An extension of genetic algorithms in which the population consists of a co-operating set of rules (i.e. a rulebase) which are to learn to solve a problem given a number of test cases. Between each generation the population as a whole is evaluated and a fitness is assigned to each rule using the bucket-brigade algorithm or other credit sharing scheme (e.g. the Pitt scheme). These schemes aims to reward or punish rules which contribute to a test case according to how good the total solution is by adjusting the individual rules fitness. At the end of the test data a new generation is created using a genetic algorithm as if each rule were independent using its own fitness (measures may be taken are taken to ensure a given rule only appears once in the new population).