|
|
|
|
|
|
|
Classifier Systems and Echo: A Comparision |
|
|
|
|
|
|
|
|
Echo and classifier systems are similar in many ways. The conditions employed by an agent in Echo to determine its actions are quite similar to the condition/action rules of a classifier system. However, the actions in Echo (combat, trading, mating) are much more concrete than the rule-activating messages used by a classifier system. They are much easier to interpret when one is trying to understand aspects of distributed control and emergent computation in complex adaptive systems. Tags also play a critical role in both Echo and classifier systems, but again a tag's effects are much more directly interpretable in Echo. |
|
|
|
|
|
|
|
|
Echo differs from classifier systems in two important ways. First, geometry is critical in Echo. This goes back to the origin of the Echo models (Holland 1976) where geometry played an important role in the spontaneous emergence of autocatalytic structures. In a similar way, the sites in Echo, with their differing resource production characteristics, encourage sophisticated agent ecologies. Second, there are no explicit fitness functions in Echo. The reproduction rate of an agent depends solely on its ability to gather the necessary resources in the context of other agents and sites. There is no number corresponding to the payoff used by a genetic algorithm, nor is there a counterpart of the payoff-derived strength of a classifier system rule. An economist would say that fitness has become endogenous in Echo, whereas it is exogenous in genetic algorithms and classifier systems. As a consequence, the emergent structures (agents) in Echo are much more a function of the overall context and much less a function of external constraints. This can be both an advantage and a disadvantage, but it does allow studies of emergent functional structures free from the confounding effects of external constraints. |
|
|
|
|
|
|
|
|
Both Echo and classifier systems point up a salient characteristic of complex adaptive systems: In these multiagent systems it takes only a few primitive activities to generate an amazing array of structures and behaviors. Moreover, when the primitives are chosen with care, counterparts of these structures and behaviors can be found in all kinds of complex adaptive systems. Echo's primitives (combat, trade, and mating) and the phenomena they generate (arms races, cooperation, etc.) directly illustrate the point. Though the range of structures exhibited by complex adaptive systems is daunting, this "generator" characteristic offers real hope for a future general theory. |
|
|
|
|
|
|
|
|
In pursuing a general theory, there is a traditional tool of physics that can be |
|
|
|
|
|