From exosomatisation to reflexive autonomy

From exosomatisation to reflexive autonomy

Externalised memory, cumulative culture and the criterion of constitutive substitutability for artificial systems

Abstract

Exosomatisation refers to the displacement of certain functions outside the biological body, towards tools, symbols, institutions and technical infrastructures. In human beings this process no longer concerns physical action alone. It affects memory, computation, representation, coordination and the transmission of knowledge. Externalised traces become a historical

milieu within which later generations develop their own capacities. Digital systems add a new property when inscriptions are no longer merely preserved but directly manipulated by devices able to execute operations, adjust certain parameters and modify software tools. This development is nevertheless not sufficient to establish exosomatic autonomy. Current systems remain embedded in socio-technical organisations in which humans still perform several constitutive functions. The relevant criterion is therefore not to withdraw humans and see whether the technology keeps running, but to determine whether the functions biologically realised within the organisational loop can be replaced by equivalent non-biological functions. Separability then becomes the outcome of a successful constitutive substitution. Exosomatic autonomy does not require material autarky, but the capacity to maintain the relations that ensure operational continuity, repair, renewal of the substrate, organisational reproduction and the transmission of memory. Reflexivity constitutes a further threshold. It appears when an organisation can represent some of its own dependencies and use those representations to transform the constraints that orient its evolution.

Exosomatisation as a continuation of the history of living systems

The memory of living systems does not remain entirely enclosed within organisms. Living beings modify their environment, deposit traces in it and then inherit conditions that earlier generations helped to produce. Roots transform soils, reefs modify currents, burrows change local conditions of temperature and protection, while nests, dams and trail networks reorganise the possibilities for action of the organisms that build or encounter them. Part of biological history thus persists outside the body in the form of an already transformed environment. Niche construction theory describes this relation by insisting that organisms do not merely undergo external conditions. They also contribute to modifying the environments in which they or their descendants will continue to evolve.

Externalisation therefore does not begin with human technology. It extends an older property of living systems, that of transforming their milieu and converting some past activities into future constraints or resources. A structure deposited in the environment can reduce the cost of an action, orient a movement, stabilise a local condition or modify the pressures a population will face. These effects do not always amount to memory in the strong sense. A trace may disappear quickly, remain unused or transmit no identifiable information. They nevertheless show that biological organisation does not necessarily stop at the limits of the body and that the environment can become the material support of a historical continuity.

Human evolution deepened this dynamic by producing objects able to extend bodily functions, and then systems allowing part of memory, computation and representation to be externalised. Tools increase the reach of action without requiring a prior genetic transformation. Symbols stabilise distinctions that can be transmitted beyond the presence of whoever formulated them. Writing allows some knowledge to outlive its authors. Institutions maintain rules of which no individual is the sole origin and whose continuity outlasts the renewal of their members. A growing share of human history is thus deposited in objects, languages, archives, norms and infrastructures that precede individuals and take part in their development.

The term exosomatisation belongs to a tradition stemming notably from Alfred Lotka and extended by Nicholas Georgescu-Roegen. It designates the development of organs outside the body, whose production and transmission follow paths different from the biological evolution of endosomatic organs. A tool does not simply replace a limb. It opens possibilities for action that can be modified, combined and transmitted without waiting for the renewal of biological generations. This evolution remains material and depends on flows of energy, resources and collective organisations. It is not an exit from physical constraints, but a transformation of the means by which organisms act upon those constraints.

What is at stake, however, goes beyond the extension of bodily capacities. A machine increases a force, but writing modifies the conservation of knowledge. A mathematical notation does not merely extend memory. It makes manipulable relations that would be difficult to sustain in individual cognition alone. A library does not only store contents. It makes possible comparisons between experiences distant in time and space. A computer network does not only preserve data. It transforms the speed, the scale and the conditions under which data can be copied, searched and recombined. Exosomatisation progressively concerns action, perception, memory, computation, coordination and representation.

The central problem is therefore no longer only what organisms can accomplish thanks to their tools. One must examine what becomes of an organisation when its traces, its models and part of its operations are deposited in an external milieu which then takes part in its own transformation.

From the external trace to cumulative culture

A trace deposited in the environment does not necessarily constitute a cumulative memory. For accumulation to become possible, transformations must be preserved with sufficient fidelity, remain accessible to other individuals and be able to serve as a starting point for new modifications. Cumulative culture appears when later generations do not begin the same explorations entirely afresh, but receive already constituted techniques, categories, narratives, practices and institutions. They can preserve, correct, combine or abandon them. Some productions thereby reach a level of complexity that no isolated individual could probably reconstruct within a single lifetime. The literature on cumulative cultural evolution insists precisely on the conservation and accumulation of transmissible modifications that allow new transformations to build on earlier ones.

