Autopoiesis of the artificial: from systems to cognition

Abstract

In the seminal work on autopoiesis by Varela, Maturana, and Uribe, they start by addressing the confusion between processes that are history dependent and processes that are history independent in the biological world. The former is particularly linked to evolution and ontogenesis, while the latter pertains to the organizational features of biological individuals. Varela, Maturana, and Uribe reject this framework and propose their original theory of autopoietic organization, which emphasizes the strong complementarity of temporal and non-temporal phenomena. They argue that the dichotomy between structure and organization lies at the core of the unity of living systems. By opposing history-dependent and history-independent processes, methodological challenges arise in explaining phenomena related to living systems and cognition. Consequently, Maturana and Varela reject this approach in defining autopoietic organization. I argue, however, that this relationship presents an issue that can be found in recent developments of the science of artificial intelligence (AI) in different ways, giving rise to related concerns. While highly capable AI systems exist that can perform cognitive tasks, their internal workings and the specific contributions of their components to the overall system behavior, understood as a unified whole, remain largely uninterpretable. This article explores the connection between biological systems, cognition, and recent developments in AI systems that could potentially be linked to autopoiesis and related concepts such as autonomy and organization. The aim is to assess the advantages and disadvantages of employing autopoiesis in the synthetic (artificial) explanation for biological cognitive systems and to determine if and how the notion of autopoiesis can still be fruitful in this perspective.