From ontology to ecology: AI as cognitive infrastructure and socio-technical mediation

Abstract

This article proposes an analytical shift from ontology to an ecology of socio-technical mediations, conceiving artificial intelligence (AI) not as an autonomous agent, but as a cognitive infrastructure that reconfigures the couplings between knowledge, decision-making, and social coordination. In contrast to debates focused on the nature of AI, this work asks what AI does in contexts of information asymmetry and model ambiguity, and how its technical architectures mediate relationships of power, visibility, and capacity for action. Drawing on contributions from distributed cognition, activity theory, socio-technical studies, critical data studies, and the literature on algorithmic justice, this paper develops a normative framework that identifies cognitive mediation, institutional design, and auditable governance as central pillars for the legitimacy of algorithmic systems. A minimal formalization of the socio-technical system is introduced, incorporating informational value, epistemic-procedural risks, organizational costs, and energy externalities. The article identifies key institutional operators—documentation, contestability, interoperability and portability, and energy budgets—and links them to auditable normative criteria based on capabilities, relational equality, non-domination, and public reason. The main contribution lies in offering an operationalizable theory that transcends technosolutionism and normative skepticism, arguing that the legitimacy of AI depends on institutional governance mechanisms that guide its design and use toward the expansion of human capabilities and the reduction of domination, making it accountable, enabling, and socially legitimate.