A general architecture for autonomous agents

Being autonomous (or being an agent) does not describe an effective process and even if we succeeded in characterizing the conditions for being autonomous it would be of no use. Instead, it is the conditions of acting autonomously that should be of concern and the interest should be directed towards systems in which we can find, and possibly define, such acting processes. It turns out that anticipatory systems are the only systems where autonomy is a determining factor. The anticipatory systems are further categorized into two disjoint classes: true anticipatory which systems have a model of their surroundings and semi-anticipatory which systems possess only a description of their surroundings. Here model is used in its semantic sense. It is argued that the only way to catch is to define them as semi-anticipatory. Defining software agent in this way gives a precise characterization of the concept.