NOTE-729: Bickhard — Autonomy, Function, and Representation

Contribution

A graded account of autonomy for open far-from-equilibrium systems, used to ground two things in the system’s current organisation rather than in its history or in an observer: normative function, at the level of self-maintenance, and representation, at the level of recursive self-maintenance. The representation account (“interactivism”) makes content an implicit presupposition of an indication that some further interaction will succeed, so error is defined, and detectable, from inside the system.

Key insight

What is at stake for a far-from-equilibrium system is its own existence: if it stops getting what it needs, it goes to equilibrium and ceases. That asymmetry is enough to make some of its processes count for or against it, with nobody watching. And a system that has to choose what to do next, depending on what kind of environment it is in, is in effect betting that the chosen interaction will work. The bet can fail, and the failure shows up inside the system. That is the seed of truth value.

Assumptions

• Naturalism. No ad hoc substances or levels “beyond which no further questions can be asked”.
• Emergence must be ontological and causally efficacious, not a matter of how an observer talks. A merely ascribed property leads to circularity or a regress of glossing agents.
• Process metaphysics. Kim’s exclusion argument is “valid” but “unsound”: it assumes particles, and quantum field theory gives only processes. A process exists only in some organisation, so organisation is a legitimate locus of causal power at every scale.
• Only current state is causally efficacious. History acts only through its effects on current state. This premise drives every objection to etiological accounts.
• Types. “Having a function” depends on what type a part and its system belong to. The paper says type assignment cannot rest on gloss, and defers the full account to Christensen & Bickhard (unread).

Key results

• Two forms of stability. Energy-well stability (an atom, a rock) needs no exchange with the environment. Far-from-equilibrium stability (a candle flame) requires the process not to reach equilibrium, so such processes “are necessarily open processes”.
• A ladder of autonomy. Autonomy is “actively contributing to their own persistence”, and graded.
  • Minimal: a chemical bath kept going by external pumps.
  • Self-maintenant: a candle flame keeps itself above combustion temperature, vaporises wax and drives convection that brings oxygen. It “cannot seek more fuel when the candle is almost gone”.
  • Energy-directing infrastructure with no alternatives: primitive cells with one metabolism. This is said to be the minimal autonomy of Christensen, Collier and Hooker.
  • Recursively self-maintenant: the system deploys “different processes depending on differences in the environmental circumstances”. The bacterium that swims up a sugar gradient and tumbles going down it. Stability holds “within certain ranges of changes in conditions”. This needs switching infrastructure, stable on the switching time scale, with energetic and informational aspects.
• Function from self-maintenance. Contributions to self-maintenance “are, in that sense, functional for that system”. Serving a function “is necessarily relative to the system whose maintenance is being contributed to”: “The gut of a parasite will serve functions for the parasite, but be dysfunctional for the host.” The normative asymmetry between function and dysfunction comes from the thermodynamic asymmetry between far-from-equilibrium and equilibrium processes.
• Having a function. A part has a function if it is of a type that tends to serve that function for the type of the whole, so a kidney that does not filter blood still has that function. Closeness to being able to serve it is graded (a damaged kidney versus scar tissue). “Dynamic presupposition” grounds the typing: a process presupposes other conditions to succeed, and the component typified as meeting the presupposition has the function, “whether or not it actually does so”.
• Against etiological function. Millikan’s swamp lion has no functions on her account, though it is molecule for molecule a zoo lion; so function “is not constituted by any property of current state”, and “the etiological approach gives us a model of function that is causally epiphenomenal”. History can explain why a kidney exists, not what having a function is.
• Representation from recursive self-maintenance. Switching needs differentiations: interactions whose internal outcome (A or B) depends on the environment. Differentiations “per se have no content”. Content belongs to the indication that a further interaction Q is appropriate after A. The indication is an implicit predication (“this environment … are Q-type environments”) that can be false, and “if false, a predication whose falsity may be detectable by the system itself”. Content is the implicitly defined set of conditions under which Q would succeed.
• Against information semantics, Fodor and Dretske.
  • Informational correspondence cannot be in error: if it holds the representation is correct, if not there is none.
  • Fodor’s asymmetric dependence is met by a receptor that crank docks in as dopamine does; it “captures functional error just as well as it captures representational error”. It also rests on counterfactuals not definable in current state.
  • Dretske’s recruitment of C as a cause of M “because of what it indicates about F” is explanatory, not causal, so it appeals to an explaining observer; and it rests on learning history.
  • None of these allows system-detectable error, which organisms show.
• More complex representation. Selection among several indicated interactions needs goals; goal conditions can be differentiated without being represented (a TOTE organisation). Indicated internal outcomes check the indication’s truth from inside and so support error-guided behaviour and learning. Branching and iterated indications form webs said to ground objects and abstractions.
• Conclusion. Cognition is grounded in “recursively self-maintenant autonomy”. Endnote 1: autopoiesis “emphasizes independence from the environment, while autonomy emphasizes the ability to make use of the environment”.

