Enactive Regulation Theory: How Cognitive Systems Sustain Sense-Making Under Drift

-Nicholas Davis, PhD



Summary

Enactive approaches to cognition characterize sense-making as the ongoing maintenance of viable engagement between an agent and its environment, but they remain underspecified with respect to how such engagement is sustained over extended timescales in non-stationary conditions. This paper introduces Enactive Regulation Theory (ERT) as a theoretical extension of enaction that places regulation under drift at the center of adaptive sense-making. ERT reframes drift not as noise, error, or task change, but as a temporally extended signal of declining organizational coherence. Adaptation, on this view, is not primarily a matter of learning or optimization, but of regulating the organization of engagement itself. The paper formalizes this account through three necessity theorems establishing that sustained sense-making under drift requires detecting temporally extended misalignment, regulating coherence across multiple timescales, and organizing adaptation at the level of regulation rather than parameter adjustment within fixed frameworks. By articulating regulation as a core enactive process, ERT thickens enactive temporality and clarifies how cognitive systems—biological and artificial—remain viable under continual change.

1. Introduction: Why Enaction Needs a Theory of Regulation

Enactive approaches to cognition have fundamentally reshaped how adaptive systems are understood. By rejecting representationalism and emphasizing organism–environment coupling, enaction reframes cognition as an ongoing activity of sense-making rather than a process of internal model construction (Varela, Thompson, & Rosch, 1991; Di Paolo, Rohde, & De Jaegher, 2010). Cognition, on this view, is not computation over symbols but the maintenance of viable engagement with the world through embodied action (Clark, 1997). Meaning is enacted through activity, and normativity arises from the conditions under which such activity remains viable (Di Paolo, 2005). This viability is not static but structurally precarious: enactive organization persists only through ongoing management of internal tensions and fragilities (Beer & Di Paolo, 2023).

This shift has had wide-reaching consequences across cognitive science, philosophy of mind, and human–machine interaction. Enaction has provided a compelling alternative to optimization- and representation-centered models, foregrounding situatedness, embodiment, and the relational nature of cognition (Thompson, 2007; Hutto & Myin, 2017). Yet despite these strengths, enaction has remained comparatively underspecified at the level of regulatory dynamics.

Enactive theories acknowledge breakdown, adaptation, and reorganization as central to cognition. Moments of disrupted coupling—when action no longer “grips” the world—are treated as pivotal sites of sense-making (Varela et al., 1991; Di Paolo & De Jaegher, 2012). However, existing accounts provide limited guidance on how systems detect the need for reorganization, how such reorganization unfolds across time, and how coherence is preserved when change is gradual, overlapping, or recurrent rather than abrupt. As a result, enaction excels at describing what cognition is, but is less precise about how it sustains itself under conditions of continual transformation.

This limitation becomes increasingly salient in domains characterized by non-stationarity. Biological organisms develop across lifespans marked by growth, learning, aging, and environmental change (Kelso, 1995). Artificial systems are increasingly deployed in open-ended environments where assumptions encoded at design time steadily lose validity (Quiñonero-Candela et al., 2009; Žliobaitė et al., 2015). Human–AI interactions unfold over prolonged engagement, during which goals, styles, and forms of coordination evolve (Suchman, 1987; Licklider, 1960). In such contexts, change is not episodic or exceptional; it is structural and cumulative. Systems do not merely encounter perturbations—they undergo drift.

Drift poses a distinct challenge for enactive accounts. Unlike discrete breakdowns, drift may erode coherence gradually, allowing systems to function locally while becoming globally misaligned. By the time overt breakdown occurs, reorganization may be disruptive or catastrophic. Without an explicit account of regulation under drift, enaction risks leaving unexplained how sense-making remains viable across extended timescales.

This paper introduces Enactive Regulation Theory (ERT) as a theoretical framework that addresses this gap. ERT holds that cognition is sustained through drift-sensitive regulation: the ongoing modulation of internal organization in response to emerging loss of coherence across time. Rather than treating adaptation as learning, error correction, or optimization, ERT reframes it as regulation—an activity that preserves viability by reorganizing structure when existing modes of engagement lose their fit with the environment (Ashby, 1956; Conant & Ashby, 1970).

