A method known as "model pruning" can be used to pinpoint and remove neurons that consistently contribute to biased responses ...