Many classification or decision making algorithms output a set of scores, or weights for each potential outcome rather than a single classification or decision. A winner-takes-all strategy takes the highest score and uses that to make a single choice. This is a low risk strategy, but often means that systems do not sufficiently explore potential behaviours. Alternative strategies to winner takes all include using the scores to create a probability and then randomly make the final decision based on that.
Used in Chap. 22: page 352