ERT does not replace enaction. It extends it. By formalizing regulation as a primary enactive process, ERT articulates mechanisms that have remained implicit in enactive theory while preserving its core commitments: non-representational sense-making, organism–environment coupling, and normativity grounded in viability (Di Paolo, 2005; Thompson, 2007). In doing so, ERT thickens enactive temporality, shifting the focus from situated interaction alone to historically regulated sense-making. ERT is not a model, not a mechanism, and not a learning rule. It is an organizational theory specifying what any adaptive system must regulate in order to remain viable under drift.

Research Questions

This work is guided by the following theoretical questions:

Rather than offering empirical hypotheses, this paper addresses these questions through conceptual analysis and theoretical synthesis, drawing on enactive cognition, dynamical systems theory, and cybernetic notions of regulation.

Contributions

This paper makes four primary contributions:

Together, these contributions provide a principled framework for understanding how cognitive systems—biological and artificial—remain coherently engaged with their environments when change is not an exception, but the rule.


2. Related Work: Drift, Adaptation, and Regulation

Enactive Regulation Theory (ERT) draws on and intersects with several established research traditions concerned with adaptation, non-stationarity, and regulation. This section situates ERT relative to adjacent work in machine learning, cybernetics, and enactive cognitive science. The goal is not comprehensive comparison, but to clarify how ERT addresses a shared problem—adaptation under change—from a distinct organizational perspective.

2.1. Drift in Machine Learning and Adaptive Systems

In machine learning and adaptive systems, drift is typically studied under the headings of concept drift, dataset shift, and non-stationary learning. Concept drift frameworks analyze changes in the statistical relationship between inputs and outputs over time, distinguishing between gradual, abrupt, and recurring shifts (Gama et al., 2014; Widmer & Kubat, 1996). Recent surveys emphasize the structural inevitability of drift in streaming and open environments (Lu et al., 2020; Hinder et al., 2024). Dataset shift similarly characterizes deviations between training and deployment distributions that undermine predictive validity (Quiñonero-Candela et al., 2009). Continual and online learning approaches attempt to maintain performance under such conditions through incremental updates, memory replay, regularization, or adaptive windowing (Bifet & Gavaldà, 2007; Parisi et al., 2019; Shalev-Shwartz, 2012).

Across these approaches, drift is generally treated as a problem to be corrected. Adaptation is framed as restoring predictive accuracy, minimizing regret, or maintaining task performance despite changing data distributions. Even when multiple models or ensembles are employed, the emphasis remains on selecting or updating models to best fit the current environment.

What is largely absent from this literature is a notion of organizational coherence. Drift is detected primarily through performance degradation rather than through internal instability, fragmentation, or loss of coordination. As a result, adaptive responses focus on parameter updates within a fixed organizational structure rather than on reorganization of that structure itself. This limits their ability to support long-duration adaptation in environments where change is overlapping, recurrent, or structurally transformative rather than task-bounded (Žliobaitė et al., 2015). Drift detection techniques treat distributional change as a signal requiring monitoring rather than immediate correction, reinforcing the view that drift is informative rather than purely defective (Hinder et al., 2024).

Limitation addressed by ERT: Existing adaptive learning frameworks treat drift as an external disturbance to be corrected, lacking mechanisms for regulating internal organization in response to sustained loss of coherence.

2.2. Regulation in Cybernetics and Dynamical Systems

Cybernetic and dynamical systems traditions have long emphasized regulation as a central principle of adaptive behavior. Early cybernetics, particularly the work of Ashby and Conant & Ashby, established that effective regulation requires sensitivity to the system being regulated and the conditions under which regulation remains possible (Ashby, 1956; Conant & Ashby, 1970). Control theory formalized regulation in terms of feedback loops, setpoints, and error correction, while dynamical systems theory modeled adaptation through attractors, stability, and phase transitions (Haken, 1983; Kelso, 1995).

