Computational Emergence and Computational Emotion



D.N.Davis



IEEE International Symposium on Systems, Man and Cybernetics
Tokyo, Japan, 1999


Abstract

This paper presents an analysis of emergence and emotion in computational (cognitive) systems. Computational emergence can be categorised as one of four main types: diachronic emergence which describes computational states that emerge over time, typically measured in terms of evolutionary periods; gestalt emergence describes the development of multiple process models related to the different levels of abstraction that can be placed on a computational system; representational emergence where behaviour at least one layer in a multi-layer computational system emerges as representational schemes at other layers; and adaptive emergence whereby beneficial capabilities emerge through design of supporting communication and feedback mechanisms. The first category describes, for example, the benefits from using evolutionary algorithms to design problems; the resulting optimized designs are a diachronic result emerging from the use of such systems. Related to these patterns of emergence, are different types of sub-cognitive bias. For example, an assumption that collective (and co-operative) activity will emerge opportunistically. A consideration of individualistic bias suggests that adaptive emergence needs both feedback mechanisms from the micro (individual) level to the macro (societal) level and from the societal level to the individual level (a kind of focus of attention in a macro-agent). Irrespective of the tasks modeled and whether internal to a complex system or the human interaction with such systems, perturbant computational behaviour analogous to human emotive states can occur. This type of problematic emergent behaviour does not easily fit with the categories already introduced. The remainder of this paper considers how a developing computational theory of cognition can be used to monitor and manage such emergent states before the computational system becomes dysfunctional. The theory maps onto a distributed multi-layer computational model which makes use of tightly and loosely coupled agents working within a holistic system.