An open learning system

Abstract

In this paper we discuss an approach for achieving high adaptivity in complex dynamic environments through learning. Unlike in traditional approaches, where adaptive systems need to be strongly equipped with knowledge about their environment and possible adaptation strategies, we propose a direction where the level of original preconceptions is kept low. In our model, both the knowledge and the strategies emerge dynamically and flexibly, during the system´s normal operation. The proposed model also includes a mechanism that systematically replaces old preconceptions with more accurate ones (e.g. to observe new features). The proposed principles were evaluated in a theoretical world via simulation; where the adaptive system was challenged to win an amoeba game against opponents with different strategies, and changing game rules (3-7 long series required to win).