|
|
|
|
|
|
|
theory, but without theory all is lost. The broadest hope is that the simulations will suggest well-informed conjectures that offer new directions for theory, while the theoretician can test deductions and inductions against the simulations. Only then can we fully reincarnate, for complex adaptive systems, the cycle of theory and experiment that is so fruitful for physics. |
|
|
|
|
|
|
|
|
While interesting models of complex adaptive systems can be built with classifier systems, and classifier systems have indeed been used for this purpose (Marimon et al. 1989), there is an advantage to having a simpler model that places the interactions in a simpler setting, giving them sharper relief. Echo is one such model, properly a class of models, designed primarily for gedanken experiments rather than precise simulations. Echo provides for the study of populations of evolving, reproducing agents distributed over a geography with different inputs of renewable resources at various sites (see Figures 17 and 18). Each agent has simple capabilitiesoffense, defense, trading, and mate selectiondefined by a set of "chromosomes." Though these capabilities are simple, and simply defined, they provide for a rich set of variations illustrating the four kernel properties of complex adaptive systems previously described. Collections of agents can exhibit analogues of a diverse range of phenomena, including ecological phenomena (e.g., mimicry and biological arms races), immune system responses (e.g., interactions conditioned on identification), evolution of metazoans (e.g., emergent hierarchical organization), and economic phenomena (e.g., trading complexes and the evolution of "money"). |
|
|
|
|
|
|
|
|
A precise description of Echo begins with definition of the individual agents (see Figure 19). The capacities of an agent are completely determined by a small set of strings, the "chromosomes," defined over a small finite alphabet. In the simplest Echo model, this alphabet consists of four letters {a,b,c,d}, called resources,and there are just two classes of chromosomes, tag chromosomes and condition chromosomes. The tag chromosomes determine the agent's external, phenotypic characteristics, and the condition chromosomes determine an agent's responses to the phenotypic characteristics of other agents. |
|
|
|
|
|
|
|
|
There are just three tag chromosomes in the simplest model: (1) offense tag, (2) defense tag, and (3) mating tag. It is convenient to think of the tags as displayed on the exterior of the agent, counterparts of the signature groups of an antigen or the trademarks of an organization. These tags are a kind of identifying address, quite similar to the tags employed by classifier systems. There are also just three condition |
|
|
|
|
|