Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Harry Potter illustrating The Mirror Protects the StoneScene still / Harry Potter

The Mirror Protects the Stone

Quirrell cannot retrieve the Philosopher’s Stone; Harry sees his reflection pocket it and it appears on him. Dumbledore later explains that only someone wanting to find the Stone but not use it could obtain it.

Access control is keyed to latent intent rather than a stated request: the dangerous capability is released only when the user’s objective is compatible with non-use.

The mirror has magical, perfect mind-reading and no false positives; Dumbledore’s rule is explained only afterward.

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. Capability Evals
  2. AI Control
  3. Knowledge Elicitation
  4. Behavioral Alignment