Reflection Processes Help Integrate Simultaneous Self-Optimization Processes

Biological systems have evolved many disparate mechanisms for continual local adaptive improvement of their own behaviors in their various ecological niches (operational environments). If we are to build engineered systems with anything like the robustness and flexibility exhibited by biological systems, then we need to examine their self-improvement processes in detail, and determine which of them are feasible for us to implement in our engineered systems. In this paper, we concentrate on optimization, and note simply that optimization is hard in an uncertain dynamic environment, self-optimization is harder, and that even once we have a repertoire of local optimizations, we still have the problem of reconciling their results to some kind of global behavior decision. We suggest that computational reflection is a method for computational systems to gain the perspective needed to integrate the results of different local optimization decisions. We argue that it provides an initial strategy for reconciling different local selfoptimization processes by feeding them into a semantically rich set of reflection processes. The purpose of this paper is to stimulate more communitywide discussion about the problems of reconciling multiple selfoptimization processes within a complex engineered system.