Representative Social Choice: From Learning Theory to AI Alignment