Agency, in its minimal sense, is a System’s capacity to actively maintain its own existence: to uphold a System Boundary, to keep a model of its Environment, and to act on that environment so as to preserve its own coherence against decay.
A system with this capacity is an Agent.
Minimal agency has three ingredients:
- a boundary – its Markov Blanket – that separates the system from its environment while allowing exchange across it;
- a self-model – the system embodies expectations about itself and its relevant environment, maintaining Information Closure;
- action – the system does not merely undergo its dynamics but counteracts perturbations that threaten its persistence. In agential systems, the trajectory through the State Space is an outcome of goal-directed behaviour – the agent actively seeking its attractor states – rather than mere gradient-following.
This is deliberately weaker than the folk notion of agency: no consciousness, intention or deliberation is required. A bacterium swimming up a nutrient gradient is an agent in this sense; so is a cell maintaining its membrane potential. Minimal agency marks the threshold at which it becomes more informative to describe a system by what it does to persist than by what it is made of – cf. Individuals are patterns in larger systems.
Agency is a Scale-free Abstraction: the same pattern – boundary, self-model, self-preserving action – recurs in cells, organisms, ecosystems and Social Systems. A movement or organisation that maintains a coherent self-model, a problem space it navigates and boundaries it holds is a collective agent in a precise, non-metaphorical sense.
This is a claim about agency, not experience: agency at every scale is not panpsychism.
References
- Barandiaran, Di Paolo & Rohde (2009), “Defining agency: individuality, normativity, asymmetry, and spatio-temporality in action”
- Friston (2013), “Life as we know it”
- Krakauer, Bertschinger, Olbrich, Flack & Ay (2020), “The information theory of individuality”