The dangers in algorithms learning humans' values and irrationalities

Rebecca Gorman, Stuart Armstrong

For an artificial intelligence (AI) to be aligned with human values (or human preferences), it must first learn those values. AI systems that are trained on human behavior, risk mi…