What if the AI does exactly what its owner wants, but harms everyone else? The kill switch works, the owner is in control, and they are happy with the results. What would make them press it??