Survival of the mutable: architecture of adaptive reactive agents

Abstract

An agent is defined as any device that perceives a certain environment through stimuli and acts upon it as to achieve a certain goal. There is a plethora of theories, architectures and languages in the literature aiming at how much an agent may be improved at performing a task. However, the majority of them focuses on the internal agent function itself instead of adopting a macroscopic, broader view of what the term “intelligent” means in the long run. In this paper we take a bio-inspired route and describe how the simplest reactive agent can be boosted towards improvements at performing complex tasks by making it mutable. We provide a mathematical framework to support such features. Conceptually, the addition of a mutability layer does not break the existing paradigms and allows hybrid approaches as a means to achieve better results.