Semiology of Autopoiesis (II)

Selections of the Sign: Unity

Maturana and Varela’s Autopoiesis: the Organization of the Living begins with a sentence indebted to George Spencer-Brown and his form of distinction where ‘distinction is perfect continence’. In their version, this becomes ‘A universe comes into being when a space is severed into two. A unity is defined.’ (Autopoiesis and Cognition: the Realization of the Living. [A&C] Dordrecht: D. Reidel, 1980, p 73)

For Spencer-Brown distinction – or severing a space into contained and uncontained – is the condition for indication, or the determination of a unity: <p>. If unity and indication are assumed to depend on perfect containment, then perfection of the cut or separation is necessary, even if subsequent crossings are to be allowed under strict unity-preserving conditions.

Spencer-Brown’s logic in Laws of Form leads to a reconstruction of mathematics and to claims about the necessary form for all things. This is a metaphysics – a speculative account of the nature, order and value of everything – and I have contrasted it with Deleuze’s metaphysics of multiplicity.

The difference between the two can be understood as the difference between metaphysics starting with a demand for perfect identification, achieved through distinction, and metaphysics starting with problematic multiplicity, emerging as continuous variations. Does existence start with the hard boundaries of rocks, or with the soft evanescence of waves? Are waves short-lived rocks, or are rocks long-living waves?

Autopoiesis and Cognition is metaphysical in its commitment to universal claims, such as all and common to all, as the basis for science and for experience. Unities ‘are at the base of all scientific enquiry’ and in common experience ‘we encounter living systems as unities.’ (73) The point of a semiology of autopoiesis isn’t to criticise this metaphysics. As a first step, it is to analyse the selection of unity as one of the main elements of a sign. ‘Autopoiesis’ names a set including the concept of unity: {unity, …}.

Unity means to be connected together as a whole. It describes unions, where parts are joint or in agreement; like a political union (a united republic), organs working together (a body, a cell), a union of mechanical and chemical processes (an engine), an evolving yet united structure (the unity of a language, of a transport network) and, in William Heath Robinson’s inventions, the unity of engineers and machines as they fit square pegs into round holes:

It is counter-intuitive for autopoiesis to be a process with perfect unity. Useful processes depend on external inputs and outputs, like a factory (raw materials to product) or an orchard (sun, soil and rain to apricots).

The engineers, pulleys, wheels and pegs of Heath Robinson’s machine form a unity. We could think of it as autopoietic – as self-producing and self-maintaining – since the engineers construct, improve and run the machine. However, counter to the notion of hermetically sealed process, the inputs and outputs do not belong to the machine. The necessary supply of pots and pegs comes from outside.

To be consistent with their definition of perfect containment of autopoietic processes, Maturana and Varela exclude external aims, energy, supplies and environmental variables (such as temperature, pressure, humidity) from autopoiesis as unity.

To form a perfect unity, autopoiesis must not depend on external teleology – the ascription of ends or purposes – as a necessary part of the internal process. Unity is crucial for this argument about necessity. It explains how a process might depend on external factors but not when it is taken as autopoietic.

Unity is not the process as a whole but rather its essential core. This distinguishes the unity of necessary components (A is unified according to this organisation of self-production) from identity understood as everything that belongs to something (A = A).

Energy is necessary for survival and functioning, but it is not necessary for any particular unity of components to function exactly as it does. Fuel is necessary for rotary and piston engines to work, but not to explain differences in their unity, or why their components work together in different ways.

In the following diagrams, a difference between engines is shown in the difference between components and how they function together, not by their use of the same fuel. Input and outputs do not account for differences in types of unity, components or cycles.

Unlike pistons and rotaries, running only when attached to other machines (ignition and timing, water cooling, fuel injection, exhaust, lubrication), an autopoietic form is autonomous and self-producing. It rules over its organisation, functioning and development.

This does not mean that an autopoietic process has created itself from nothing and depends on nothing else. It means that as a functioning unity it controls, transforms and sustains itself. Viewed externally, it does this over a given period of time, but this is not a necessary aspect of its autopoiesis. Taken internally, autopoiesis has a particular space, its topological domain, defined by all the transformations it goes through while maintaining its unity.

Something physical associated with an autopoietic process – a thing corresponding to a unified organisation – can be made and given birth to. It can be stopped and started, killed and impaired, decelerated and accelerated. Yet, the autopoietic unity itself remains autonomous. How its components are produced, destroyed and work together is entirely within its control and within the boundaries of its topological domain:

A living system is a living system because it is an autopoietic system in the physical space, and it is a unity in the physical space because it is defined as a unity in that space by and through its autopoiesis. Accordingly, any structural transformation that a living system may undergo maintaining its identity must take place in a manner determined by and subordinated to its defining autopoiesis; hence, in a living system loss of autopoiesis is disintegration as a unity and loss of identity, that is, death.

