|
EPSRC Network on Evolvability in Biology & Software SystemsSoftware Evolution and Evolutionary Computation Symposium Abstracts
University of Hertfordshire, Hatfield, U.K.
|
CHRISTOPHER LANDAUER KIRSTIE L. BELLMAN
Aerospace Integration Science Center
The Aerospace Corporation, Mail Stop M6/214
P.O.Box 92957
Los Angeles, California 90009-2957, USA
cal@aero.org, bellman@aero.org
In this talk, we will combine our own studies of integration infrastructure, Computational Semiotics, and intelligent system architectures to describe an approach to constructing self-modifying computing systems that aims to support many of the important robustness and adaptibility properties of biological systems, while providing an excellent testbed for studying the most mysterious of those properties (the creation of new functionality).
We will begin by discussing some of the differences between evolutionary processes in biological systems, evolutionary programming approaches to computing complicated functions, and software evolution, or maintenance of computational systems. We will explain why we believe that biological evolution is such a remarkable process, from which we might be able to learn many important principles of resilience and viability for Constructed Complex Systems, which are large heterogeneous systems managed or mediated by computing systems. We will then describe several study directions of our prior research, and show how they address many important issues in the understanding of how evolution might help us develop more effective Constructed Complex Systems.
First, we describe our Wrapping approach to integration infrastructure, and show how it provides many of the desirable properties of evolutionary systems. The Wrapping approach is a computationally reflective, knowledge-based, integration infrastructure for Constructed Complex Systems, which has been shown to be a very flexible method for integrating both newly created software and also legacy software and systems. We show how the Problem Posing Programming Paradigm unifies diverse programming and modeling paradigms, and provides a uniform method of communication and interaction among all components of a Constructed Complex System, and therefore allows very different styles of processing to occur in such a system.
Next, we turn to Computational Semiotics, which is our term for the study of the use of symbols and symbol systems by computing systems. We explain our ``Get Stuck'' theorems, show how they focus our attention on the symbol systems and representational mechanisms that are used both to represent and to cause system behavior, and also to define and interpret those same symbol systems and representational mechanisms.
Then, we describe the architecture of a system that is entirely explicitly self- and model-defined, in that all parts of the system have explicit models that are accessible to and interpretable by the system. This property means that the system has a complete model of its own behavior (down to some level of detail), which allows the system to make informed judgments about its own behavior in its execution context, where that behavior may be inadequate, and what kinds of changes may result in better performance.
Finally, we engage in some informed speculation about how our approaches and
methods might help us to address the most mysterious of the properties of
biological evolution, which is the development of entirely new structures and
capabilities, loosely contained in the notion of the evolution of complexity.
Some part of this work is speculative, but most of our conclusions are based on actual systems that the authors have built or designed. Demonstrations of some of these systems will be available at the conference.