These frameworks provide powerful tools for understanding stability and change, but they typically define regulation relative to fixed goals or control variables. Even in accounts of self-organization or metastability, regulation is described in terms of maintaining trajectories, equilibria, or predefined objectives.

ERT departs from this tradition in two key ways. A biologically grounded account of regulation clarifies this move. Regulation, on this view, is not external control but internal constraint modulation that preserves systemic organization from within (Bich et al., 2016). ERT rejects fixed setpoints as the basis of regulation, replacing them with viability and coherence as dynamically maintained conditions. Critically, it treats regulation as operating on organizational structure, not merely on outputs or state variables. Coherence, in ERT, is not a control variable to be optimized but a relational property indicating whether internal dynamics remain mutually supportive over time.

Limitation addressed by ERT: Classical regulation frameworks emphasize control relative to fixed objectives, leaving under-theorized how systems reorganize their own structure when the conditions of regulation themselves change.

2.3. Enaction and Temporality

Enactive approaches to cognition characterize sense-making as the active maintenance of organism–environment coupling grounded in viability rather than representation (Varela et al., 1991; Thompson, 2007). Autopoietic theory emphasizes organizational closure and self-production, while participatory sense-making extends enaction to social interaction, highlighting coordination, breakdown, and repair (Di Paolo, 2005; Di Paolo & De Jaegher, 2012). Across these accounts, breakdown plays a crucial role: moments of disrupted coupling motivate reorganization and renewed sense-making.

ERT is continuous with these commitments but addresses a temporal gap. In many enactive accounts, breakdown is treated primarily as an episodic event—a local disruption in organism–environment coupling that prompts reorganization and renewed sense-making (Varela, Thompson, & Rosch, 1991; Di Paolo, 2005; Di Paolo & De Jaegher, 2012). Such breakdowns play a central role in motivating adaptive restructuring, particularly in participatory and interactional contexts. However, less attention has been given to gradual, cumulative misalignment that unfolds across extended interaction histories without discrete rupture (Thompson, 2007; Di Paolo et al., 2017). Drift introduces precisely this form of temporal pressure: coherence may erode incrementally while local engagement remains functional, delaying reorganization until instability becomes structurally significant.

By treating drift as a temporally extended regulatory signal, ERT thickens enactive temporality beyond situated interaction. Temporal modulation of intention and meaning is not merely forward-moving but retro-modulatory, shaping present sense-making through past and anticipated trajectories (Di Paolo, 2015).  Sense-making becomes sensitive not only to immediate perturbations but also to the accumulation of tension, instability, and misalignment across time. Regulation is rendered explicit rather than implicit, clarifying how systems remain viable when change is slow, overlapping, or recurrent.

Limitation addressed by ERT: While enactive theories recognize breakdown and reorganization, they lack a principled account of how gradual loss of coherence is detected and regulated across extended timescales.

Across machine learning, cybernetics, and enactive cognition, drift and regulation are acknowledged but framed differently: as error to correct, deviation to control, or breakdown to repair. Enactive Regulation Theory integrates these perspectives by reframing drift as a signal of organizational incoherence and regulation as the process that sustains sense-making under continual change. This positioning clarifies both ERT’s lineage and its distinct theoretical contribution.

3. Enaction and the Limits of Learning-Centric Adaptation

Enactive cognition emerged in explicit contrast to learning-centric and representational theories of mind. Rather than modeling cognition as inference over internal representations, enaction characterizes cognition as the enactment of a meaningful world through embodied activity (Varela et al., 1991; Thompson, 2007). Meaning arises not from correspondence between internal models and external states, but from the norms of viability governing organism–environment coupling (Di Paolo, 2005).

Within this framework, adaptation is not primarily a matter of updating representations or minimizing error. It is a matter of maintaining viable engagement. Breakdowns in coupling—moments when action no longer yields expected affordances—signal the need for reorganization rather than correction (Di Paolo & De Jaegher, 2012; Thompson, 2007). Such breakdowns are not failures of prediction but disruptions in the conditions that sustain sense-making.

