What are some of the most carefully thought-out objections to the need for AI alignment?