The Idea of Optimization for Self-Improving Systems

Biological systems have multiple goals integrated within the system. In their interactions with others and the environment, they are constantly solving the SISSY challenge of being Self-Integrating and Self-improving Systems. That is, as is expected for SISSY systems, biological systems do not just establish fixed interfaces with new components or new relationships and configurations among established components / entities but rather they must constantly alter relationships and configurations to match shifting goals and the largely uncontrollable demands of dynamic environments. In this exploratory paper, we point out some of the characteristics of these biological systems that impact how we set up the optimization problems we use to have our systems reach objectives and to assess what improvement might mean. The purpose of this paper is to invite a community-wide discussion on what biological-like solutions are feasible for computational systems and what we may or may not want to adopt for our SISSY systems. We start with a discussion of the biological properties of multi-goal systems and discuss a representative method to model multiple goals in multi-agent systems, using the sustainable foraging problem as the focus. We then discuss some of the mathematical challenges of adopting a more biological approach which commonly has massively parallel systems that are always multi-goaled and operating in diverse timescales. We conclude with some suggestions for how to start very different patterns of “optimization” for satisfying objectives and achieving incremental improvement.