They should just work with local medical ethics boards at Stanford.

AI interactions with humans could be considered a class of medical experimentation as a thought experiment. Would have interesting output, and the system already exists...