Measuring AI Agent Autonomy: Towards a Scalable Approach With Code Inspection

Peter Cihon, Merlin Stein, Gagan Bansal, Sam Manning, Kevin Xu

Track: Sociotechnical

Keywords: AI agents, autonomy, code inspection, code review, AutoGen, assessment, evaluation, AI systems

TL;DR: Autonomy can be roughly assessed with code inspections, as shown empirically with AutoGen applications

Abstract: AI agents are AI systems that can achieve complex goals autonomously. Assessing the level of agent autonomy is crucial for understanding both their potential benefits and risks. Current assessments of autonomy often focus on specific risks and rely on run-time evaluations – observations of agent actions during operation. We introduce a code-based assessment of autonomy that eliminates the need to run an AI agent to perform specific tasks, thereby reducing the costs and risks associated with run-time evaluations. Using this code-based framework, the orchestration code used to run an AI agent can be scored according to a taxonomy that assesses attributes of autonomy: impact (actions, environment) and oversight (orchestration, human-in-the-loop, observability). We demonstrate this approach with the AutoGen framework and select applications.

Submission Number: 80