Classifier systems are an intriguing idea that, so far, has been hard to make work. The difficulties fall into three areas that I call the Discovery Problem, the Cooperation Problem, and the Time Problem. Here I will address the Discovery Problem, but the others deserve some mention. The Cooperation Problem arises because in a classifier system, classifiers must often cooperate ---and yet are
... [Show full abstract] also in competition under the GA. Classifiers may cooperate diachronically, forming chains along which payoff flows. Each member of a chain is dependent on the others: the early on the late and vice versa. So if the success of one member leads it to reproduce to the detriment of another member, it in effect shoots itself in the foot. Classifiers also may cooperate synchronically, sharing in the proper representation of a situation at one time. The same dilemma holds: a classifier's ultimate fitness may well depend on its contemporaries also being there---yet GA competition acts against this. Coope...