A new study shows that chess discussions of "black," "white," "attack," and "threat" can trigger red flags from hate speech algorithms.