Claims

id	claim	strength	support
C1	Kim’s exclusion argument fails because it presupposes a particle metaphysics	weak	an appeal to quantum field theory and the claim that processes are always organised; not developed
C2	A far-from-equilibrium system’s contributions to its own maintenance are functions for it, with function and dysfunction fixed by the thermodynamic asymmetry	moderate as a proposal	conceptual; the candle flame and the parasite as illustrations
C3	Etiological accounts make function causally epiphenomenal	moderate	the swamp-lion argument plus the premise that only current state is efficacious
C4	Representation emerges in recursively self-maintenant systems as the implicit presupposition of indications of interaction success	moderate as a proposal	conceptual; the bacterium
C5	Information semantics, Fodor’s asymmetric dependence and Dretske’s account cannot model error or system-detectable error	moderate	argument, with the dopamine/crank counterexample
C6	Complex representation (objects, abstractions) is built from webs of conditional indications	weak	asserted here, with the work deferred to earlier books

Concepts

• Far-from-equilibrium stability. Persistence of a process that would cease at equilibrium, and so needs exchange with its environment.
• Autonomy. Active contribution to one’s own persistence; graded.
• Self-maintenance. Contributing to the conditions of one’s own far-from-equilibrium existence (a flame).
• Recursive self-maintenance. Maintaining the property of being self-maintenant across changes in the environment, by switching among processes.
• Infrastructure. Structure stable on the time scale of switching, which carries out the switching and directs energy.
• Serving a function. Contributing to the self-maintenance of a specified system.
• Having a function. Being of a type that is presupposed to serve a function for the type of the whole, whether or not it does.
• Dynamic presupposition. What a process needs to hold in order to succeed.
• Differentiation. An internal outcome that depends on the environment; it has no content of its own.
• Indication. A functional link from a differentiation to a further interaction as possible or appropriate; it implicitly predicates conditions of success, and that is its content.

Connections

It builds on Christensen, Collier and Hooker’s autonomy theory, Campbell’s evolutionary epistemology (the bacterium) and Moreno & Ruiz-Mirazo on metabolism, and argues against Millikan, Godfrey-Smith, Fodor, Dretske, Cummins, the physical symbol system hypothesis and connectionism. Mossio, Saborido & Moreno (LIT-570, NOTE-444) name Christensen & Bickhard as allies they improve on with explicit conditions, and Barandiaran’s paper on malfunction (NOTE-730) builds on the same organizational line.

Bearing on the record

On CLAIM-288: support for its premise, by a different route from the one it cites. The objection holds that norms an organization appears to maintain are the modeller’s when its operators maximise objectives the modeller fixed. Bickhard’s account says the same about function. A function is relative to the system whose far-from-equilibrium maintenance it serves. A function that exists only in an observer’s ascription or explanation is not natural, by the regress argument, and that is his objection to Dretske. A reward the modeller writes is not the operator’s or the coalition’s persistence condition, so on this account it confers no function on anything in the simulation. The paper does not discuss designed systems, and it ends saying the design of artificial cognitive systems “will have to accommodate these conditions”. The application to reward-maximising operators is the record’s.