However, many contemporary adaptive systems, including those inspired by enaction, continue to rely implicitly on learning-centric mechanisms. Learning modifies parameters, representations, or policies within an assumed organizational structure. Even when deployed online or continuously, learning typically presupposes that the structure within which updates occur remains appropriate (Clark, 2013; Sutton & Barto, 2018). This assumption reflects a residual commitment to stability at the organizational level, even when content is allowed to change. From an ecological-enactive perspective, cognition is not best understood as inference toward optimal predictions but as skilled regulation of organism–environment fit (Bruineberg et al., 2018).

The learning-centric assumption becomes fragile under drift. When the environment’s structure changes in ways that invalidate prior organizational commitments, learning alone is insufficient. Parameter updates may compensate locally while global coherence erodes. Error may remain bounded even as the system’s internal organization fragments across time—a phenomenon observed in continual learning systems where performance metrics fail to capture deeper instability (French, 1999; Parisi et al., 2019). In such cases, learning delays rather than prevents breakdown.

From an enactive perspective, this constitutes a category error. Learning addresses content within organization; it does not address the organization itself. When the problem is loss of coherence, the solution is not improved learning but regulation—deciding when to preserve, modify, or abandon existing modes of engagement (Ashby, 1956; Di Paolo, 2005).

Enactive Regulation Theory (ERT) clarifies this distinction by explicitly separating learning from regulation. Learning is treated as a subordinate process that operates within regulated regimes, while regulation is positioned as the primary adaptive activity that determines when learning is appropriate, suppressed, or redirected. This reframing aligns closely with enactive commitments to viability, breakdown, and reorganization, while addressing a critical blind spot in learning-centric adaptive accounts: the absence of mechanisms for regulating organizational coherence under drift.

4. The Problem of Drift in Enactive Systems

Drift is widely discussed in machine learning, signal processing, and adaptive control, yet it is rarely treated as a foundational theoretical concern. In most frameworks, drift is conceptualized as noise, distribution shift, or task change—an external complication that degrades performance and must be corrected (Widmer & Kubat, 1996; Gama et al., 2014; Žliobaitė et al., 2016).

ERT adopts a different stance. Drift is not an anomaly imposed on otherwise stable systems. It is an intrinsic feature of systems that operate over time in structurally coupled environments. Drift arises whenever the assumptions embedded in a system’s organization gradually lose alignment with the environment’s evolving structure (Varela et al., 1991; Thompson, 2007; Di Paolo et al., 2017).

In enactive terms, drift corresponds to a loss of coherence in coupling. Actions no longer stabilize expected patterns. Internal dynamics become increasingly fragmented. Coordination across timescales weakens (Kelso, 1995; Thelen & Smith, 1994). Importantly, these changes need not manifest as immediate failure. Systems may continue to function locally while accumulating global incoherence—a phenomenon observed in both biological adaptation and artificial learning systems (Smith & Thelen, 2003; Parisi et al., 2019). Habits, for example, may remain locally stable while becoming globally maladaptive under shifting conditions (Ramírez-Vizcaya & Froese, 2019).

This creates a problem for adaptation strategies that rely on error or performance as primary signals. Error is local and outcome-based; it does not capture the gradual erosion of organizational fit. A system may minimize error while drifting structurally, compensating moment-to-moment without addressing deeper misalignment. By the time failure becomes visible, reorganization is often disruptive or catastrophic (Ashby, 1956; Di Paolo, 2005; Kirkpatrick et al., 2017).

ERT therefore reframes drift as a regulatory signal, not a defect. Drift signals that the system’s current organization is no longer sustaining coherent sense-making across time. The appropriate response is not correction, but reorganization—adjusting the structure through which engagement is enacted (Di Paolo et al., 2017; Thompson & Stapleton, 2009).

