AI Systems, from Objects to Subjects: Comparing Frameworks from AI Literature and Social Sciences

Artificial Intelligence (AI) systems have predominantly been studied with regard to their impact on social, political and economical domains. The emphasis of their role within larger assemblages clarifies the depth of their interaction and mediation of social relationships, power structures, and cultural norms, calling for holistic approaches to research but also to a refinement of the conceptual understanding of these systems. Notwithstanding their contribution as tools for research, many perspectives on their agency fall short of a rigorous interpretation of their role as social subjects and its implications for empirical work. However, reconceptualizing AI systems considering recently acquired capabilities is an important step towards research on how AI systems represent and influence social phenomena. AI literature itself provides valuable insights into understanding AI systems as social subjects by highlighting their emergent properties and technical capabilities and by mobilizing frameworks from different disciplines. A more comprehensive perspective can be then reached when considering theoretical traditions from social sciences and philosophy, such as actor-network theory, posthumanism and assemblage theory.By acknowledging the complementarity of AI research and social science perspectives, unique contributions of these frameworks in comprehending AI's role as social subjects are underscored, while also shedding light on unknowns and topics of speculation. The decisive role of AI systems as subjects in contemporary society, and related limitations, both theoretical and practical in nature, in areas like interpretability, explainability and access, are recognized. It is also posited that social scientists are in an advantageous position to face said challenges, by virtue of sharing similar difficulties and uncertainties in their field and having developed a versatile toolbox for tackling them. Strong theoretical foundations are however needed to adequately make use of this expertise and identify opportunities and boundaries in research methods.