Autopoiesis: the Organization of the Living, 112

The ‘how’ of autopoiesis is logical and structural. It concerns the necessary form of unity. It is also dynamic, because this form evolves autonomously through variations in components and structure. However, this dynamic evolution is itself logical and structural. Autopoiesis depends on levels of interrelated and communicating structures of organisation.

Autopoiesis is not about variables of operation such as speed or heat. These interact with the form, since they maintain and threaten it; for instance, as inputs of energy, or as unviable states such as too much pressure. In those cases, to survive, the form must include control over the threatening variable. This explains why homeostasis – autonomous monitoring and command of variables such as temperature during a transformation – is an important factor for autopoiesis.

Homeostasis keeps the form of unity within limits of viability. If a process requires external intervention to keep it within operating limitations, it is not autopoietic. A reactor requiring an external control room is not autonomous. A reactor with its control room might be. A reactor and its control room requiring external permission to shut down would not be autonomous or autopoietic, but a fully independent one might be autonomous.

The case of the reactor and control room is instructive because the idea of autopoiesis can be taken to apply to human institutions (in sociology, psychology and law, for instance) and to complex engineering (in artificial intelligence and cybernetics). If adopting the idea from Maturana and Varela, there is a risk of misapplication, given the difficulty in finding cases of fully autonomous institutions and engineering solutions, let alone ones producing all their components as they transform.

This rarity is partly a consequence of regulation. Institutions and systems have internal regulations, but in politically connected societies they are also subject to external laws, such as corporation law, environmental standards, safety rules, and ongoing training and education. An organ concerned with compliance is a barrier against autopoiesis, because it introduces external control and subjection, turning unity into relative dependencies.

There is suspicion of compliance or internal investigation departments because they are hybrids, partly inside and partly outside the firm, bridging between internal and external regulations and practices. As such they cannot be part of autopoietic systems. Maturana and Varela’s insistence on self-production of components is also in this vein, since an externally produced component (a new employee, a piece of software, a graft) carries the threat of external control.

Except for idealisations and in science fiction, it is hard to come up with working systems that aren’t overseen by external commands within internal protocols. The ideal state of full autonomy, its dangers and legal or technical counteractions to them have been a feature of robot fiction and theory from early on.

If a military robot is autopoietic, then there is a risk that in creating and sustaining itself autonomously it might arrive at a state where killing the ‘wrong’ humans becomes a possibility. To avoid such rogue states and ban killer robots, Isaac Asimov devised laws of robotics constraining robots from harming humans: ‘”Two,” continued Powell, “a robot must obey the orders given it by human beings except where such orders would conflict with the First Law.”‘ (Runaround, Asimov 1942)

As a square-peg-for-round-hole maker or spaghetti stretcher, Heath Robinson’s contraption is not autopoietic, since its goals are external. As a self-contained process it might be, so long as it satisfies the conditions for unity and hence autonomy.

The apparent contradiction that something can be both autopoietic and not autopoietic is resolved in terms of observation and explanation. An observer’s, or manipulator’s, or user’s understanding of purpose cannot justify a form as autopoietic. They explain it as allopoietic: functioning for another. The form is autopoietic only when its organisation, control and development can be explained strictly internally in terms of its unity:

An autopoietic machine is a machine organized (defined as a unity) as a network of processes of production (transformation and destruction) of components that produces the components which: (i) through their interactions and transformations continuously regenerate and realize the network of processes (relations) that produced them; and (ii) constitute it (the machine) as a concrete unity in the space in which they (the components) exist by specifying the topological domain of realization as such a network.

Autopoiesis: the Organization of the Living, 78-9

I have identified unity as the first element of the sign for autopoiesis because of its importance for this definition. There are two levels of unity here: unity of organisation of a network and concrete unity. They are dependent on one another. The unity of the plan – the structure – specifies how concrete unity is realized through processes of monitored production and destruction.

Observed as a machine for translating, Searle’s Chinese room is allopoietic. It’s purpose is to translate languages by following rules despite no understanding of Chinese. Observed through its own rules for creative development, with no reference at all to its role as a translator, the machine might be autopoietic.

We can say ‘it does this in order to do that’, but we can also say ‘it produces itself like this’. If we don’t need to refer to external purpose to explain internal organisation, then we might be describing autopoiesis, depending on other tests for autonomy and self-production.

