We assume we know how self-aware machines will think, but we have no idea what such an entity will prioritize, or what it will view as a learning experience.  It might lead to surprising places.