A three-dimensional typology of agency for advanced AI systems

Recent research from arXiv introduces a three-dimensional typology to categorize the agency of advanced AI systems. This framework moves beyond existing approaches that focus primarily on normative or highly agentic systems, offering a more nuanced understanding by considering moral/legal, individual/collective, and human/non-human aspects of AI agency.

Why it matters

Understanding the different types of agency attributed to or exhibited by advanced AI systems is critical for developing robust governance frameworks and managing emerging risks. This typology provides a foundational conceptual tool for organizations to analyze and anticipate the ethical, legal, and operational implications of increasingly autonomous AI.

Key insights

• Existing research on AI agency often focuses on normative concepts or particularly agentic systems.
• No prior comprehensive framework addresses the diverse types of agency instantiated by advanced AI systems, especially non-moral forms.
• A new typology is proposed, based on established theories from philosophy, ethics, legal theory, and sociology.
• The typology defines agency across three dimensions: its nature (moral or legal), its mode (individual or collective), and its locus (human or non-human).
• Combining these dimensions yields eight distinct profiles of AI agency.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.20041