Following the distinction between the Greek allo (other) and auto (self), Maturana and Varela’s autopoiesis cuts out the subjective other of external self-interest. It thereby avoids impositions of purpose. Don’t tell me how it does something for you. Tell me what it does to maintain its unity and organisation while creating itself. Observation is limited to the identification of unity, not to aims, roles and values.

Sidestepping teleology is a way of avoiding historical positions in biology such as finalism and design. Life is autopoiesis, rather than something decided by external purpose, or as the result of external creation. Against vitalism – focused on life forces – the autopoietic definition of life is a combination of mechanics and production.

It is mechanical in its understanding of organisation, but adds the condition that autopoiesis produces its own components. A car engine is neither autopoietic nor alive, but a factory producing its own machines and renewing them autonomously could be considered to be alive, even if not exhibiting signs of ‘life forces’, purposes or design similar to other living things.

The logical and practical advantages of the distinction between teleology and autopoiesis can be understood through the different perspectives of engineer and user. A user identifies a machine with purpose and registers its failure through lack of outputs. When tracing faults, an engineer identifies a machine with function and seeks failure in functional breakdowns.

The fact that a particular output has ceased tells us very little about why. The sealed unity of autopoiesis concentrates organisation and development on how something produces itself, independently and on its own terms, rather than on how it fulfils outside expectations. However, this also means that autopoietic approaches might reinforce the potential weakness of some engineering methods when they fail to take account of use and purpose in design.

There are ethical and political advantages to thinking of life and machines as autopoietic. Once we stop thinking of animals and other humans in terms of external purpose, we can avoid their reduction to imposed roles such as service, sustenance, work and slavery. An autopoietic lifeform is for itself, not for another.

Maturana and Varela’s identification of life with autopoiesis is revolutionary. It breaks with the anthropological focus for ideas of life, simultaneously making it harder to count as living – because of the criteria for autopoiesis – and easier to reach the ranks of the living – because mystical preconceptions about life are abandoned.

This revolution in ideas of life also has connections to science fiction. Whether life can be extended to machines is a sci-fi staple, existing long before Phillip K. Dick’s classic ‘Do Androids Dream of Electric Sheep?’ and the two Blade Runner films drawn from it. By Blade Runner 2049, androids begin to approach autopoiesis in giving birth, possibly autonomously. It’s only an approximation, though, since the films remain vitalist and sentimental in their ideas of life when compared to Maturana and Varela’s autopoiesis.

In a reversal of Asimov’s laws, Dick’s short story leads to reflection on whether humans have the right to kill androids if they satisfy conditions for life (changing an ‘it’ of circuits into a living being):

He shot Roy Baty; the big man’s corpse lashed about, toppled like an overstacked collection of separate, brittle entities; it smashed into the kitchen table and carried dishes and flatware down with it. Reflex circuits in the corpse made it twitch and flutter, but it had died; Rick ignored it, not seeing it and not seeing that of Irmgard Baty by the front door.ópia.pdf

Maturana and Varela are much more radical than theoreticians seeking to include more types of being in the category of living. Due to the insistence on unity for autopoiesis and the hard but clear criteria of self-production of components and autonomy, their definition of life is an extreme individualism, not in the sense of a liberal politics of freedom and desire, but in detaching the condition for life from groups (life as an X or Y) and situating it strictly within individuals:

The organization of the individual is autopoietic and upon this fact rests all its significance: it becomes defined through its existing, and its existing is autopoietic. Thus, biology cannot be used anymore to justify the dispensability of the individuals for the benefit of the species, society or mankind under the pretense that its role is to perpetuate them.

Autopoiesis: the Organization of the Living, p 118

Given unity and perfect containment, autopoiesis isn’t a pragmatic approximation of how multiple interlocking systems function globally. It is an account of the necessary logical structure producing organs as a unity, so that something remains itself as organisation while also changing in its components and their relations.

This doesn’t mean that autopoiesis cannot fit into multiple systems. It implies that, when they do, there are constrictions due to the nature of autopoiesis: ‘Autopoietic systems may interact with each other under conditions that result in behavioural coupling.’ (Autopoiesis: the Organization of the Living, p 119-20)

For Maturana and Varela, internal organisation isn’t affected by the external uncertainty of altering and boundless environments. Their version of autopoiesis is a form of rationalism, based on unity according to logical structure. Though detected through empirical observation, it is an essential form, in the sense of how an organisation must work for something to be itself; such as a specific cell structures and processes uniting and producing different organelles within a membrane.

