Xabier E. Barandiaran, Ezequiel Di Paolo and Marieke Rohde (2009), Adaptive Behavior 17(5):367–386, special issue on agency (eds. Rohde and Ikegami).
The brief’s citation (Adaptive Behavior 17(5), 2009) is correct.
Key takeaways
- Three conditions, not one. Individuality (the system distinguishes itself from its environment, without an observer doing it), interactional asymmetry (it is the active source of modulations of its coupling) and normativity (it regulates that coupling against norms it generates). Each is necessary; none, and no pair, is sufficient (§2.4, Table 1).
- Individuality is a precondition, not agency. A kitten warmed by its mother is an individual whose norm is met, but the mother drives the coupling, so it is not acting. A person with Parkinson’s tremor is an individual and the source of the movement, but the movement answers to no norm (Table 1).
- Asymmetry is modulation, not causation. Energetic and statistical (correlational) criteria each fail on clear cases: the gliding bird and the diver versus the faller. Asymmetry is defined instead as the system changing a subset p of the constraints Q on an otherwise symmetrical coupling, Δp = H_T(S) (Eqs. 1–3).
- The generative definition (§4). S is an agent for a coupling C with environment E iff S is an open autonomous system (a network in which every process depends on and enables another, and would run down in isolation) and S modulates C adaptively, that is, so as to maintain some of its own processes. Autonomy here is Varela’s, not personal autonomy.
- Consequences for models (§6). Sensorimotor coupling alone is too weak for agency, and “systems that only satisfy constraints or norms imposed from outside (e.g. optimization according to an externally fixed function) should not be treated as models of agency”.
Standing in the record
Filed on 2026-10-03 at the owner’s request, in the batch on agency, self and control, as the record’s organizational definition of what an agent is. No anthology topic holds it; it names robotics and AI as consumers of the definition but carries no instruction for machine-learning practice.
Agent versus individual. The record keeps agent, individual, self and
person apart (agency, individuation, self, personhood, ADR-024).
This paper argues for the first cut from inside one theory: individuality
is a condition on agency, and agency adds asymmetry and self-generated
norms. It is tagged agency first and individuation second for that
reason. It is not tagged self or self-governance. Its “self” is the
autos of an autonomous organization, and its “autonomy” is
organizational (Varela 1979), not the personal autonomy of Christman’s
survey LIT-297 or the self-endorsement of the self-determination theory
filings LIT-559 and LIT-560. The words coincide; the subjects do not.
Where it bears on what the record holds:
- Biological Autonomy (LIT-536). The acknowledgments thank Alvaro Moreno, and §§3–4 rest on the autonomy tradition that book later systematises (its chapter 4 is “Agency”). The paper’s autonomy is wider than the book’s metabolic one: “we do not restrict or reduce autonomy to the domain of metabolism”, and component processes may be neural, sensorimotor or social.
- Friston, “Life as we know it” (LIT-526). Two of the paper’s moves bear directly on the Markov-blanket account of the lifelike. It rejects statistical measures of influence as sufficient for asymmetry, and it requires that individuality be self-defined by an organization rather than drawn by an observer, which is the regress argument from Jonas in §2.1. LIT-526 individuates by a conditional-independence partition chosen by spectral clustering. Bruineberg et al. (LIT-603) borrow this paper’s term “interactional asymmetry” to describe the arrow structure a Friston blanket assumes. The paper predates LIT-526 and does not discuss it.
- Organizational closure. The definition’s statement 1.1 (each process depends on and enables another) is the operational closure that Montévil & Mossio (LIT-582) make formal as closure of constraints, and its normativity is the one Mossio, Saborido & Moreno (LIT-570) ground functions in; the paper cites “Mossio et al. 2009” for it.
- Participatory sense-making (LIT-571). Its “sense-making” is the same notion, from the same group: a deviation from a self-generated norm makes an event significant for the agent.