TERM-063: agent, as an individual that is the active source of modulations of its coupling with its environment and regulates them against norms its own organization sets (Barandiaran, Di Paolo and Rohde)

Definition

THEORY-071, from Barandiaran, Di Paolo and Rohde (LIT-566). A system is an agent on this definition when:

1. Individuality. It distinguishes itself from its environment, with no observer doing it. Here that means organizational closure (TERM-070).
2. Interactional asymmetry. It is the active source of modulations of its coupling with the environment.
3. Normativity. It regulates that coupling against norms its own organization sets.

A11’s form: “A system qualifies as an agent under this criterion when its own maintained organization establishes conditions of continued functioning and actively modulates its coupling with the environment in accordance with those conditions.”

Why it rivals TERM-044, and does not complement it. The two give different verdicts on cases the record already holds:

• The thermostat with its designers passes TERM-044 (CASE-044: it would be redesigned, so its policy would adapt). Here it fails: the norm is the designers’, not set by its own organization, which is CLAIM-288’s point about norms a modeller fixes.
• A trained policy’s training process is an agent on Kenton et al.’s account, which is TERM-044’s third test (THEORY-527). Here it is doubtful, with no organization of its own that sets norms (CLAIM-367, CASE-113).
• A cell in passive osmosis is an individual and not an agent here, because it is not the source of the modulation (THEORY-071’s own case). TERM-044 likely agrees for another reason, its description and adaptation tests, so the two part on who is an agent, not only on why.

So the two cannot both be the unqualified “agent” of the distributed-agency line. ADR-046’s lifecycle fits: a Proposed candidate rivals the Active term it would replace for the same use. A12 put the choice: “If proposed as replacements for TERM-044, then rivals: [TERM-044] is appropriate. If instead they are separate dimensions that can coexist, complements is the eventual relationship.” If the owner wants both in force as dimensions, the relation becomes complements, and this title must stop saying “agent” unqualified.

What would settle it. THEORY-071’s own promotion test: measures of closure and of normativity, and an asymmetry criterion that does not depend on how the modeller splits parameters from variables, applied “almost mechanically” to a protocell model, an evolved robot controller and a homeostat, with verdicts that match LIT-566’s cases without being tuned to them. And QUESTION-032 and QUESTION-093, which ask what makes a factor an agent and which autonomy it needs.

What it is not

• Not a homeostat. THEORY-583’s cybernetic Self counts any homeostat. This definition does not.
• Not TERM-044’s third test, policy adaptation (THEORY-527, THEORY-801). “Agent, as an adaptive policy mechanism”, which A11 and A12 also proposed as a rival, is that test taken alone: TERM-044 narrows it rather than rivalling it, and it is not filed.
• Not a subject. THEORY-071’s “autonomy” and “self” mean organizational closure, not a subject or self-governance.
• Not a change to the claims stated in TERM-044 (CLAIM-155, CLAIM-157, CLAIM-167, CLAIM-168, CLAIM-170 and CLAIM-174). They are unchanged.

Note of 2026-10-29: what holds it back

It rivals TERM-044, which is Active, and the test that would choose between them has not been run. No measure of closure, of normativity, or of interactional asymmetry independent of how the modeller splits parameters from variables has been applied to a protocell model, an evolved robot controller and a homeostat (THEORY-071’s promote_when). The other route is the owner’s ruling that the two are complementary dimensions, which would also retitle this term away from the unqualified “agent”.