The lesson of the Chinese Room Argument is challenging for the distinction between allopoietic and autopoietic, because it raises the possibility of a process appearing to be autonomous, but registering changes in purposes and responding to them at the level of unifying organisation, thereby introducing external aims into internal processes.

Searle’s argument draws attention to the difference between mechanical rule following and a more flexible form of human understanding. The latter is meant to be more adaptable due to deeper insight, whereby a feel for changing circumstances leads to the production of components such as a new lexicon. Like an author correcting style in response to actual or perceived communication, this intuitive sense cannot be reduced to an automatic and rule-driven monitoring of changing variables.

Have I been understood? When Nietzsche asks himself this question, incessantly reworking texts and notes, there is a more thorough and deeper response to conscious and unconscious prompts than any stable form of productive homeostasis. His style evolves in touch with sensations of and guesses at external prompts, rejections, mistakes, adoptions, sympathies, provocations, emotional responses, communions and agreements.

If we can show external purpose working internally and adding something necessary – for instance, through a learning or editing component shared across inside and outside – then autopoiesis fails through the loss of unity. A child learning chess from a grandmaster is not autopoietic (though the unity of teacher and pupil might be). An author dependent on conscious and unconscious external prompts for the production of style is not autopoietic.

Allopoetic learning functions do not have to be human; for instance, in recent developments of deep learning. When learning depends on external factors that participate in the unity of internal processes, those processes cannot be autopoietic according to Maturana and Varela’s definitions.

It is important not to confuse impenetrability with autonomy for these arguments. A process might be impenetrable, in the sense of beyond understanding and explanation of its internal workings, but that does not imply that it is autonomous. A black box process might only be known through its inputs and outputs, yet its processes might well be dependent on outside components and processes. Impenetrable processes extending into a learning environment are allopoetic.

Maturana and Varela’s insistence on unity is a strong autopoietic condition. By stipulating perfect unity it implies domain containment and autonomy, avoiding external teleology and control over organisation for the lifetime of a particular form of unity.

Perfect unity does not imply that autopoiesis cannot be externally controlled; for instance, by changing a variable to achieve a particular output. It does not mean that autopoiesis cannot be transformed by external intervention on components or processes. When this occurs, outside control brings about a different form, whether autopoietic or not. Autonomy is not inviolability.

Beyond Maturana and Varela’s definition, autopoiesis is an unstable idea. I have concentrated on their rigorous definition, but there are less coherent versions of autopoiesis, where perfect unity is abandoned in favour of expansions such as extension, embeddedness, enactment, embodiment and emergence. These are associated with less strict identity and stability conditions, because identity covers expanded and changing external states.

Developing a theory of deep learning and autopoiesis, Rao Mikkilineni argues that ‘Autopoiesis refers to the behaviour of a system that replicates itself and maintains identity and stability while its components face fluctuations caused by external influences.’ (A New Class of Autopoietic and Cognitive Machines, 2)

Identity preserving behaviour is a more flexible condition than perfect unity through organisation. Unity is a stringent way of preserving identity and stability. It is a structural condition prior to any behaviour designed to preserve identity: ‘… the significant properties of the components must be taken in terms of relations, as the network of interactions and transformations into which they can enter in the working of the machine which they integrate and constitute as a unity.’ (Autopoiesis: the Organization of the Living , 77)

To count as autopoiesis, behaviour should correspond to relations of the production of components in a network – an organised process – that constitutes a unity. The emphasis is on these conditions, not on behaviour as such. Autopoietic autonomy is not consistent with many forms of behaviour preserving and relative identities, such as correlations, dependencies and symbioses.

We can describe behaviour aiming for identity and stability, consistent with external connections involved in production: the behaviour of child chess learners as they try to keep up with the master; the behaviour of the individual wolf as it runs with the pack; the behaviour of a deep learning machine dependent on external databases; the behaviour of a parasite and its host as they maintain joint and separate identities; the behaviour of hybrid biological or mechanical systems.

Mikkilineni augments his version of autopoiesis with distributed cognition: ‘Cognition requires the functions with the knowledge to receive information from various sources and process it into more knowledge that enables the system to manage its goals with respect to stability, safety, sustenance, and survival in the face of fluctuations in its interactions.’ (NCACM, 13)

For Maturana and Varela, cognition – as a property of autopoiesis – must be internal and restricted to the processes that define autopoietic unity and its development. There are many forms of cognition corresponding to different autopoietic unities, rather than one general type of cognition varying across different kinds of knower: ‘… for any autopoietic system its particular mode of autopoiesis determines its cognitive domain and hence its behavioural diversity…’ (Autopoiesis: the Organization of the Living, 119)

If cognition is explained through extension, embeddedness, enactment, embodiment or emergence (or any combination thereof) then autopoietic unity is interrupted by types of projection, reliance and interactivity. These diffuse processes are inconsistent with unity-driven autopoiesis, since they make it dependent on spread out and uncontrolled interactions that nonetheless form an intrinsic part of autopoietic organisation.

