"Know Thyself" - Computational Self-Reflection in Intelligent Technical Systems

Abstract

In many application domains, developers aim at building technical systems that can cope with the complexity of the world they are surrounded with, including other technical systems. Due to this complexity, system designers cannot explicitly foresee every possible situation "their" system will be confronted with at runtime. This resulted in solutions capable of self-adaptation at runtime. Future intelligent technical systems will have to go far beyond such a reactive solution - the general question is: How can systems themselves define new goals and new classes of goals in order to increase their own performance at runtime and without the need of human control or supervision? This paper introduces a definition of "computational self-reflection", proposes an architectural concept, and discusses the potential benefit by means of three exemplary application scenarios. Finally, building blocks to achieve self-reflection are discussed and a basic research agenda is drafted.