http://starcat.sdsu.edu - UNDER CONSTRUCTION
We at the Science of Evolving, Adaptive Systems (SEAS) lab at San Diego State University are focused on the development of software which produces ongoing behavior that adapts to a continually changing environment and that evolves appropriate new functionality over the course of its interaction in that environment. There are several characteristics desired from the systems we develop; we follow Steels [1996] in our articulation of these. First, we wish to create evolving, adaptive systems which exhibit self-maintenance. The ongoing activity of such a system should include tending to its own functioning and the adjustment of its parameters that support its continued behavior in an environment. Second, we seek to build systems that are adaptive, which is to say that the particular behaviors manifest by the system are in constant flux, as dictated by the needs of the system in its current environment. We believe that the ability to interact with an environment and be affected by the consequences of behavior in that environment is essential for success in building evolving, adaptive systems. The models we develop must be embodied in this sense. Third, we intend our systems to demonstrate information preservation. This can be stated as the capability for representational behavior (among the other system behaviors); however, we insist that representation is a side-effect of behavior--an emergent consequence of the system's internal dynamics which can be used to guide further system activity--rather than a static entity that attempts to imbue the system with a priori domain knowledge. To say that information is preserved in such a system is to expect persistence over time of certain of those dynamic structures left behind by the system's activity. These persistent, emergent, dynamic structures can be coupled to the environment through the behavior they foster, and we call this behaviorally coupled representation. Finally, and most challenging, in our developed systems we hope to see the same kinds of spontaneous increase in complexity exhibited by complex systems in nature. We believe that systems that realize these characteristics will necessarily be complex systems, by which we mean precisely the following: that their behavior arises from the local interactions of a large number of interacting agents. We are interested in discovering systems capable of self-generation of symbolic activity--the pairing of emergent structures with specific features or phenomena in the environment and the associated behaviors. Furthermore, we note that natural systems that exhibit these characteristics are autopoeitic--meaning that the environment can trigger changes in behavior, but the particular shape of those changes derives from the internal dynamics of the system when impinged upon by the environment and is not directly selected thereby. Among other things, this gives these systems individuality through their history of interaction with their surroundings. We draw on a growing body of research that supports these notions, including that of Mitchell [1993] and Hofstadter [1995], Maturana and Varela [1980], Clark [1997], Holland [1986], Prigogine [1984]. We believe it is important to realize the models developed from these ideas in testable software for a variety of actual problem domains in order to refine and develop further the science of evolving, adaptive systems.