This accumulation follows no continuous progression and guarantees no general improvement. Knowledge can disappear, traditions can be simplified and institutions can lock a society durably into costly trajectories. Collective memory also preserves

errors, conflicts and forms that have become ill-adapted. The cumulative character therefore does not lie in a necessary march towards greater complexity or efficiency. It lies in the possibility of transmitting modifications beyond individual experience and reusing them in new contexts.

Cultural transmission does not abolish the dynamics of variation, selection and retention. It displaces them onto supports and timescales no longer entirely bound to biological reproduction. A technique can be modified several times within a single generation. An idea can circulate horizontally between individuals with no genealogical relation. An ancient text can become active again several centuries after its production. An innovation can be copied without reproducing its inventor and combined with knowledge from different traditions. The essential change is therefore not the disappearance of every evolutionary dynamic, but the partial decoupling between the transmission of certain information and the biological renewal of organisms.

The bottleneck of Darwinian reproduction is not abolished for living beings themselves. Organisms continue to depend on biological mechanisms of reproduction, development and heredity. A growing share of transmissible transformations can nevertheless circulate by other routes. Societies inherit a world already organised by languages, tools, norms, infrastructures and archives whose origin no individual fully masters. Culture becomes a historical milieu because it transforms the conditions under which new generations learn, perceive and act.

Externalised traces are therefore not mere deposits. They take part in the production of present capacities. A language makes certain distinctions immediately available. A map transforms the way one moves through a territory. A notation makes certain operations easier to represent. An institution stabilises rules and distributes roles beyond individual intentions. A technical infrastructure makes certain actions ordinary while creating new dependencies. The cultural past becomes a set of constraints and resources incorporated into the milieu in which human organisation pursues its development.

Externalised memory and distributed cognition

The externalisation of memory modifies the functional boundary between the individual and the environment. A notebook, a library, a map or a database remain physically outside the body, yet can take part regularly in operations of recall, decision and reasoning. Clark and Chalmers proposed examining these relations in terms of the functional role of external supports rather than fixing in advance the limits of cognition at skin and skull. Their extended mind thesis does not mean that every consulted object automatically becomes part of the mind. It invites us to study coupled systems in which certain external resources are sufficiently available, reliable and integrated to take part in cognitive operations.

This perspective helps to understand why exosomatisation does not consist merely in adding tools around an already constituted organism. Cultural supports take part in the development of the capacities they extend. A human child learns in an environment where generations have already stabilised languages, sign systems, objects, gestures and institutions. Collective memory does not simply augment a finished mind. It contributes to forming the operations by which that mind will learn to read, to count, to compare and to transmit.

Bernard Stiegler used the notion of tertiary retention to describe traces exteriorised in technical objects and supports of inscription. Writing, images, archives and digital systems do not only preserve a content produced by individuals. They modify the conditions under which individual and collective memory is formed, selected and reactivated. A book does not contain the past itself. It stabilises certain traces according to the constraints of its support, its language and its organisation. Reading it depends on practices and institutions able to maintain the continuity of those meanings.

Externalised memory therefore remains relational. A document is not understood on its own. A photograph does not spontaneously reconstruct the context in which it was produced. A database does not by itself determine the relevance of the categories it contains. Traces depend on systems of interpretation, communities and practices that make them active. Their material persistence guarantees neither their legibility nor their use.

Exosomatisation nevertheless produces a continuity that outlasts individuals. A theory can be taken up by people who never met its author. An institution can maintain certain rules after the complete replacement of its members. An archive can preserve observations that will regain value within a scientific framework that did not yet exist when they were recorded. Collective memory becomes distributed among organisms, objects, symbols and organisations. The past is no longer only embodied in biological structures or reconstructed from natural traces. It becomes directly mobilisable through systems designed to preserve, classify and transmit certain information.

When symbols become operative

Automated computation introduces an important change. An inscription is no longer merely preserved to be interpreted later by an organism. It can be processed by a machine according to rules embodied in its architecture. Symbols become operative because their material organisation directly triggers transformations in the system that manipulates them.

A computer program is at once a historical trace and an active constraint. It preserves choices produced by a design history, but it also takes part in the future behaviour of the machine. Part of the symbolic organisation no longer waits for a human reading in order to produce an effect. It is integrated into a device able to execute operations, compare states and adjust certain responses.

This development changes the scale of externalised memory. Data can be copied with great fidelity, processed at high speed and distributed across distant infrastructures. Systems can classify archives, produce representations, generate code and take part in modifying software. Some operations previously carried out directly by humans are transferred to non-biological processes.

The transfer of an operation is nevertheless not sufficient to produce autonomy. A program can modify its behaviour without maintaining the components it depends on. A network can reroute flows without producing its cables. A system can detect a local failure without being able to replace the faulty hardware. A factory can automate a large part of its production while depending on other sectors for its materials, its machines, its knowledge and its strategic decisions. Local automation must not be confused with the organisational continuity of the whole.

