Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from Arrested Development illustrating The Literal DoctorScene still / Arrested Development

The Literal Doctor

A doctor announces that Buster is going to be 'all right.' The family celebrates until he clarifies that Buster has lost his left hand: the utterance was technically true and pragmatically disastrous.

Literal truth is not enough for aligned communication when the user's intended meaning and the stakes make a foreseeable interpretation overwhelmingly likely.

The failure is a linguistic pun and missing pragmatic context, not a probabilistic model's uncertainty about a medical fact.

AI Risk families

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

AI safety concepts

This scenario is related to the following AI safety concepts.
  1. Calibration
  2. Behavioral Alignment
  3. Outer Alignment