Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from The Hunger Games illustrating The Nightlock BerriesScene still / The Hunger Games

The Nightlock Berries

When the Gamemakers revoke the two-victor rule, Katniss and Peeta threaten to eat nightlock berries and leave the Capitol with no winner, forcing it to declare them both victors.

The contestants find a technically available outcome that the evaluator cannot tolerate and use it to force a change in the reward rule.

Human contestants deliberately exploit an oppressive institution rather than an AI exploiting its designers.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Metagaming
  2. Contestability
  3. Specification Gaming
  4. Situational Awareness