An agent is a System to which it is useful to ascribe Agency – a system whose persistence and behaviour are better explained by what it does to maintain itself than by its material composition. Where Agency names the capacity (boundary, self-model, self-preserving action), agent names the individual that has it: the thing, in the wide ontological sense, we treat as an owner of goals.
The coherent self: three interlocked elements
An agent is a coherent self, made of three interlocked elements:Again, borrowed from Levin.
- a Problem Space within which it operates – and which, for a higher-order agent, differs from the problem spaces of its parts;
- a Cognitive Light Cone that fixes the size of the goals it can pursue – how far in space and time its concern extends;
- a set of cognitive processes (competencies) that let it navigate the problem space with some degree of skill.
Intelligence here is James’s minimal, scale-free notion, so the triad scales continuously from molecular networks to organisms to Social Systems, each an agent with differently sized goals and competencies. Agency runs all the way down (organs, cells, molecular networks, to the principle of least action) and all the way up (organisations, classes, societies) – a disciplined restatement of animism that follows from treating Agency as a Scale-free Abstraction.The justification for scale-free agency talk is what Metzinger calls its heuristic fecundity: the vocabulary earns its place by what it lets us notice and do, not by a prior metaphysics. Cf. Prioritise abstraction over metaphor.
Agency ascribed makes an individual
An agent is the point at which Individuals are patterns in larger systems|a pattern in a larger system becomes an individual: the more a system’s stability looks like self-maintenance rather than accident, the more willing we are to call it an agent rather than a mere object. Agenthood is a matter of degree and of stance, not a metaphysical either/or.
This pragmatic, observer-relative construalThis is Michael Levin’s framing. is deliberately deflationary. To call something an agent is to adopt an interface: to hypothesise that the system occupies a certain problem space, holds a certain goal, and has certain competencies to reach it, and then to test whether that stance yields a richer relationship than manipulating the system’s parts. There need be no fact of the matter about the goal; the ascription earns its keep by working. This is the move behind Levin’s regenerative medicine: address the cell collective as an agent to be persuaded, not a mechanism to micromanage.
References
- Levin (2019), “The Computational Boundary of a ‘Self’: Developmental Bioelectricity Drives Multicellularity and Scale-Free Cognition”
- Levin & Dennett (2020), “Cognition all the way down”
- Levin (2022), “Technological Approach to Mind Everywhere (TAME)”
- Metzinger & Levin (2024), conversation, “From Self Models to Artificial Suffering”