CarlShulman comments on Friendly AI and the limits of computational epistemology