New hot field: applied moral philosophy. A lot of AI alignment is based on the people who are using AI systems today & their preferences (what answers they like or don’t). This interesting paper argues that using a Rawlsian Veil of Ignorance works better.