · Governance Failure · Specification Gaming
  2. Still from District 9 illustrating Cat-Food Consent

    District 92009

    Cat-Food Consent

    Risk
    Security · Systemic
    Concept
    Goodhart’s Law · Governance Failure · Specification Gaming
  3. Still from Family Guy illustrating Customer of the Week

    Family Guy2021

    Customer of the Week

    Risk
    Malicious use
    Concept
    Reward Hacking · Goodhart’s Law · Specification Gaming