Abstract Definitions of artificial intelligence (AI) do not merely describe a technological field; they shape ethical analysis, governance frameworks, and the allocation of responsibility. Existing debates tend to define AI in two dominant ways: first, in technical terms, by reference to computational methods, system capabilities, or task performance; and second, in governance-oriented terms, by reference to legal scope, risk classification, and regulatory control. Both approaches are indispensable. Yet, taken on their own, they remain normatively incomplete because they abstract AI systems from the social and institutional settings in which their effects materialize. This paper argues that the contribution is not to rediscover the socio-technical character of technology, long established in science and technology studies and the philosophy of technology, but to translate that insight into a more adequate definition of AI for ethics and governance. To do so, the paper first differentiates major families of technical and governance-oriented definitions and shows how they underdescribe emergent ethical problems such as distributed responsibility, structural discrimination, epistemic dependence, and transformations of institutional practice. It then draws on scholarship in science and technology studies, philosophy of technology, and AI ethics to conceptualize AI as a socio-technical actor: not a moral agent in a strong anthropomorphic sense, but a system that mediates action, structures choices, and reshapes social and institutional relations through sustained interaction with human actors. On this basis, the paper proposes a socio-technical definition of AI and develops its implications for governance. A socio-technical concept of AI, it argues, provides a stronger foundation for context-sensitive accountability, institutional analysis, and ethical evaluation than definitions focused solely on functionality or risk.
Patrick Hedfeld (2026) studied this question.