A New Artificial Intelligence Makes Mistakes—on Purpose

A chess program that learns from human error might be better at working with people or negotiating with them.