This reframing has two major consequences. First, it shifts the adaptive focus from performance optimization to coherence maintenance, aligning adaptation with viability rather than task success (Di Paolo, 2005; Moreno & Mossio, 2015). Second, it foregrounds temporality: adaptation becomes a trajectory-sensitive process rather than a sequence of isolated updates. Cognition, on this view, is not merely situated—it is historically regulated (Varela, 1979; Smith & Thelen, 2003).

By identifying drift as a central condition of adaptive systems and regulation as the mechanism that responds to it, ERT provides the conceptual foundation needed to extend enaction into domains of long-duration, non-stationary operation.

5. Core Commitments of Enactive Regulation Theory (ERT)

Enactive Regulation Theory (ERT) advances a set of commitments that together define a distinct account of adaptation, sense-making, and long-term cognitive organization. These commitments do not introduce new metaphysical claims beyond enaction, nor do they depend on specific computational mechanisms. Instead, they articulate the organizational principles required for systems to remain coherently engaged with their environments under conditions of drift. 

5.1 Regulation Is Primary, Learning Is Subordinate

The first commitment of ERT is that regulation, not learning, is the primary adaptive process. Learning modifies parameters, representations, or associations within an existing organizational structure. Regulation determines whether that structure remains viable at all. Agency, on enactive accounts, is defined precisely by adaptive regulation across time, including spatio-temal asymmetries that sustain identity (Barandiaran et al., 2009). When environments are stable or slowly varying, learning may suffice. Under drift, however, learning alone cannot determine when its own assumptions have become invalid.

ERT therefore treats learning as conditional. Learning occurs within regulated regimes and is modulated by regulatory state. In some contexts, learning may be suppressed to preserve coherence; in others, it may be amplified or redirected. Regulation decides when learning is appropriate, not merely how it proceeds. This distinction clarifies a common ambiguity in adaptive systems. Systems often appear to “learn their way out” of breakdowns, but closer inspection reveals the presence of implicit regulatory mechanisms—such as resetting internal models, switching between regimes, or reallocating memory resources—rather than pure parameter updating within a fixed structure (Ashby, 1956; Conant & Ashby, 1970; French, 1999; Kirkpatrick et al., 2017; Parisi et al., 2019). In continual learning systems, for example, strategies that prevent catastrophic interference frequently rely on architectural separation, memory rehearsal, or consolidation constraints, which function less as learning improvements and more as forms of structural regulation (French, 1999; Parisi et al., 2019). Similarly, adaptive control frameworks incorporate mode switching and reset mechanisms when operating conditions exceed the stability bounds of a current regime (Åström & Murray, 2008). These cases suggest that what appears as incremental learning is often scaffolded by regulatory operations that determine when and how learning is permitted to proceed. In biological systems, homeostatic plasticity stabilizes neural function when Hebbian learning destabilizes it, illustrating how regulation governs learning processes rather than replacing them (Turrigiano, 2012). ERT makes these mechanisms explicit and elevates them to first-class theoretical status.

5.2 Drift Is a Regulatory Signal, Not a Performance Defect

ERT’s second commitment is that drift is treated as an endogenous regulatory signal rather than an error condition. Performance-based signals such as prediction error or reward loss are insufficient for guiding long-term adaptation. They are local, outcome-dependent, and insensitive to gradual erosion of organizational coherence. Drift, by contrast, reflects cumulative loss of fit between a system’s internal organization and the structure of its environment.

In ERT, drift is detected through coherence measures: indicators of instability, fragmentation, or inconsistency across time. These measures are not limited to accuracy or reward; they include temporal persistence, cross-scale alignment, and stability of internal dynamics. By treating drift as informative rather than pathological, ERT enables anticipatory regulation. Systems reorganize before catastrophic failure, adjusting structure in proportion to the magnitude and persistence of coherence loss. Adaptation thus becomes graded and continuous rather than reactive and disruptive.

