ResearchPod Summary
Agency is a central concept in biology, cognitive science, and artificial intelligence, yet it remains conceptually unstable. It is often used as a catch-all term for behaviors ranging from simple self-maintenance to complex, goal-directed anticipation. Current approaches generally fall into three categories: treating agency as an irreducible property of life, identifying it with robust self-stabilization, or universalizing it through formalisms like free-energy minimization. The authors argue that these approaches either risk vitalism, fail to distinguish agents from simple physical systems like thermostats, or suffer from explanatory overreach by applying the same formalisms to everything from pendulums to organisms.
To resolve these issues, the authors build on relational biology and the concept of semantic closure—a state where a system's components not only maintain the system but also interpret the constraints that define it. The core innovation of this paper is the introduction of time. By associating the constitutive processes of a semantically closed organization with distinct characteristic timescales, the system unfolds into an out-of-sync dependency structure. This structure can be formally modeled as a history-dependent, revisable Asynchronous Dynamic Bayesian Network. This temporal unfolding is not merely a detail; it is the mechanism that allows a system to move from simple persistence to genuine anticipation.
By analyzing these temporalized organizations, the authors derive a hierarchy that clarifies the thresholds of agency:
This framework allows for a more precise, biologically grounded definition of agency that avoids the pitfalls of both classical computationalism and purely descriptive enactivism. It provides a path toward understanding how simple chemical systems might transition into fully autonomous, agentive organisms.
AI-generated third-party summary by ResearchPod. Not official content or an endorsement by the paper authors or affiliated organizations.