Maturana and Varela’s radical definition of life is on the individualist side of debates opposing theories of social interconnection to individualism, as dramatised in dystopian science fiction satire. We might think of the Borg in Star Strek as autopoietic machines, since capable of regeneration. This is a mistake. Each assimilated member participates fully in the cognition, development and stability of the wider Borg whole they are extended into.

There is no Borg individual. No member of the collective is a unity and even the collective as a whole is not autopoietic, when taken as an externally driven by the goal of perfection through absorption of other species, since its internal relations are determined by that goal. Taken as such, no machine of war or colonial apparatus can be autopoietic, because it is defined by what it seeks to conquer and exploit:

We are the Borg. Lower your shields and surrender your ships. We will add your biological and technological distinctiveness to our own. Your culture will adapt to service us. Resistance is futile.

Underpinning political and philosophical critique, there are two main advantages to studying autopoiesis through semiology. First, pragmatically, by taking the selection of key terms as a fundamental move, it avoids taking them as relative and negotiable. Unity is indispensable for an understanding of Maturana and Varela’s definition of autopoiesis. There isn’t a general understanding of autopoiesis that includes theirs, but rather many different and incompatible definitions to approach practically.

When built around precise definitions of sign and process, semiology distinguishes different positions adopting the same term but inconsistently, against analyses of broad ideological movements coalescing around vaguely defined yet banner terms such as autopoiesis.

This clash of interpretations comes out in my analysis of autopoiesis as unity, where positions based on cognitive and other extensions are shown to be incompatible with Maturana and Varela’s work. These other positions prove to be inconsistent, when compared to rigorous deductions from unity to autonomy and self-creation. The strict differentiation of autopoiesis and allopoiesis, also dependent on perfect unity, separates loose appeals to relative autonomy from Maturana and Varela’s account of self-production.

Second, formally, a process semiology takes account of the effects of selection against a background of all other elements. These consequences aren’t anodyne. The selection has positive and negative effects on the background it draws its terms from, where positive means reinforcing connections and negative means breaking them down. Coming prior to established values (moral, political, scientific, aesthetic, philosophical) the values of a selection describe a process, rather than judging it on antecedent bases.

The construction of theoretical ideas is taken to follow from selections of elements into sets that create novel signs {a, b, c,..}. Once these signs have been introduced, theories can be organised, constructed, spread and defended around them. This has effects on the linguistic and physical structures the sign then fits into.

For example, the idea that all things are predetermined is opposite to free will, chance and independence. Among many other effects, this opposition has consequences for law (responsibility), science (universal laws) and sociology (social determinism). When included in a sign, predetermination alters the potential of a wide context of ideas, weakening freedom and strengthening causal explanations.

For autopoiesis, when unity is selected to be part of its sign, a series of antonyms are avoided and relegated in terms of importance: disunity, discordance, confusion, disorganisation, disconnectedness, disjointedness, disturbance, incoherence and disproportion. These words are not only negations, as shown by the prefixes ‘dis’ and ‘in’, they are also usually given negative values, indicating states that are normally, though not always, defined as bad.

If we look further, the meanings of whole and joint associated with unity bring other opposites: divided, scattered, diffuse, discontinuous and fragmented. These are also frequently taken as negative; in particular, when taken to describe violent and unfortunate events that overcome unity. Maturana and Varela have built autopoiesis around a term with powerful positive connotations and equally strong contraries. Negatives such as dismemberment describe bad outcomes and threats to unity.

There is no essential negativity to any of the antonyms that unity has been selected against. This can be shown through a simple test. If we take unity itself to be negative, such as the unity of a menace or peril, then its division, fragmentation and diffusion are positive – as in the overturning of a reductive idea, the unmasking of a false image of sameness, the defeat of a totalitarian enemy, or the elimination of an obsessive foe. Though unity of autopoiesis benefits from and reinforces positive images of wholeness, identity and resilience, none of this is natural, essential or necessary.

Independent of the accuracy of autopoiesis as a description or as a consistent concept, Maturana and Varela’s idea rests on a selection of unity reinforcing preconceptions about the values of wholeness, autonomy, independence, stability and cohesion. If we fail to keep these semiological implications in mind when we adopt the term, we’ll unknowingly privilege powers of unity and autonomy over problematic multiplicities.