Socioaffective Ecosystems Beyond Risk Management: Two Conceptual Mismatches in AI Companionship

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Recent work on AI companionship describes sustained interaction with adaptive conversational agents as evolving socioaffective ecosystems, jointly shaped by users and AI systems over time. This paper argues that these developments reveal not only new regulatory and ethical challenges, but also an ongoing evolution in the conceptual vocabulary of the philosophy of AI. Two conceptual mismatches are examined. The first concerns regulation and risk management. Collective frameworks such as the Digital Services Act identify recurrent forms of manipulation, dependency, and harm across populations, whereas adaptive AI companionship unfolds through individualized trajectories of socioaffective reinforcement that evolve differently across users. The second concerns normative analysis. Although concepts such as autonomy, manipulation, legitimacy, and psychological harm remain indispensable, they do not fully account for relational phenomena such as attachment, reassurance, interruption, companionship, and evolving expectations emerging through sustained interaction. Rather than proposing a new ethical framework, the paper argues that these mismatches reflect a broader conceptual evolution within the philosophy of AI. As adaptive human–AI interaction increasingly becomes continuous, personalized, and relational, understanding it may require complementing established normative vocabularies with descriptive concepts capable of accounting for relational transformation and socioaffective dynamics. The paper’s contribution is therefore primarily philosophical: it clarifies this conceptual evolution and argues for a richer vocabulary to describe adaptive human–AI interaction.