A Taxonomy of Agents: Intro & Request for feedback
A new project aims to unify the fragmented definitions of 'agency' across AI, economics, and biology.
Researcher Jonas Hallgren, writing on the Equilibria Substack, has initiated a public research project to create a comprehensive taxonomy of 'agents.' The core premise is that the term 'agent' carries vastly different meanings across scientific fields like control theory, economics, biology, cognitive science, and AI research. The project adopts philosopher Daniel Dennett's 'intentional stance,' treating agency not as a fixed property but as a useful compression strategy that observers use to predict system behavior. The goal is to systematically map why each field compresses the concept differently based on its unique predictive challenges.
The project's methodology involves publishing a series of short posts, each dedicated to a specific domain such as control theory, economics, evolutionary systems, developmental biology, and machine learning. For each field, the team will distill its canonical conception of an agent, identify essential features, and explain why that model works for its purposes. Crucially, these initial characterizations will be verified and corrected through recorded conversations with domain experts. The final artifacts could include a detailed comparison table mapping fields against agentic features (like goal-directedness or memory) or a synthesized academic paper, with the specific output shaped by community feedback.
- Adopts Dennett's 'intentional stance,' framing agency as a predictive compression strategy used by observers.
- Will publish verified analyses across 6+ domains including economics, biology, and AI via expert interviews.
- Seeks public feedback on outputs, which may include a comparative feature matrix or a formal paper.
Why It Matters
Clarifying foundational concepts is crucial for effective collaboration between AI researchers, economists, and biologists building intelligent systems.