5.3 Cognition Is Sustained Through Regime Organization

A third commitment of ERT is that cognitive coherence is maintained through regimes rather than monolithic organization. A regime is a temporally extended mode of organization that stabilizes sense-making under particular conditions. Regimes are not tasks, models, or policies. They are patterns of coordination that remain viable as long as coherence is maintained. ERT assumes that multiple regimes may coexist. Environmental structure is rarely uniform, and forcing all engagement through a single organizational form leads to internal conflict and drift accumulation. Regime plurality allows systems to preserve distinct modes of sense-making and transition between them when coherence degrades.

This commitment directly addresses the problem of catastrophic interference. Rather than protecting representations or freezing parameters, ERT prevents interference by maintaining organizational separation. Forgetting, when it occurs, is a regulated shift in dominance rather than an uncontrolled loss of structure.

5.4 Regulation Operates Across Multiple Timescales

ERT’s fourth commitment is that regulation is inherently multi-temporal. Drift unfolds over time. Some coherence losses emerge rapidly, while others accumulate slowly across extended interaction histories. Regulation must therefore operate across multiple timescales, from moment-to-moment stabilization to long-term reorganization. ERT rejects purely reactive adaptation. Instead, it emphasizes trajectory-sensitive regulation: decisions are informed by trends, persistence, and recurrence, not just instantaneous signals. This enables systems to distinguish between transient perturbations and genuine structural change.

Multi-timescale regulation also supports continuity. By retaining memory of prior regimes and organizational solutions, systems can re-engage past structures when conditions recur, rather than relearning from scratch. Adaptation thus becomes historical rather than episodic.

5.5 Coherence, Not Optimality, Is the Primary Norm

The final commitment of ERT is that coherence, rather than optimality, is the fundamental norm governing adaptation. Many adaptive frameworks assume that cognition aims to optimize some objective—prediction accuracy, reward, or efficiency. ERT rejects this assumption as insufficient for systems operating under drift. Optimality presupposes stable criteria; coherence does not. Coherence refers to the system’s ability to remain organized such that its internal dynamics support ongoing engagement with the environment. This normative orientation does not reintroduce external goals; it arises from self-determining organization itself, which grounds purposiveness in constraint closure rather than in imposed objectives (Mossio & Bich, 2017). A system may be locally suboptimal and yet globally coherent. Conversely, a system may achieve high short-term performance while drifting toward breakdown.

By prioritizing coherence, ERT aligns with enactive accounts of viability while providing a concrete regulatory criterion. Adaptation is successful not when performance is maximized, but when sense-making remains organized across change. Taken together, these commitments define Enactive Regulation Theory as a framework for understanding adaptation under drift:

ERT does not prescribe specific algorithms or architectures. Instead, it specifies the organizational constraints any system must satisfy to remain coherently adaptive in non-stationary environments. These commitments set the stage for examining how ERT can be instantiated computationally and empirically, and how it reframes longstanding debates in adaptive cognition and artificial intelligence.

5.6. The Enactive Regulation Theorems

The following theorems formalize the central claims of Enactive Regulation Theory by making explicit the organizational conditions required for sustaining sense-making under drift. Rather than offering empirical predictions or algorithmic prescriptions, these theorems articulate necessity claims: constraints that any system—biological or artificial—must satisfy in order to remain coherently engaged with its environment across extended, non-stationary interaction. Each theorem isolates a distinct aspect of regulation under drift, together providing a structured account of why adaptation cannot be reduced to learning, optimization, or parameter adjustment within fixed organizational frameworks.

5.6.1 The Drift Signal Theorem

In systems operating under gradual, non-stationary change, performance-based error signals are insufficient for detecting loss of coherence; drift must instead be detected as a temporally extended signal of organizational misalignment. Systems that rely exclusively on outcome-based error will detect breakdown only after organizational coherence has already degraded.

5.6.2 The Temporal Coherence Theorem

Sense-making is sustained only through regulation that operates across multiple timescales; moment-to-moment adaptation alone is insufficient to preserve coherence under drift. Systems lacking multi-timescale regulatory organization may remain locally responsive while becoming historically incoherent.

