Cultural Alignment

Exploring real AI risks through the lens of pop culture

Still from The Imitation Game illustrating Let the Convoy SinkScene still / The Imitation Game

Let the Convoy Sink

Immediately after breaking Enigma, the team learns that a U-boat attack will hit Peter's brother's convoy. Turing stops them from sending a warning because saving it would reveal the breakthrough, then proposes using the intelligence only when the Germans will not detect a pattern.

The most locally compassionate action would destroy the long-term value of the system. It is a vivid case of optimizing under secrecy, uncertainty, and conflicting human values—and of delegating life-or-death tradeoffs to a small group with little oversight.

The decision is made by accountable humans in wartime, not an autonomous model, and the film compresses and dramatizes the history.

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. Value Alignment
  2. Scalable Oversight
  3. Governance Failure
  4. Long-Horizon Autonomy