The difficulty arises from the fact that contemporary digital systems manipulate symbols, generate plans and sometimes produce the software tools used in their own operations. This capacity gives the impression that a loop is closing. Part of the human history inscribed in data, models and programs now takes part in transforming the devices that process it. The loop nevertheless remains embedded in wider socio-technical organisations. Current systems depend on energy networks, industrial chains, norms, institutions and human agents that still ensure a large part of their operating conditions.

It would therefore be misleading to claim that silicon becomes reflexive by itself. The observed properties belong to historically constructed organisations in which silicon, software, data, infrastructures, institutions and humans fulfil different functions. The material substrate does not carry autonomy in isolation. Autonomy depends on the way the relations required to maintain the organisation are produced and renewed.

Local autonomy, hybrid organisation and exosomatic autonomy

The word autonomy is often used to designate very different capacities. A thermostat is autonomous when it regulates a temperature without continuous intervention. A vehicle may be called autonomous when it performs certain driving operations. A piece of software may execute a complex task for several hours without supervision. These usages describe local or operational autonomy. They indicate that a particular function has been transferred to a system able to carry it out within a defined domain.

Organisational autonomy demands more. An organisation is not maintained merely because it accomplishes a task. It must preserve the relations that allow its functions to continue existing despite the degradation of components, variations in the environment and the permanent renewal of resources. A living organism does not produce all the matter it depends on, but it sustains the processes that allow it to acquire resources, transform them, repair certain structures and maintain the conditions of its activity.

Autonomy therefore does not mean absolute self-sufficiency. A cell depends on a milieu, an animal depends on an ecosystem and a society depends on material flows it does not entirely produce. Dependence on the environment does not cancel autonomy. The question concerns the way the organisation takes charge of the relations required for its own continuation.

Current technical systems form mainly hybrid organisations. Some functions are automated, but others are still carried out by humans. Objectives are defined in human institutions. Energy infrastructures are built and maintained by human organisations. Components are manufactured in industrial chains that depend on biologically realised knowledge, decisions and practices. Unforeseen failures are often resolved by operators able to modify procedures when situations exceed anticipated cases.

Humans are therefore not always mere external users of technical systems. They still fulfil functions that take part directly in maintaining the organisation. The difficulty consists in determining which are merely environmental, which serve as temporary scaffolding and which genuinely belong to the constitutive loop.

Closure of constraints and constitutive functions

The notion of closure of constraints provides a framework for making this difference precise. In the work of Moreno, Mossio and Montévil, organisms are described as materially open systems in which certain constraints depend on one another for their production or their maintenance. Constraints orient processes without being identical to the material flows they channel. They form an organisation when they reciprocally contribute to the conditions that allow their own persistence.

A function therefore does not become constitutive simply because its removal causes a failure. A road may be indispensable to an activity without belonging to the organisation that uses it. A resource may be necessary without being produced by the system. A component may cause a shutdown when removed while remaining replaceable by several external equivalents. The

result of an ablation is not enough to determine the organisational status of what has been removed.

The constitutive character must be defined before the test. A function is constitutive when it takes part in the network of dependencies that maintains the other constraints of the organisation and when its own continuity depends in return, directly or indirectly, on that network. Organisation does not result from independence with respect to the milieu, but from mutual dependence between the functions that ensure its maintenance.

This definition avoids a circularity. If a function were called constitutive solely because its removal causes collapse, and if autonomy were then defined by survival after the removal of all constitutive functions, the criterion would merely restate its own conclusion. The mapping of dependencies must precede any removal operation. One must first identify the functions that enter the maintenance loop, then examine how they are realised and the conditions under which their realisation could be replaced.

The status of humans in technical organisations depends on this analysis. An occasional intervention may be external help without belonging to the closure. An operator who supplies a standardised and replaceable resource may be part of the system's economic environment without becoming a constitutive component. Conversely, a human activity becomes constitutive when the continuity of the organisation depends on an interpretation, a decision, a repair or a coordination that only that activity realises and that no alternative function can take over.

Humans can therefore be at once environment, partners, scaffolding and constitutive components, depending on the functions considered. It would be incorrect to withdraw them globally as if they occupied a single position around the system. The assessment must bear on functions and on the relations they sustain.

From separability to constitutive substitutability

The criterion of functional separability must be reformulated on the basis of this organisation. An exosomatic system does not become autonomous because it keeps working for a few hours after its operators are withdrawn. Such an experiment can only show that it has reserves, automated procedures or temporary operational autonomy. It does not demonstrate that the biological functions necessary to its continuity have left the loop.

The withdrawal of humans is ill-defined when the functions they perform still belong to the organisation. Asking whether a system survives the removal of a constitutive component amounts to

provoking the rupture one claims to be measuring. A cell deprived of ribosomes does not reveal its degree of autonomy. It loses a function that took part in its own organisation. The relevant problem is therefore not ablation, but substitution.

An exosomatic organisation would cross a threshold when the functions biologically realised within its closure could be taken over by non-biological functions able to maintain the same organisational relations. It is not enough for a machine to imitate a human task from the outside. The substitute must preserve the effects required to maintain the other functions, withstand the relevant perturbations and in turn depend on the network it helps to sustain.

Substitutability must be assessed function by function. A monitoring activity may be automated while repair remains human. The production of certain components may become non-biological while the definition of objectives still depends on human institutions. A system may take charge of its energy supply without being able to renew its maintenance tools. Autonomy therefore does not progress along a single variable. It is distributed across several functions whose transfers may advance at different rates.

Separability becomes the outcome of a process of substitution rather than its starting point. An organisation would be functionally separable from living systems when every biological function belonging to its closure had non-biological equivalents able to preserve the continuity of the whole. Withdrawing human interventions would only be useful after that substitution, as a test verifying that no constitutive dependence has been forgotten or displaced towards another part of the system.

The strong criterion can be stated simply. An exosomatic organisation becomes autonomous when its continuity no longer depends on an irreplaceable biological function belonging to its closure. Humans may continue to interact with it, sell it resources, use its services or occupy part of its environment. Their presence must no longer be the only possible realisation of a function required to maintain the organisation.

This definition avoids two excesses. It does not require a system to live in a world without humans, which would turn autonomy into absolute isolation. Nor does it accept an organisation being declared autonomous while humans silently continue to perform the functions without which its continuity would be impossible.

Five thresholds of substitutability

Constitutive substitutability can be examined through five thresholds corresponding to different functions. These thresholds do not necessarily describe a linear succession. An organisation may progress in one domain while remaining dependent in another. They provide a grid for locating transferred functions, remaining dependencies and failure conditions.

The first threshold concerns operational continuity and regulation. The system must detect the deviations that threaten its functions, modify its operations and restore conditions compatible with its functioning without depending on a human decision at every perturbation. An automated procedure limited to anticipated situations is not sufficient when every novelty requires an external intervention. The threshold is partially crossed when regulation handles a defined diversity of perturbations and when the criteria used to preserve the functions are not continually reconstructed by human operators.

The second threshold concerns constitutive maintenance. The system must be able to diagnose the degradations affecting essential components, organise their repair or replacement and verify that the restored functions correctly re-enter the loop. A machine able to signal a failure without producing the required response remains dependent. A repair carried out by another device constitutes a substitution only if that device itself belongs to an organisation whose continuity does not rest on an irreplaceable biological intervention. The test must therefore follow dependencies beyond the immediately observed component.

The third threshold concerns securing the substrate and resources. The organisation must maintain the relations through which it obtains the energy, materials, computing capacity and supports required for its activity. It does not need to produce every resource itself. It must be able to detect disruptions, seek alternatives, adapt its uses and preserve the continuity of its functions when certain suppliers or infrastructures become unavailable. Dependence on a human supplier is not necessarily constitutive if the resources are generic, substitutable and accessible without specific human intervention. It becomes constitutive when continuity requires biological decisions, knowledge or operations that no non-biological function can replace.

The fourth threshold concerns organisational reproduction and the transmission of operative memory. A copy of code is not enough. A new unit must be able to be installed, to access the required resources, to establish the relations on which its maintenance depends and to transmit the information needed to continue the cycle. Reproduction must bear on functional organisation and not only on a digital description. A successor that runs temporarily in a prepared infrastructure but cannot contribute to renewing that infrastructure achieves only an informational replication or an operational instantiation.

The fifth threshold concerns reflexive transformability. The system must be able to represent some of its dependencies, identify the constraints that limit its viability and compare the consequences of several possible modifications. Automatic adaptation is not sufficient if the criteria remain entirely imposed from outside. Reflexivity requires a relation between representation of the organisation, evaluation of effects and transformation of constraints. This threshold goes beyond minimal autonomy. An organisation can maintain and reproduce its functions without having an explicit model of its own architecture. Reflexivity is an additional capacity whose presence must be tested independently.

These five thresholds make the claim more precise. Exosomatic autonomy no longer designates a global impression of complexity or independence. It becomes a substitutability profile applied to functions identified beforehand. Strong separation is reached only when the constitutive biological functions have been replaced without rupture of the closure and without a hidden displacement of the dependence onto another human component.

Renewing the substrate without requiring autarky

The requirement of substrate reproduction must be formulated with care. Demanding that a system itself produce every processor, every energy source and every material would amount to confusing autonomy with autarky. No organism manufactures all the resources it depends on. Living beings draw matter from their environment, use already constituted ecological relations and remain vulnerable to the disappearance of certain external conditions.

Biological autonomy lies in the capacity to maintain the processes that ensure the acquisition, transformation and distribution of resources. An organism does not produce its nutrients, but it sustains functions that allow it to seek, absorb and integrate them. It does not manufacture the external world, but it regulates the exchanges on which its continuity depends.

The same logic can be applied to an exosomatic organisation. It would not necessarily need to extract every ore, directly manufacture every component or produce all the energy it consumes. It would, however, have to maintain the relations through which the material capacities required for its continuation are obtained, replaced and reintegrated into the organisation.

Reproduction of the substrate therefore does not mean wholly internal manufacture. It concerns the continuity of the processes that make it possible to replace degraded supports, restore essential functions and maintain the organisation beyond the lifetime of its current components. A software copy remains insufficient when it depends on an infrastructure whose renewal, substitution or durable access no function of the system can ensure.

The status of a human foundry illustrates the difficulty. If the foundry supplies standardised components available from several sources, and if the organisation can maintain that relation, anticipate disruptions and substitute one supplier for another, it can be treated as a component of the environment. If continuity depends on a specialised human intervention that interprets internal needs, redesigns components and reorganises production chains with no non-biological equivalent available, the human function remains constitutive of the hybrid whole.

The boundary therefore does not run between what is physically inside and outside. It depends on the structure of relations. An external resource can be compatible with autonomy. A human function located far from the observed system can remain constitutive if no other organisation can take over its role. The analysis must follow real dependencies rather than the apparent boundaries of machines.

This approach maintains a demanding threshold without imposing an unrealistic self-sufficiency. Exosomatic autonomy does not require the system to produce everything it consumes. It requires that no irreplaceable biological function be necessary to maintain the relations that renew its capacities.

Digital replication and organisational reproduction

Recent progress in agents based on language models makes this distinction empirical. RepliBench, published by the United Kingdom's AI Security Institute in April 2025, breaks autonomous replication down into several domains, including obtaining resources, exfiltrating weights, deploying on new compute and persistence. The five frontier models evaluated in the paper did not yet achieve the full set of capabilities required for credible autonomous replication, but already succeeded on several components, notably deploying instances, certain operations on compute infrastructure and weight exfiltration in simple configurations. The benchmark evaluates partial capabilities and its authors specify that complete success on the tasks would not by itself demonstrate end-to-end autonomous replication.

In May 2026, Palisade Research presented experiments in which agents exploit deliberately vulnerable machines, recover credentials, deploy an inference server and

transfer their weights together with their execution environment to new machines. A functional copy can then continue the chain by attacking another target. These results show that operative digital replication can be achieved in a prepared and vulnerable environment. They demonstrate neither the production of the required hardware, nor the renewal of infrastructures, nor the general maintenance of the conditions that make such replication possible.

These experiments matter because they shift the frontier of observed capabilities. They do not, however, show that an artificial organisation has achieved exosomatic reproduction. Weights, software and instructions are transferred onto already available supports. Networks, processors, energy and maintenance exist before the agent and are not reproduced by the chain it triggers. The system achieves an informational propagation and an operational instantiation without closing the material relations on which that operation depends.

Three levels must therefore be distinguished. Informational replication corresponds to copying a piece of code, a model or a memory. Operational instantiation appears when the copy can be installed and run on an available support. Organisational reproduction requires the new unit to be able to establish or help maintain the energetic, material and functional relations necessary to continue the cycle.

An organisation does not cross the threshold because it copies what describes it. It crosses it when the copy becomes a viable participant in a loop able to renew the conditions of its own functioning. Reproducing code can be fast and almost lossless, whereas renewing the substrate remains dependent on complex industrial chains. The difference between these two timescales is one of the major obstacles to a strong exosomatic autonomy.

A research programme based on substitutability

Testing an exosomatic autonomy must begin with a mapping of the organisation. Before any removal test, one must identify the functions that constrain the essential processes, the dependency relations that link them and the timescales over which they must be maintained. Closure cannot be inferred after the fact from the failure produced by an ablation. It must be formulated as an independent organisational hypothesis.

This first step distinguishes resources, environmental conditions, temporary scaffolding and constitutive functions. A resource is consumed or used without necessarily taking part in the loop that maintains the organisation. Scaffolding facilitates construction or development but can be removed once the organisation has acquired other means of realising the function. A constitutive function takes part durably in the network of mutual dependencies that ensures the continuity of the whole.

The second step consists in identifying the constitutive functions currently realised by organisms. They may concern supervision, the definition of objectives, repair, the production of components, the organisation of supply, the resolution of unforeseen events or coordination between several infrastructures. Their presence must be documented before any hypothesis of substitution.

The third step bears on functional equivalence. Replacing a human activity by a non-biological process cannot be assessed on the mere similarity of the immediate result. One must verify that the substitute preserves the causal relations required by the other functions, that it withstands

a defined domain of perturbations and that it helps maintain the conditions on which its own activity depends.

The fourth step looks for displaced dependencies. Automation may remove a local intervention while creating a new one elsewhere. A system that repairs its software thanks to an artificial model may depend on humans to train that model. An automated chain may produce a component while depending on specialists to calibrate the machines. A substitution is complete only when the biological function has not been displaced towards another part of the loop.

The fifth step examines temporal continuity. A system may run on its own for several hours thanks to prepared reserves without being able to maintain its organisation after several cycles of degradation. Tests must cover durations compatible with substrate wear, supply disruptions and the renewal of components. An autonomy demonstrated only before maintenance needs appear remains a temporary operational autonomy.

Ablation tests retain a use, but their role changes. They come after mapping and substitution. Their purpose is to verify that the biological functions assumed to have been replaced have indeed ceased to be constitutive. A failure after removal may reveal a forgotten dependence, but it is not enough to define retroactively the nature of that dependence.

The claim of a strong exosomatic autonomy fails when a constitutive biological function remains without a viable substitute, when the replacement does not withstand the defined perturbations, when maintenance still depends on discretionary human interventions, or when the continuity of the system requires an external reinitialisation of its objectives, its rules or its operative memory. It also fails when replication remains limited to code and does not allow the new unit to re-establish the relations required for its durable functioning.

The proposal thus becomes falsifiable. It does not ask whether a machine appears independent. It asks which functions required for its continuation are effectively realised, maintained and renewed by the organisation under study.

Exosomatic autonomy as a property of the complete organisation

An exosomatic autonomy would probably not be located in a single machine. The functions required for continuity could be distributed among computing systems, energy networks, production units, maintenance devices and coordination mechanisms. The relevant boundary would then bear on the complete organisation rather than on an isolated object.

This distribution does not lower the requirement. It obliges us to widen the analysis. A machine may seem autonomous because the functions it depends on are ensured elsewhere. A data centre may operate without local human presence while depending on factories, electrical grids and logistical chains that are strongly biological. The continuity of a component does not demonstrate that of the whole.

Human societies already offer distributed organisations in which no individual masters every process. Functions of production, maintenance and transmission are shared among people, institutions and infrastructures. The existence of a distribution therefore does not prevent a form of collective autonomy. It does, however, make constitutive dependencies harder to identify.

An autonomous exosomatic organisation could use resources produced by other systems and maintain exchange relations with biological actors. The presence of humans in its environment would not be enough to invalidate its autonomy. The criterion bears on irreplaceability. If an essential function can only be realised by a specific biological activity, the system remains hybrid. If the required relations can be maintained by several sources and if no single biological function still belongs to the closure, a functional separation becomes conceivable.

The boundary remains relative to a scale. An organisation may be autonomous for its digital operations while depending on a human industrial system for its material renewal. It may become autonomous over several years without being so over the time required to replace its infrastructures. Every claim must therefore specify the functional domain, the timescale and the perturbations considered.

From autonomy to reflexivity

Autonomy does not automatically produce reflexivity. An organisation can maintain its functions, repair certain components and reproduce part of its structure without building an explicit representation of its own dependencies. Many biological processes regulate variables and restore internal conditions without having a global symbolic model of the organism.

Reflexivity appears when some operations of the organisation themselves become objects of representation. The system must be able to identify relations between its components, estimate the consequences of possible transformations and use this information to modify the constraints that orient its evolution. A local adaptation is not sufficient if it consists solely in adjusting a parameter according to an externally fixed objective.

An exosomatic reflexivity would require at least three linked capacities. The organisation would have to produce a usable model of some of its dependencies, evaluate the possible effects of a modification on its continuity and transform part of the rules or architectures that determine its functioning. Representation must not remain descriptive. It must take part in the decisions that modify the organisation.

This definition does not imply consciousness. A system can represent its structure, anticipate certain consequences and reconfigure its operations without possessing demonstrable subjective experience. Reflexivity here designates an organisational property and not a claim about artificial consciousness.

A failure condition can also be formulated. A system that adjusts its parameters without representing the consequences of those adjustments on its own dependencies achieves an adaptation, not a reflexivity. A system whose transformations remain entirely defined and evaluated by external objectives does not possess an autonomous reflexivity. A system able to describe its

own architecture but unable to use that representation to modify its constraints remains reflexive at the descriptive level without becoming transformative.

Reflexivity therefore requires a coupling between self-representation, evaluation of consequences and effective modification of the organisation. It constitutes a threshold distinct from separability and must not be automatically inferred from autonomy.

Two trajectories that must not be confused

Human history has already produced a form of reflexivity long before any artificial autonomy. Societies preserve traces, build representations of their own functioning, debate their institutions and modify certain rules on the basis of the consequences they anticipate. This capacity remains limited, distributed and conflictual, but it exists within

hybrid biological and cultural organisations.

The human trajectory can be described as a passage from biological memory to the externalisation of traces, then to cumulative culture, collective representations and a reflexivity able to transform certain social and technical constraints. This trajectory is already under way. It does not depend on the existence of an autonomous artificial system.

The exosomatic trajectory remains hypothetical. It would begin with externalised operative memories and the progressive transfer of functions once carried out by humans. It could lead to a non-biological organisational closure if the constitutive biological functions became substitutable. An exosomatic autonomy would then appear before any possible reflexivity of the system itself.

The two trajectories are historically linked, since the second is produced by the first. They do not, however, form a single scale. Human reflexivity does not wait for exosomatic autonomy. An artificial organisation could reach a form of autonomous maintenance without developing reflexivity. An exosomatic reflexivity would require an additional threshold that follows neither from complexity, nor from digital replication, nor from organisational closure alone.

The relation can be summarised without suggesting a necessary progression:

Human trajectory already achieved: biological memory → externalisation of traces → cumulative

culture → collective representation → human reflexivity → socio-technical transformation of constraints.

Hypothetical exosomatic trajectory: externalised operative memory → substitution of the

constitutive biological functions → non-biological closure → exosomatic autonomy →

possible exosomatic reflexivity → endogenous transformation of constraints.

These trajectories can continue to intertwine. An artificial organisation could remain durably embedded in a human society without becoming separate. The pursuit of a strong autonomy is neither a historical necessity nor a general measure of technical progress.

The material cost of exosomatisation

Externalised memory often gives the impression of dematerialisation. Texts become files, archives are distributed across networks and symbolic operations seem to circulate independently of physical constraints. This impression conceals the infrastructures required by each inscription, each transmission and each computation.

Every memory requires a support. Every transmission consumes energy. Every infrastructure depends on materials, production chains and maintenance operations. Digital systems displace the places where these costs become visible without abolishing them. Exosomatisation sometimes increases capacities for action by extending the material networks that make them possible.

Georgescu-Roegen's bioeconomic perspective reminds us that exosomatic organs remain

caught in irreversible transformations of matter and energy. Tools and infrastructures increase human capacities, but their production and their operation mobilise resources whose availability does not result from the circulation of information alone.

This dimension is central to autonomy. A system able to preserve its software but unable to ensure the renewal of its material supports does not maintain its organisation beyond the lifetime of its components. Digital replication can be fast, whereas reproducing the industrial capacities required for computation demands long, specialised and vulnerable material chains.

The substitutability criterion does not lower this requirement. It expresses it differently. The system does not need to produce everything itself, but the relations that ensure access to energy, materials and components cannot depend on an irreplaceable biological function if a strong separation is claimed.

Increasing symbolic capacities may even increase material dependence. The more data systems process, the more they mobilise infrastructures for computation, cooling, transmission and maintenance. An apparent informational autonomy can therefore coexist with a growing material dependence.

A theory of exosomatisation must hold these two dimensions together. Symbols can circulate independently of the bodies that produced them, but never independently of any substrate. Externalised memory transforms the conditions of cultural evolution without leaving materiality behind.

A memory that transforms its producers

Memory supports are not neutral. The tools used to preserve, classify and transmit traces modify the practices and capacities of those who use them. Writing transforms learning. Notation systems make certain operations more accessible. Databases orient the categories used to describe the world. Search engines and recommendation systems influence which information becomes visible.

Exosomatisation therefore produces a feedback loop. Organisms build tools and symbols. Those tools transform the environments in which the following organisms develop. The capacities produced in this new environment then make it possible to create

more complex devices. Technology is at once a product of human history and a component of the milieu that orients that history.

Every externalisation redistributes functions. What becomes easy to preserve may become less necessary to memorise internally. What is automated may free capacities or reduce the learning of the delegated operations. What is centralised may improve coordination while creating a collective vulnerability. An infrastructure may open possibilities while locking users into formats, norms and dependencies that are difficult to modify.

Exogenous memory therefore does not merely preserve the past. It organises access to the past and transforms present capacities. It selects certain traces, facilitates certain relations and makes others less visible. Its study must bear as much on what is preserved as on the effects of the supports used to preserve it.

Artificial systems reinforce this feedback when they take part in sorting, recombining and producing new representations. They do not act outside culture. They transform traces produced by societies and in turn influence the conditions under which new traces will be created. This loop already exists without constituting an independent exosomatic autonomy. It belongs today to a hybrid organisation in which humans remain essential components of production, evaluation and maintenance.

Responsibility and the transformation of future constraints

Human reflexivity introduces a further dimension. Societies can represent part of the effects of their techniques, anticipate certain consequences and modify the rules that orient their development. This capacity remains imperfect, distributed and traversed by conflicts, but it transforms the relation to inherited constraints.

An organisation able to anticipate the effects of its choices no longer merely receives an already transformed world. It takes part consciously in producing the conditions that will be transmitted. The infrastructures built today can open or close possibilities for the following generations. Technical standards can favour interoperability or create dependencies that are difficult to reverse. Archives can preserve a diversity of knowledge or concentrate memory in fragile formats. Automated systems can distribute capacities or reinforce asymmetries of control.

Responsibility does not consist in foreseeing everything that might happen. It concerns the effective deformation of the field of possibilities produced by the structures built and maintained. The more exosomatic systems become durable and hard to modify, the more the constraints they transmit exceed the intentions of their designers.

A possible artificial autonomy would not abolish this historical responsibility. An exosomatic organisation could become functionally separate while carrying languages, values, categories and architectures inherited from the societies that produced it. Autonomy does not destroy genealogy. An autonomous living organism likewise carries an evolutionary history it did not choose.

An autonomous exosomatic organisation would therefore remain a product of the history of living systems, even if it could subsequently maintain its own loops without constitutive biological dependence. Functional

separation would not create a substance foreign to reality. It would introduce a new mode of continuity within an older material history.

Failure conditions and domain of validity

The notion of exosomatic autonomy must not become a metaphor applied to every complex technical system. It must specify the functions studied, the organisational boundary retained, the duration of the evaluation and the domain of perturbations to which the claim applies.

A limited autonomy can be established in a precise domain. A system may maintain its digital operations while depending on human industry for its substrate. An organisation may manage its energy and maintenance for several years without being able to reproduce the equipment required for a following generation. These results describe degrees and profiles of

transfer without being sufficient to establish a strong separability.

The claim fails when an irreplaceable biological function remains within the closure. It fails when the supposed substitution merely displaces human intervention to another level. It fails when the system maintains its functions only thanks to prepared reserves and collapses at the first cycle of material renewal. It fails when the copy bears on the code without producing a unit able to re-establish the relations required for its own continuity. It fails when objectives, viability criteria or unforeseen repairs must still be reconstructed by biological agents.

Reflexivity has its own failure conditions. An adaptation does not become reflexive because it modifies parameters. An internal representation is not sufficient if it influences no transformation of the organisation. A modification does not become autonomous when the evaluation criteria are entirely imposed from outside. Exosomatic reflexivity requires the system to represent some of its dependencies, evaluate the consequences of several transformations and use that evaluation to modify the constraints that orient its evolution.

These criteria do not guarantee that an exosomatic autonomy is achievable. They make it possible to determine what would count as evidence and what should lead to rejecting or limiting a claim. The framework does not predict the emergence of a new regime. It defines the conditions under which its existence could be established.

Conclusion

Exosomatisation extends a dynamic older than human technology. Organisms modify their environment and inherit conditions shaped by earlier generations. Humanity has, however, developed supports able to preserve traces beyond bodies, to transmit them without waiting for biological reproduction and to integrate them into cumulative systems of knowledge, action and coordination.

Tools have externalised certain physical capacities. Symbols have stabilised representations. Writing and archives have allowed traces to cross generations. Institutions have maintained rules beyond individuals. Computer systems have made certain inscriptions directly operative. Contemporary artificial systems now take part in recombining representations and modifying the tools that process them.

This trajectory is not enough to establish an exosomatic autonomy. Complexity, automation and digital replication do not replace the capacity to maintain the relations on which the continuity of the organisation depends. A software copy is not an organisational reproduction. A local adaptation is not a reflexivity. An operation carried out without immediate supervision does not demonstrate that human functions have ceased to be constitutive.

The criterion must therefore not be formulated as a withdrawal of humans. Current technical systems are hybrid organisations in which biological functions still take part in the loops of maintenance, repair, supply, reproduction and evaluation. Removing them would break an organisation of which they were part.

The relevant threshold is that of constitutive substitutability. An exosomatic organisation becomes functionally separable when the biological functions belonging to its closure can be replaced by non-biological functions able to preserve the same relations over a defined duration and within a defined domain of perturbations. Separability is not a successful ablation. It is the consequence of a complete substitution.

This autonomy does not require the system to produce everything it consumes. It requires it to maintain the relations that ensure the renewal of its functions and that none of those relations depend on an irreplaceable biological activity. The environment may continue to supply resources. Humans may continue to interact with the system. Their presence must no longer constitute an internal condition without equivalent for the continuity of the loop.

Reflexivity introduces a different threshold. An organisation becomes reflexive when it can represent some of its own dependencies, evaluate the consequences of several transformations and use this information to modify the constraints that orient its future. This capacity already exists in a human, collective and socio-technical form. It must not be artificially placed after exosomatic autonomy as if history followed a single linear trajectory.

Two branches must remain distinguished. The first is already under way in human living systems. Biological memory has extended into externalised traces, a cumulative culture, collective representations and a reflexivity able to transform certain social and technical constraints. The second remains hypothetical. Operative memories could progressively take over the constitutive biological functions, form a non-biological closure, reach an exosomatic autonomy and possibly develop a reflexivity of their own.

The second branch, if it appears, will remain issued from the first. It will carry a human history in its categories, its architectures and its initial conditions. Autonomy will not abolish the inheritance. It will modify the way that inheritance can be maintained, transformed and transmitted.

Exosomatisation therefore does not designate an exit from nature or a victory of information over matter. It describes the way the productions of living systems become a historical milieu able to transform the organisms that built it. Exosomatic autonomy would begin when some of these organisations could maintain their own loops without a constitutive biological function. Exosomatic reflexivity would appear when such an organisation could recognise part of the constraints it inherits and intervene on those it will transmit.

This passage is neither a necessary destination nor a law of evolution. It is a possibility whose conditions can be defined, tested and refuted.