Abstract

Context: Attempts to conceive artificial agents are manifold. Often, however, basic aspects are overlooked, which in this case seem to be implemented already in current technologies. Problem: Agency is often conceived individualistically. Agents are seen as separate entities. This contribution sets out to conceive agents as emerging from the interaction of a multitude of agents. I claim that, somehow paradoxically, agents become agents only in the course of interaction. Method: A conception for conceiving constructivist agents illustrated with multi-agent simulations and artificial neural networks. I stress three aspects: double contingency, plasticity and granularity, which I consider relevant for conceiving artificial constructivist agents and which seem to find certain analogies in the methodology of artificial neural networks. Results: Constructivist agents should be conceived as emerging concurrently as a multitude of agents, providing one another with “irritations” to which agents need to adapt in order to evolve agency. In order to facilitate adaptation, they need what I call plasticity. In its turn, in order to guarantee sufficient adaptability, plasticity needs a certain degree of granularity. Implications: Researchers involved in the technical development of contemporary machine learning methods may not be aware that fundamentals of their methodology conform well to the epistemology of constructivism, just as constructivists may not be aware that their epistemology finds analogies in the technique of artificial neural networks. This contribution tries to raise these awarenesses. Constructivist content: Agents are conceived as developing agency on the basis of their on-board means. There is no need for assuming any “outside” but only “irritations” generating dissonances in the on-board means.