5.6.3 Enactive Regulation Theorem

Any system capable of sustaining sense-making over extended interaction in a non-stationary environment must regulate the coherence of its own organizational dynamics under drift. Adaptation based solely on learning, optimization, or parameter updating within a fixed representational or objective framework is insufficient to ensure continued sense-making over time.

Together, these three theorems articulate the core claims of Enactive Regulation Theory by specifying the conditions under which sense-making can be sustained under drift. The Drift Signal Theorem establishes that loss of coherence cannot be reliably detected through performance-based error alone, requiring drift to be treated as a temporally extended signal of organizational misalignment. The Temporal Coherence Theorem follows by showing that responding to such signals demands regulation across multiple timescales, since moment-to-moment adaptation is insufficient to preserve coherence across extended interaction histories. The Enactive Regulation Theorem integrates these insights into a necessity claim: any system that remains capable of sense-making in non-stationary environments must regulate the coherence of its own organizational dynamics rather than relying solely on learning or optimization within a fixed framework. Taken together, the theorems reframe adaptation as an organizational achievement sustained through regulation under drift, rather than as incremental improvement toward stable objectives.

Here, drift refers to temporally extended misalignment between a system’s current organization and the evolving structure of its coupling with the environment. Regulation denotes processes that modulate how adaptation itself is organized—altering modes of engagement, sensitivity, and participation—rather than merely adjusting parameters within a fixed learning objective. Sense-making is understood in the enactive sense, as the ongoing enactment of meaningful relations that support system viability.

6. Discussion: Regulation, Drift, and the Future of Enactive Sense-Making

This paper has argued that enactive approaches to cognition require a more explicit theory of regulation in order to fully account for adaptive sense-making under conditions of continual change. While enaction has successfully reframed cognition as embodied, relational, and normatively grounded in viability, it has remained comparatively underspecified with respect to how coherence is sustained across extended temporal drift. Enactive Regulation Theory (ERT) addresses this gap by articulating regulation as a primary enactive process—one that operates on organizational structure rather than on representational content or performance variables.

The contribution of ERT is not to revise enaction’s core commitments, but to thicken them temporally. By foregrounding drift as a first-class phenomenon and regulation as the mechanism that responds to it, ERT shifts the focus of enactive cognition from situated interaction alone to historically extended sense-making.

6.1 Regulation as an Enactive Primitive

A central implication of ERT is that regulation must be treated as an enactive primitive rather than as a secondary or derivative process. In many adaptive frameworks, regulation is subsumed under learning, control, or optimization. In contrast, ERT argues that regulation is the activity that determines when learning is appropriate, which modes of engagement remain viable, and how organizational structure should be reorganized when coherence degrades.

This reframing aligns closely with biological intuition. Living systems do not simply learn continuously; they regulate learning, plasticity, and persistence in response to their own stability and fragility. By making this regulatory activity explicit, ERT provides a principled way to describe how sense-making systems maintain continuity without freezing, and adapt without dissolving their own organization.

6.2 Drift as a Structuring Condition, Not a Pathology

Another key contribution of ERT is the reconceptualization of drift. In much of machine learning and adaptive systems research, drift is treated as a nuisance or failure mode—an external disturbance that must be corrected in order to restore performance. ERT instead treats drift as a structuring condition of real-world cognition.

From an enactive perspective, environments are not stable backdrops against which cognition unfolds. They change continuously, often in ways that are gradual, overlapping, and recurrent. Drift captures this reality more faithfully than episodic notions of task shift or breakdown. By treating drift as an informative signal of declining coherence rather than as noise or error, ERT reframes adaptation as an ongoing negotiation with historical change.

This move has important consequences. It explains why systems can remain locally functional while becoming globally misaligned, why breakdown is often delayed rather than sudden, and why catastrophic reorganization is frequently the result of unregulated drift rather than abrupt perturbation.

