Creativity in Complex Systems


    One of the most attractive features of complex evolutionary systems is their creativity: novel solutions to difficult problems emerge spontaneously. Evolutionary arms races illustrate this phenomenon particularly well. Although computer models have been developed for many complex systems, very few that exhibit the open-ended creativity of evolutionary arms races.
    We argue that the primary reason for this lack of creativity is that the features upon which creativity is typically built exist at a level below that of the issues that the model was intended to examine. For example, for a toxin and anti-toxin to evolve (as in an evolutionary arms race) in a computer model, the biochemical mechanisms that the toxin effects must be built into the model in advance. But normally those mechanisms are at a level significantly below that which the model was intended to examine, which is the external interaction among entities.
    An evolutionary arms race is creative precisely because it exploits features of entities that were not anticipated as being used in the way the arms race found to use them. Evolutionary creativity is difficult to model because the complexity of the mechanisms that would have to be built into the model in advance is typically overwhelmingly large.
    We argue that the only feasible alternative to building inordinately complex physical and chemical details into a model in advance, details which are unlikely to be used in any event, is to allow the evolutionary process to evolve mechanisms that are open ended in some meaningful way. This differs from many evolutionary models in which (a) parameters are allowed to vary but (b) the basic functioning of the agents in the models are set in advance. Genetic programming is one framework that allows such open ended evolution. So far its use has been limited primarily to problem solving in the tradition of genetic algorithms rather than in co-evolutionary agent based models.


Russ Abbott
California State University, Los Angeles
Department of Computer Science
rabbott@calstatela.edu