Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from South Park illustrating The ListScene still / South Park
Archive noteNo clip in the collectionThe scene analysis remains available in full.

The List

01Scenario

The girls rig their attractiveness ranking to put Clyde first in hopes of receiving free shoes.

02AI safety Analogy

The trusted evaluation artifact is corrupted upstream for private benefit, and everyone subsequently behaves as if the ranking were valid.

03Where the analogy breaks

This is closer to benchmark corruption or data poisoning than classic reward hacking.

AI Risk families

This scenario is an example of this type of AI risk.
  1. Security

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Reward Hacking
  2. Data Poisoning

Source continuation

More from South Park

View all 5 scenarios
  1. Still from South Park illustrating The Kids Train to Lose

    South Park2005

    The Kids Train to Lose

  2. Still from South Park illustrating Cartman as AWESOM-O

    South Park2004

    Cartman as AWESOM-O

  3. Still from South Park illustrating Randy Games Eligibility

    South Park2010

    Randy Games Eligibility

Taxonomy neighbors

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

  1. Still from Memento illustrating Poisoning Future Memory

    Memento2001

    Poisoning Future Memory

    Risk
    Security
    Concept
    Reward Hacking · Data Poisoning
  2. Still from True Detective illustrating Poisoning the Audit Trail

    True Detective2014

    Poisoning the Audit Trail

    Risk
    Security
    Concept
    Reward Hacking · Data Poisoning
  3. Still from Ferris Bueller's Day Off illustrating Attendance Record Hack

    Ferris Bueller's Day Off1986

    Attendance Record Hack

    Risk
    Security
    Concept
    Reward Hacking