We're really working hard on opening Pandora's box here, but it is so interesting and probably too late to stop, so...onward!
One thing I see is a lot of concern for matching the values of diverse users and minorities. It sounds nice, but it is not really possible. You can't even get agreement inside of the United States on values let alone places like China, the Muslim world, Russia and Africa. You probably wouldn't enjoy dealing with an AI programmed to the values and preferences of the Iranian mullahs, for example. They feel the same way about us, of course.