1. Human level AI requires the ability of the agent to reason about its past, present, future and hypothetical choices.
2. What an agent can do is determined by its environment rather than by its internal structure.
3. Having choices is usefully distinguished from the higher capability of knowing about them.
4. What people can do and know about what they can do is similar to what robots can do and know.
AI needs a more developed formal theory of free will, i.e. the structures of choice a robot can have and what it can usefully know about them.