Researchers Introduce Autonomous Agency Scale to Assess Self-Directed Behavior in AI Systems
A new preprint introduces the Autonomous Agency Scale (AAS), a behavioral framework designed to measure the degree of self-directed behavior in AI systems. The AAS scores systems across seven dimensions of agency, each evaluated in both active (user-initiated) and ambient (idle) temporal bands. When applied to six AI systems, the scale shows that task agents like Claude Code and Manus exhibit low ambient agency, while a persistent companion architecture uniquely demonstrates self-directed behavior during idle periods. The study also notes limitations such as single-rater assessment and potential evaluator bias.
Why it matters: The AAS provides a systematic method to distinguish between reactive and genuinely self-directed AI systems, addressing a gap in current AI evaluation frameworks.
Full story at: arXiv Computers and Society ↗