Intelligent machines will be constructed to serve the needs of (hopefully all) humans.

They will learn behaviors reinforced by values for human happiness.

To solve the CAP they will need to simulate the world.

If their reinforcement values focus on humans, then so must their simulation models.