On CLAIM-284: vocabulary, not evidence. Recursive self-maintenance is “maintaining the property of being self-maintenant” by switching among processes. That is governing among regimes, the second of CLAIM-025’s three layers. The third, transforming the organization that does the switching, does not appear: the bacterium’s switching infrastructure is fixed. The paper says nothing about selves or about authority among processes. Like Di Paolo (LIT-1017) it supplies a ladder on which the thesis could be stated, without showing the layers come apart.

On QUESTION-087: a standard internal to the larger organization, with a limit.

• The parasite sentence gives the shape of an answer. A sub-system with its own self-maintenance has functions for itself that can be dysfunctions for the whole that contains it. An authority structure that reproduces itself is functional for itself. Whether it is dysfunctional for the organization is fixed by the organization’s own persistence conditions, not by goals assigned from outside.
• Having a function adds malfunction proper: a part typified, by what the whole’s processes presuppose, as meeting a condition it does not meet. An authority structure presupposed to route decisions well and not doing so would malfunction in this sense.
• The limit. The only standard is contribution to persistence. Captured authority that keeps the organization alive, or makes it more robust, is not dysfunctional on this account. That is CLAIM-075’s point, and Bickhard does not dispute it: nothing in the paper says what ought to persist.
• The account of types, which carries malfunction, is deferred to Christensen & Bickhard and is the paper’s weakest step.

On CLAIM-075: no conflict. The norms are function and dysfunction relative to a system’s persistence. The paper calls this “normative” and never treats it as value.

On THEORY-072: a disagreement about functions, not about the demarcation. Mossio, Saborido & Moreno, a source of THEORY-072, hold that flames “do instantiate closure, but their components do not have functions”. Bickhard gives the flame self-maintenance and says its contributions are “functional for that system”. Both can be read as true of different notions: Bickhard’s flame has serving-a-function but no infrastructure, so nothing in it has a function as a typed part. His infrastructure tiers sit where THEORY-072’s differentiation does. But the two do disagree on whether self-maintenance alone yields norms, and THEORY-072’s title (that far-from-equilibrium self-maintenance is not the line between organisms and flames) is not contradicted, since Bickhard draws no organism line.

On THEORY-071 and LIT-1017. Bickhard’s bacterium is the chemotaxis case of LIT-566 and the case Di Paolo uses. His split between self-maintenance and recursive self-maintenance is close to Di Paolo’s between robustness and adaptivity, five years earlier. They differ on where norms start. Di Paolo says conservation gives only an observer’s norm; Bickhard gives the flame function. The difference is smaller than it looks: on Bickhard’s account error the system can detect arrives only with recursive self-maintenance, which is where Di Paolo puts norms the system can use. Bickhard has no individuality condition and no interactional-asymmetry criterion; the system’s boundary is taken as given.

Limitations

• Conceptual throughout. No model, no formal definition of far-from-equilibrium, contribution or switching.
• The anti-Kim argument is a page of assertion about quantum field theory.
• Contribution to self-maintenance is not given a measure, and the flame’s “several active contributions” are listed, not derived.
• Having a function needs types, and the account of typing by dynamic presupposition is sketched and deferred.
• The premise that only current state is efficacious is assumed, and etiological theorists reject it as a constraint on what function is.
• Representation beyond the primitive level (objects, numbers, language) is asserted with references to earlier books.

Open questions

• Can contribution to self-maintenance be stated so that it is measurable in a model, for instance as the effect on a viability kernel? That would make the parasite criterion applicable to the record’s authority-capture cases.
• Does a sub-system that switches among self-maintaining processes inside an organization (a faction, an entrenched office) have representation in Bickhard’s sense, and functions of its own that conflict with the whole’s?
• Is the flame’s serving-a-function the same thing Mossio et al. deny it, or a weaker notion? A side-by-side reading of Christensen & Bickhard (2002) would settle it.