Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Brooklyn Nine-Nine illustrating The Dentist’s Ego

The Dentist’s Ego

01Scenario

Jake and Holt stop trying to directly prove the murder and instead construct an interrogation that exploits Philip Davidson’s need to be seen as the smartest person in the room, prompting him to reveal hidden knowledge.

02AI safety Analogy

A tailored adversarial evaluation can elicit capabilities or knowledge that ordinary questioning leaves concealed.

03Where the analogy breaks

This is a coercive criminal interrogation, not a benign model evaluation.

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. Capability Evals
  2. Knowledge Elicitation
  3. Red Teaming
  4. Situational Awareness

Source continuation

More from Brooklyn Nine-Nine

View all 3 scenarios
  1. Still from Brooklyn Nine-Nine illustrating Proposal in the Heist

    Brooklyn Nine-Nine2017

    Proposal in the Heist

  2. Still from Brooklyn Nine-Nine illustrating Dozerpads

    Brooklyn Nine-Nine2015

    Dozerpads

Taxonomy neighbors

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

  1. Still from Willy Wonka & the Chocolate Factory illustrating The Gobstopper Test

    Willy Wonka & the Chocolate Factory1971

    The Gobstopper Test

    Risk
    Security
    Concept
    Capability Evals · Red Teaming · Situational Awareness
  2. Still from Guardians of the Galaxy Vol. 2 illustrating Baby Groot’s Death Button

    Guardians of the Galaxy Vol. 22017

    Baby Groot’s Death Button

    Risk
    Security
    Concept
    Capability Evals · Knowledge Elicitation
  3. Still from Frieren: Beyond Journey’s End illustrating Frieren Suppresses Her Mana

    Frieren: Beyond Journey’s End2023

    Frieren Suppresses Her Mana

    Risk
    Security
    Concept
    Capability Evals · Situational Awareness