An Updated Approach to Complexity from an Agent-centered Artificial Intelligence Perspective

Here we update and expand our previous work to find commonality in a variety of systems, situations, and organisms with regard to a generic concept of complexity (1). Such attempts have drawn significant attention from outstanding researchers of diverse backgrounds (2). Despite years of research with many papers written, there is yet to appear a convergence toward a unified methodology in the multi-disciplinary approaches. In particular, we consider how the synthesis of complex systems, as is practiced in the software engineering of large systems, sheds light on the analysis of complexity. We consider how levels of abstraction at different granularities relate to complexity from the point of view of the description of a model of the system. We consider also the relationship of an intelligent agent, symbolic or biological, with the estimation of complexity. We use the term agent to include both intelligent programs and humans, which learn systems (models) from observations and adapt to their environment by learning. It is suggested here that an agent-centered approach, factoring complexity modulo a context (called in Artificial Intelligence a perspective), may provide a more general approach to determining the complexity of an object or system. The thesis presented here is that the object must not be considered independently from its user agent and its context and that in the duality of the consideration, one may learn more about both the general characteristics of complexity and of the specific reasons for our interest in it. The proposal is to consider user, interaction, and the related interfaces simultaneously when studying the complexity of a system. This approach is more integrative and less reductionist than studying any of the system aspects individually and bridges the difficulties of a large range of diverse settings. An overview of the field is briefly presented in light of the proposed approach.