Anticipation in Autonomous Systems: Foundations for a Theory of Embodied Agents

This paper outlines a theory of anticipation in autonomous systems. Our account of autonomous systems is designed to model the basic organisational form of life. Anticipation is an integral feature of the autonomy account, and is an important foundational concept for an interactivist-constructivist (I-C) theory of embodied intelligent agents. We present the basic conceptual framervork of the I-C approach to intelligence, including an account of directed processes, normativity as process closure, and self-directedness as the basis of intelligence and learning. Intelligence is understood as emerging through increasing self-directedness. Self-directed systems anticipate and evaluate their interaction flow, directively modifying the interaction process so as to achieve goals that regenerate or improve the system's autonomous closure conditions. Learning arises out of the drive to improve anticipation, which starts by being contextual, vague, and implicit, and becomes increasingly articulated and explicit as the system constructs anticipative models and goals for managing and evaluating interaction. Cognitive development occurs through self-directed anticipative learning (SDAL), in which a pushme-pullyou effect is generated as increasingly rich anticipation increases the directedness of learning by improving error localisation, context recognition and the construction of improved anticipation. The paper concludes with an introduction to a general anticipatory conception of intentional agency, and a correlative critical appraisal of Rosen's pioneering analysis of anticipation.