Anicker, F., Flaßhoff, G., & Marcinkowski, F. (2024). The matrix of AI agency: On the demarcation problem in social theory. Sociological Theory, 42, 307–328. https://doi.org/10.1177/07352751241289925
Summary
This paper tackles what the authors call the “demarcation problem” for AI agency: how, and whether, artificial intelligence systems should count as agents in social theory. Their central move is to reject any account that treats agency as an intrinsic property of a system. Instead, they argue that agency is a social status—a socially granted and continually monitored license to issue actions. On this view, whether a machine is an agent depends not on its technical capabilities but on the abilities that social practices attribute to it. The authors develop theoretical criteria for distinguishing agents from nonagents and assemble these into a “matrix of AI agency” for classifying entities according to their socially attributed abilities.
Key Contributions
- Reframes AI agency as an attributed social status rather than an inherent system property.
- Proposes theoretical criteria for demarcating agents from nonagents grounded in socially attributed abilities.
- Introduces a conceptual “matrix of AI agency” for situating AI systems within questions of agency.
- Establishes the machine-agency question as an urgent concern for contemporary social theory.
Methods
The work is conceptual and theoretical, drawing on social theory and the sociology of agency. The authors build demarcation criteria by treating agency as a license acquired and monitored through social practices, then organize these criteria into a classificatory matrix. There is no empirical study; the contribution is a reframing of the analytic categories through which machine agency is debated.
Findings
- A matrix of socially attributed abilities can be used to classify entities as agents or nonagents.
- Machine agency is contingent on social recognition and ongoing monitoring, not on technical capability alone.
- The social-status conception stands in contrast to views that locate agency in the internal properties of a system.
Connections
This paper’s social-status account of AI agency connects to broader work on how societies and institutions constitute and imagine AI systems, such as the deep-mediatization and institutional perspective in Hepp2026-oi and the study of AI’s shifting cultural and organizational roles in Stanusch2026-ec. Its emphasis on attribution over intrinsic capability offers a theoretical counterpoint to more artifact- and platform-centered analyses of generative AI in this topic cluster.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify