From organisational autonomy to reflexive autonomy

From organisational autonomy to reflexive autonomy: dimensions, architectures and the regulation of transformability

An organisation can maintain its conditions of existence, preserve the effects of its history, adjust its functioning and increase its capacity for anticipation without thereby becoming reflexive in the strong sense. Organisational autonomy and reflexive autonomy must therefore be distinguished. The first concerns the capacity of a system to take part in maintaining the relations on which its continuity depends. The second designates the possibility of making some of those relations accessible to examination, to comparison and, within certain limits, to transformation.

Reflexive autonomy is not the natural outcome of complexity. It depends on a particular organisation of memories, couplings, margins and regulations that makes certain constraints modifiable without dissolving the continuity of the system. A very complex organisation can remain rigid, locked into its routines or unable to act on the mechanisms it understands. Conversely, limited forms of reflexivity can appear in relatively simple organisations as soon as certain relations become comparable, revisable and accessible to intervention.

The framework proposed must not be understood as an already validated theory of the stages of autonomy. It is first of all a conceptual grammar intended to distinguish operations that are often confused: preserving a trace, mobilising a memory, representing a relation, intervening on a constraint and regulating the conditions of that intervention. These dimensions can reinforce one another, but they describe neither a necessary progression nor the successive degrees of a single axis. Their distinction rests on the function they perform and on the possibility that they evolve in a partly independent way.

Distinguishing dimensions without inventing ruptures

Abandoning an overly general vocabulary of thresholds does not mean that all capacities merge into a single continuum. A continuity of development does not necessarily abolish the difference between the operations performed. The distinction no longer rests on the supposed existence of a qualitative leap, but on the nature of the relation that becomes possible.

A trace designates the persistence of a modification produced by an earlier event. Memory appears when that modification can be reactivated, selected or mobilised in present functioning. Representation intervenes when a relation becomes available beyond the immediate situation in which it was produced and can be compared, transferred or used to anticipate other configurations. Interventional knowledge presupposes that a representation can orient an action on the mechanisms concerned and make it possible to examine its consequences. Reflexive regulation, finally, bears on the conditions under which those actions, rules and procedures of revision are themselves maintained, suspended or transformed.

These dimensions can progress together, but nothing requires them to. An organisation can accumulate considerable archives without integrating them into its decisions. It then has many traces but a limited functional memory. It can preserve effective routines without having an elaborate representation of the relations that make them effective. An institution can produce very precise analyses of its own functioning while being unable to modify the rules it describes. Its capacity for representation exceeds its power of intervention. An organisation may likewise have a considerable power of transformation while remaining unable to examine the criteria that orient its own changes.

These dissociations give a conceptual foundation to the distinction of dimensions. They do not yet demonstrate their empirical independence. That would have to be studied through comparisons able to show that one dimension can increase, decrease or hold steady without mechanically dragging the others along. The framework therefore formulates a hypothesis of functional separability, not an already established demonstration.

The word « threshold » should be retained only when a change of regime can actually be identified: the appearance of a new stability, the abrupt loss of a function, an irreversible shift or a dynamic of hysteresis. In other situations it is more prudent to speak of degrees, passages or functional configurations.

From trace to memory

Every organisation has a history, but not every history becomes a memory. A perturbation can leave a durable modification without that modification being reused or integrated into future responses. Memory begins when earlier effects become available to orient present functioning.

That availability takes several forms. A trace can automatically modify a response without being compared with other experiences. It can be reactivated only in a context close to the one in which it was formed. It can also be mobilised in new situations, confronted with other traces and integrated into more general regularities. Memory is therefore not a homogeneous state. It varies with the duration of preservation, the conditions of reactivation, the diversity of contexts in which it remains usable and the possibility of modifying its influence.

Preservation is not enough. An organisation that kept every modification with the same intensity would indiscriminately accumulate useful information, accidents and noise. An organisation that erased the effects of the past too quickly would remain dependent on the immediate. Functional memory rests on a tension between stabilisation, selection and revision.

Redundancy can protect certain functions, but it does not by itself explain learning. Different supports can contribute to partly common functions while retaining distinct sensitivities. That diversity sometimes makes it possible to maintain a continuity without reproducing the same limits everywhere. The organisation then preserves enough stability to persist and enough variation to modify its responses.

Memory does not represent a sudden rupture with the trace. It designates a transformation of its role: the past no longer persists merely within the organisation, it becomes available to orient its activity.

From memory to representation

Memory makes it possible to preserve differences and to compare trajectories. It does not guarantee an understanding of the relations that produce them. An organisation can repeat an effective response without identifying the conditions that make it effective. It can also modify its behaviour without distinguishing internal constraints from variations of the milieu.

Representation begins when a relation can be mobilised beyond its immediate manifestation. It need not be understood as a complete internal image of the world or of the organisation. It can be partial, distributed, embodied in rules of action or reconstructed from several traces.

Its reach depends notably on the capacity to link several temporalities and several levels. An organisation that processes only local deviations can correct its errors without understanding the global mechanisms that produce them. An organisation that observes only its aggregate results can ignore the local dynamics on which those results depend.

Integration between levels must avoid two extremes. Too weak a coupling fragments knowledge. Information stays locked inside subsystems that cannot relate their observations. Too strong a coupling propagates every variation to the whole and reduces the autonomy required for experimentation.

Multi-scale representation therefore requires an organisation able to preserve local differences while maintaining routes for comparison and coordination. It progresses as certain relations become identifiable, transferable and usable in several contexts. That progression may be gradual, but the operation performed remains distinct from mere preservation: it is no longer only a matter of being modified by the past, but of using certain relations to interpret absent or possible situations.

From representation to intervention

Representing a constraint does not mean having the power to modify it. An organisation can know the causes of a problem while remaining locked into the rules that reproduce it. This situation is frequent in institutions and companies. The knowledge exists, the effects are documented, the solutions are sometimes identified, but the mechanisms of decision, the economic dependencies or the distribution of power prevent their implementation.

Knowledge becomes interventional when it makes it possible to act on a relation and to examine the consequences of that action. That capacity demands more than a prediction. It presupposes a causal access, however limited, to the processes concerned.

Intervention too varies by degrees. An organisation can modify certain parameters without being able to transform its rules of coordination. It can reorganise procedures without touching the criteria by which its success is assessed. It can transform its means while leaving intact the ends that orient its choices.

Adapting a response, modifying a procedure, revising a rule and transforming a constitutive condition therefore involve neither the same risks nor the same capacities. These operations belong to a single continuity of transformation, but they do not bear on the same objects.

Interventional knowledge also requires a confrontation with consequences. A modification that improves a local indicator can weaken other functions or displace costs to another level. The quality of an intervention then depends on the extent of the field of effects taken into account. That extension of representation increases the possibility of anticipating and attributing certain consequences, but it does not by itself produce a moral obligation.

The constitutive core and historical lock-in

Every organisation depends more directly on some relations than on others. The constitutive core designates the set of processes whose interruption compromises the continuity of the organisation under study. It does not necessarily correspond to a centre, a management or an identifiable material component.

Identifying it requires several criteria. A function may be important because it directly supports viability. It may be so because it coordinates other functions. It may also appear indispensable only because the organisation was built around it and no longer has an available alternative.

This last possibility imposes an essential distinction between constitutive necessity and historical lock-in. An organisation may restore a function as a priority because it is genuinely indispensable. It may also mobilise all its resources to preserve a structure that has become dominant, costly or hard to replace.

The priority of restoration is therefore not a sufficient criterion. It can reveal what is vital, but also what the organisation protects out of inertia, dependence or concentration of power. An organisation can defend with equal intensity a function necessary to its continuity and a structure that compromises its future viability.

Identifying a core requires a comparison between several possible trajectories. One must examine what would happen if a function disappeared, if it were replaced or if its role were assured otherwise. A present dependency does not become necessary by the mere fact that it exists.

The assessment must also bear on several temporal horizons. A function can support immediate continuity while destroying the future conditions of viability. An economic system can maintain its production in the short term while exhausting the resources that make it possible. An institution can preserve its procedures at the price of a progressive loss of legitimacy, effectiveness or capacity to learn.

Adaptive restoration protects the functions that remain necessary to future continuity. Compulsive restoration preserves an organisation that has become unable to distinguish its own viability from the conservation of its habits. That difference cannot, however, be known with absolute certainty at the moment the decision is taken. It depends on a provisional assessment of the consequences, the alternatives available and the capacities for revision that are maintained.

Transforming the core

The constitutive core must not be conceived as an immobile foundation. The components of an organism are renewed. The people who make up an institution change. Technologies, procedures and infrastructures can be replaced.

Continuity rests less on preserving identical supports than on maintaining relations able to ensure certain functions. An organisation can therefore transform its core if it preserves the essential capacities during the transition.

Abrupt replacement increases the risk of rupture. A gradual transformation allows a new organisation to take charge of certain functions before the old one disappears. A period of overlap can then preserve continuity.

This strategy nevertheless has a cost. Two architectures must coexist, sometimes with incompatible rules. The old system can use its dominant position to absorb the resources required to build the new one. Overlap is therefore not automatically a protection. It can become a mechanism for prolonging the lock-in.

Continuity by relay does not mean that everything must be preserved until the replacement is complete. Some structures themselves become obstacles. The difficulty consists in distinguishing what must stay active to support the transition from what prevents that transition from becoming possible.

This distinction cannot be obtained through a universal rule. It requires a comparison between the functions maintained, the costs imposed, the alternatives available and the future possibilities that each trajectory preserves or destroys.

The margin of reconfiguration

Understanding the necessity of a change does not create the material conditions for carrying it out. An organisation whose resources are entirely absorbed by current operations has few possibilities of exploration.

The margin of reconfiguration designates the time, resources, skills, institutional spaces and degrees of freedom available for testing other organisations. In human systems it depends on budgets, decision rights, legal protections, possibilities of experimentation and the capacity to bear a temporary drop in output.

A large margin nevertheless guarantees no transformation. Resources can exist without being accessible to the actors able to use them. Pilot projects can be funded without having the power to modify the general rules. An organisation can encourage innovation while maintaining evaluation mechanisms that make any divergence impossible.

Transformability therefore depends on the relation between available resources, access to the rules and the possibility of integration.

An experiment entirely subject to the demands of existing operations often reproduces the constraints it seeks to overcome. An entirely isolated experiment sometimes produces solutions that cannot be integrated. The space of reconfiguration must keep a link with real operations while enjoying sufficient protection.

The margin also has a political dimension. It depends on the capacity to keep several trajectories open rather than concentrating all resources on the defence of a single organisation. A margin entirely mobilised to restore the existing system ceases to be a capacity for transformation and becomes an instrument of lock-in.

The risks of reflexivity

Reflexivity is no guarantee of progress. It increases the capacity for representation and transformation, but it can also produce new fragilities.

An organisation can become a prisoner of its own models. The more coherent its representations, the more it may be tempted to interpret events in a way that preserves that coherence. The model then ceases to serve the relation to reality and becomes a filter that protects the organisation against whatever might contradict it.

Reflexivity can also slow action down. The multiplication of scenarios, criteria and levels of analysis can make every decision contestable. An organisation that constantly re-examines its rules risks never stabilising the conditions required for learning.

Openness must therefore be modulated. Not all rules can be revised simultaneously. Operational adjustments can stay rapid, while deeper transformations require longer periods of observation.

This differentiation introduces a difficulty: what determines the rules that are provisionally stabilised at a given moment?

An answer consisting in adding levels of oversight indefinitely would produce an endless regress. Every rule would require a meta-rule, and then a rule charged with revising that meta-rule.

The vertical regress can be interrupted by a temporal limitation. A real organisation never examines all its rules simultaneously. It acts from a provisionally stabilised set while revising another. The rules of revision can themselves become accessible to transformation, but on a different timescale.

This solution does not, however, guarantee that the provisional base is appropriate. An organisation can stabilise precisely the rules that sustain its lock-in. Practical limitation produces continuity, not necessarily relevance.

The central question is therefore no longer to find an ultimate foundation, but to determine how a provisional base remains exposed to the possibility of its own correction.

No organisation can demonstrate from within that its fundamental rules are free of lock-in. It can only prevent them from becoming entirely immunised against contestation. A provisional base remains more open when it preserves return routes for the information that contradicts it, maintains available alternatives, protects part of its margins against total absorption, permits independent evaluations and makes it possible to compare its effects across several temporalities and several levels.

Conversely, a stabilisation becomes more probably compulsive when it turns every criticism into a threat, reduces the diversity of observations, concentrates the means of evaluation in the same bodies as those that take the decisions, and mobilises every available resource to prevent the appearance of alternatives.

These properties provide no absolute guarantee. They reduce the risk that an organisation confuses the conservation of its rules with the preservation of its viability.

The provisional base therefore cannot know with certainty that it is appropriate. It can only remain exposed to consequences it does not entirely control and maintain the conditions of its own revision.

From contestability to the effectiveness of contestation

Keeping a base open to criticism does not guarantee that this criticism can modify decisions. An organisation can preserve return routes for contradictory information, fund independent evaluations and maintain several alternatives without those elements exerting any real influence on its functioning. Information can be received without being judged decisive, alternatives recognised without ever being selected, and evaluations produced without modifying the rules that organise action.

Contestability therefore describes a condition of possibility, not an effective capacity for correction. It reduces the risk of immunisation without preventing inertia. An organisation can be procedurally open and substantively locked in.

The passage from available contestation to effective contestation depends on additional mechanisms. Contradictory information must be able to reach the places where decisions are taken. The actors who produce it must have sufficient capacity to demand a justification, to interrupt certain trajectories or to impose a re-examination. The results of an evaluation must be able to modify the criteria, resources or rules concerned. A contestation with no possible consequence remains peripheral information, even when it is officially acknowledged.

That effectiveness still does not guarantee that the decision finally adopted is just or appropriate. An organisation can seriously examine several objections and nevertheless maintain a destructive trajectory. No procedural device makes it possible to deduce the substantive relevance of a decision. Procedures can make errors more visible, increase their justificatory cost and preserve the possibility of a later correction. They abolish neither conflicts of value, nor uncertainty, nor the possibility of a choice deliberately unfavourable to certain viabilities.

The difference must therefore be maintained between three properties. A base can be contestable when objections can be formulated and preserved. Contestation becomes effective when it has real access to the mechanisms of decision. A decision becomes appropriate only if its consequences effectively support the conditions of viability retained. None of these levels follows automatically from the previous one.

The framework therefore does not settle the question of how to recognise a just base. It shifts the requirement towards a more limited question: which organisations prevent their own errors from becoming durably invisible or inconsequential? That shift does not close the normative problem. It only specifies the domain within which a correction remains possible.

Relativity to the boundary of observation

The notions of constitutive core, viability, memory, autonomy and lock-in all depend on the boundary adopted to delimit the organisation under study. That dependence does not concern artificial intelligence systems alone. It is a general property of the framework.

A company considered in isolation may appear autonomous in certain operations, while including its infrastructures, its suppliers, its workers or the institutions that support its activity reveals dependencies that the first delimitation made invisible. An organism can be studied as a unit, as a set of tissues or as an organisation embedded in wider ecological relations. To each change of boundary correspond other variables of viability, other constitutive relations and other possible forms of externalisation.

The boundary therefore cannot simply be assumed and then forgotten. It is an analytical decision that must be made explicit, justified and, where possible, compared with other delimitations.

This relativity does not mean that every boundary would be arbitrary. Some delimitations correspond to material discontinuities, to mechanisms of regulation, to capacities of maintenance or to stronger causal relations. They can be supported by observations. They nevertheless never become entirely independent of the question asked.

The constitutive core must thus be understood relative to a determined unit and scale. A function can be indispensable to the maintenance of a company without being favourable to the viability of the sector to which it belongs. A relation can sustain a technical system while increasing its dependence on an external infrastructure. An organisation can reinforce its local continuity by transferring its costs to the levels excluded from the initial boundary.

The analysis should therefore be repeated over several perimeters when the conclusions depend strongly on the delimitation chosen. Varying the boundaries then becomes a critical tool. It makes it possible to identify hidden dependencies, externalised costs and the properties that disappear or appear when the scale changes.

The framework does not describe an absolute autonomy. It examines degrees and forms of autonomy relative to an explicit boundary. That relativity is not a local defect to be corrected, but a condition of any analysis bearing on nested organisations.

Status of the framework and empirical hypotheses

The present framework belongs first of all to conceptual clarification. It proposes a vocabulary that makes it possible to distinguish phenomena often grouped under the general terms of learning, autonomy, adaptation or reflexivity.

That function must not be presented as an empirical validation. Defining memory as an organised mobilisation of the past, or distinguishing a constitutive necessity from a historical lock-in, is not yet a discovery about the real functioning of organisations.

The framework becomes empirical when it formulates relations liable to be contradicted by observation. Several hypotheses can be drawn out.

An organisation could have a substantial memory and an elaborate capacity for representation while remaining barely transformable when the actors who produce the knowledge have no access to the rules that structure decisions.

Spaces of experimentation that are both protected from immediate output requirements and connected to the places of decision should favour transformation more than projects entirely subject to existing routines or totally isolated from current operations.

An organisation that devotes the bulk of its margins to restoring its existing structures should find it harder to build replacement solutions, even when the limits of those structures are widely acknowledged.

A strong capacity for representation should not guarantee a better accounting for consequences when the mechanisms of evaluation reward local performance and render invisible the costs transferred to other levels.

These propositions can fail. Some organisations can transform their rules deeply despite a strong concentration of power. Protected experimental spaces can produce few effects. A small margin can sometimes accelerate a transformation by making the maintenance of the old system impossible.

Confrontation with such counter-examples would make it possible to determine whether the proposed distinctions have explanatory value or provide only a descriptive vocabulary.

Human organisations

Institutions and companies often show that knowledge is not enough to produce transformation. An organisation can have solid expertise, detailed indicators and a substantial memory while reproducing practices whose limits it knows.

The blockage may lie in the distribution of power, in economic dependencies, in existing infrastructures, in mechanisms of remuneration or in the criteria that define success. It is not, however, a general law. Some organisations fail because they have insufficient information, contradictory diagnoses or a weak capacity to learn.

The framework therefore invites us to locate the blockage rather than presume it. Does the organisation lack reliable traces, a memory able to integrate its experiences, a representation of causal relations, access to the rules, or margins allowing experimentation?

Reflexive capacity is not measured by the quantity of information held. It depends on the way that information circulates, is contested and can modify the mechanisms that orient functioning.

An expertise that remains advisory does not necessarily transform the rules. A local experiment that cannot influence general decisions remains peripheral. A leadership with considerable power can also fail when it no longer receives information able to contradict its own models.

Artificial intelligence systems

Current artificial intelligence systems have capacities of representation, learning and adaptation that can be substantial. They do not, however, have an organisational autonomy comparable to that of an organism.

The difference does not lie simply in dependence on external resources. All living systems depend on a milieu they do not entirely produce. The distinction concerns the relation of upkeep.

An organism uses external resources to maintain its own processes actively. Current artificial intelligence systems are designed, trained, evaluated, deployed, powered and repaired by human organisations. The essential conditions of their functioning are not merely external. They are organised by agents who define their objectives, select their data, modify their architectures and decide on their maintenance.

Their operational autonomy can be real within a limited domain. They can select actions, adjust certain parameters or use tools. That autonomy remains inscribed within an infrastructure whose continuity they do not themselves ensure.

Artificial intelligence systems should nevertheless not be studied as isolated entities. Their functioning belongs to socio-technical assemblages combining models, computing infrastructures, energy sources, databases, designers, operators, users, institutions and governance arrangements.

The choice of boundary therefore modifies the analysis. A model considered separately remains largely heteronomous at the constitutive level. A wider socio-technical system can ensure certain functions of maintenance, learning, control and reconfiguration. Extending the perimeter does not automatically turn the whole into an autonomous organisation. It shifts attention towards the real distribution of dependencies, of decision-making capacities and of maintenance functions.

Artificial autonomy is therefore neither a binary property nor an absolute independence. It can be partial, distributed and different depending on the functions considered. The relevant question is to determine which conditions the system effectively helps to maintain, which transformations it can initiate and which dependencies remain taken care of by other organisations.

To date, artificial intelligence systems remain mainly heteronomous at the constitutive level, even when they possess a substantial autonomy in carrying out certain tasks.

Coherence between levels and responsibility

An organisation always belongs to wider assemblages and depends on subsystems that have their own conditions of viability. A company depends on its workers, its suppliers, its infrastructures and the environments whose resources it uses. A society depends on local organisations while transforming their possibilities.

The viabilities of these levels can come into conflict. An organisation can preserve its functioning by transferring its costs to others. It can become more efficient at its own scale while degrading the conditions that support it.

Reflexivity therefore does not guarantee coherence between levels. An organisation can understand precisely the effects it produces and nevertheless choose to maintain them.

Representing consequences does not by itself create an obligation. The passage from a fact to a norm cannot be obtained by simply extending cognitive capacities.

Responsibility requires additional elements: norms, commitments, institutions, a capacity to act and a recognition of the interests affected.

Knowledge nevertheless modifies the conditions of accountability. An organisation that could reasonably have foreseen certain effects is not in the same situation as an organisation that had no means of knowing them. Representation widens the domain within which a justification can be demanded. It does not automatically produce the norm by which that justification will be judged.

Reflexivity therefore constitutes a possible condition of responsibility, not its sufficient foundation.

An organisation becomes responsible when its capacities of representation and anticipation are connected to rules that oblige it to take certain effects into account, to answer for its decisions and to integrate the consequences imposed on other levels.

That obligation may be moral, legal, political or institutional. It is not mechanically deduced from cognition.

Socio-ecological transitions as a case of application

Socio-ecological transitions illustrate the tension between immediate maintenance, structural transformation and future viability. A society can recognise that a trajectory is becoming unsustainable while remaining dependent on the infrastructures that support it.

Too rapid a rupture can weaken essential functions before alternatives are available. Too slow a transition can prolong dynamics that reduce future possibilities.

The preceding distinctions make the problem more precise. Must a function be maintained because it remains indispensable, or because a historical dependency still prevents its replacement? Do the resources devoted to restoring it protect the continuity required for the transition, or do they absorb the margins that would allow an alternative to be built?

The corridor of reconfiguration designates the period during which the old and the new can coexist without continuity being broken or transformation being indefinitely postponed.

This section is not an additional theoretical development. It shows how the distinctions between core, lock-in, margin and relay can be applied to a collective problem in which the preservation of certain functions must be articulated with the transformation of the structures that support them.

Empirical limit and the next test

The framework now has enough conceptual coherence for its main problem to be no longer internal. The distinctions between trace, memory, representation, intervention and regulation are defined by the operations they make possible. Their separability nevertheless remains a hypothesis.

The next advance can therefore no longer come from a new terminological refinement. It requires a case in which the dimensions are defined concretely enough to be able to vary separately and for that hypothesis to be able to fail.

A relevant study should begin with a single organisation, a single period and an explicitly justified boundary. For each dimension it should determine distinct observations. Traces could be identified by the documented preservation of past events. Functional memory would have to be assessed by the real influence of those events on later decisions. Representation would require showing that certain relations are recognised and used to anticipate new situations. The capacity for intervention would be established only when those representations make it possible to modify the mechanisms concerned. Reflexive regulation would bear on the possibility of revising the rules that frame those transformations.

The hypothesis of separability would be weakened if all these dimensions always varied together, if distinguishing them improved no description, or if the observations attributed to one dimension could not be distinguished from those attributed to the others. It would be strengthened by observable dissociations: an organisation that preserves more information without improving its use, that develops a more precise representation without increasing its capacity for intervention, or that increases its power of transformation while reducing its capacity to examine the criteria that orient that transformation.

A single study would not validate the framework. It would nevertheless make it possible to leave the domain of conceptual coherence and enter that of empirical constraint.

The text must therefore stop where that test begins. Its present function is to provide distinctions, to make certain confusions visible and to formulate relations liable to be examined. Its future value will depend on its capacity to survive cases in which its categories might prove redundant, insufficient or badly delimited.

Towards a reflexive autonomy

Reflexive autonomy is neither an automatic summit of complexity nor a property that would appear beyond a universal threshold.

It can be described through several functional dimensions: preserving the effects of history, mobilising a memory, representing certain dependencies, intervening on mechanisms, organising spaces of reconfiguration and regulating the conditions of that transformation.

These dimensions can reinforce one another, limit one another or evolve separately. An organisation can have a substantial memory and remain unable to modify its rules. It can transform its structures rapidly without understanding the effects produced. It can have elaborate models while refusing the information that contradicts them.

Reflexivity does not become more demanding because it reaches a higher degree of complexity. It becomes more demanding when it accepts that its own rules can never be guaranteed from an entirely internal point of view.

An organisation cannot know with certainty that its provisional base is not already a lock-in. It can only preserve the conditions that allow that base to be contested, confronted with its consequences and replaced when its effects become incompatible with the viabilities it claims to maintain.

That contestability nevertheless guarantees neither the effectiveness of criticism nor the relevance of the decision. An organisation can listen without modifying, compare without choosing and evaluate without transforming. Procedural openness reduces certain risks of blindness, but it replaces neither real access to the places of decision nor the examination of the consequences produced.

The capacity to transform one's constraints is no more sufficient to produce a responsibility. Responsibility appears when knowledge of effects is connected to norms, to obligations and to mechanisms of accountability.

Reflexive autonomy can then be defined as the capacity of an organisation to render some of its own constraints accessible to examination and transformation, to organise a revisability limited in time, to preserve a viable continuity throughout the change, and to keep open the conditions for correcting its own rules.

That capacity abolishes neither conflicts, nor uncertainty, nor constraints. It transforms the way they become visible, debatable and sometimes modifiable.

Between persisting and understanding there is not necessarily a rupture. Between understanding and transforming, no automatic continuity. Between transforming and answering for one's effects, no immediate logical passage. Between stabilising a base and knowing that it is just, no definitive guarantee.

Reflexive maturity perhaps lies not in possessing an assured foundation, but in the capacity to act without rendering one's foundations definitively inaccessible to revision.

The maturity of the framework itself now depends on the same requirement. It will no longer progress by adding a new layer of coherence, but through its exposure to situations able to resist its categories, to contradict its hypotheses and, if necessary, to impose their revision.