6.3 Implications for Artificial Adaptive Systems

Although ERT is presented as a theoretical framework, its implications for artificial systems are substantial. Many contemporary AI systems operate effectively in short deployments but degrade over time when exposed to open-ended environments. Standard responses—retraining, fine-tuning, or parameter regularization—address symptoms rather than causes. They adapt content within an assumed organizational structure without regulating that structure itself.

ERT suggests a different design target. Rather than optimizing performance under assumed stability, adaptive systems should be designed to remain coherently organized under change. This implies architectures that support multiple regimes, monitor coherence across timescales, and reorganize structure conservatively in response to drift. Enactive Drift Regulation (EDR), introduced as the operationalization of ERT, exemplifies this approach, but the theoretical implications extend beyond any single architecture.

Importantly, this reframing does not require artificial systems to possess consciousness or phenomenology. ERT distinguishes clearly between the functional role of lived regulatory feedback and its biological realization. Artificial systems can implement the regulatory function without reproducing experience, preserving conceptual clarity while enabling practical application.

6.4 Enaction Beyond Situatedness: Thickening Temporality

ERT also contributes to enactive theory itself by extending its treatment of temporality. Traditional enactive accounts emphasize situatedness: cognition unfolds in the here-and-now of organism–environment coupling. ERT retains this emphasis but argues that it is insufficient for understanding long-duration adaptation.

Drift introduces a form of temporal pressure that is not reducible to immediate breakdown. Coherence may erode slowly, across interaction histories, without producing discrete moments of failure. By treating regulation as sensitive to accumulated instability rather than only to immediate disruption, ERT provides a framework for understanding how sense-making is historically shaped and preserved.

This move bridges enaction with dynamical systems approaches that emphasize metastability, phase transitions, and multi-timescale organization, while retaining enaction’s normative grounding in viability.

6.5 Scope and Limitations

ERT is intentionally positioned as a theoretical framework rather than a comprehensive empirical theory. It does not specify particular neural, physiological, or computational mechanisms, nor does it offer testable predictions in isolation. Instead, it provides conceptual constraints and organizational principles that can guide both empirical investigation and system design.

Future work is needed to explore how ERT can be operationalized across domains, how coherence and drift can be measured in practice, and how regulatory processes interact with learning, perception, and action in concrete systems. Empirical studies—biological and artificial—will be essential for refining and challenging the framework.

6.6 Sustaining Coherent Interaction

Taken together, Enactive Regulation Theory reframes a central question of cognitive science and AI. The defining challenge for adaptive systems may not be learning faster, predicting more accurately, or optimizing better, but remaining coherently engaged with a changing world. By placing regulation under drift at the center of sense-making, ERT offers a principled account of how such engagement can be sustained.

Rather than treating drift as a defect to be eliminated, ERT treats it as the very condition that makes regulation—and cognition itself—necessary. In doing so, it extends enaction into a temporally grounded theory of adaptive organization, opening new paths for understanding biological cognition, designing artificial systems, and rethinking what it means to adapt over time.

7. Limitations and Future Work

While Enactive Regulation Theory (ERT) provides a principled account of regulation under drift, several limitations and open questions remain.

First, ERT is intentionally organizational and theoretical rather than mechanistic. It specifies what must be regulated for sense-making to remain viable under drift, but not the particular neural, physiological, or computational mechanisms by which such regulation is realized. This abstraction is a strength for theoretical generality, but it also means that ERT does not, by itself, generate direct empirical predictions. Future work should explore how ERT’s commitments can be operationalized in specific domains, including neural dynamics, embodied agents, and artificial adaptive systems.

Second, while ERT emphasizes coherence as a central regulatory quantity, coherence itself remains an open construct. Different systems may instantiate coherence in different ways—through stability of coordination dynamics, consistency across timescales, or maintenance of viable interaction patterns. Further research is needed to clarify how coherence can be measured, tracked, and compared across biological and artificial systems without collapsing it into performance metrics or error signals.

Third, ERT explicitly distinguishes itself from theories of subjective experience, yet it draws conceptual inspiration from lived regulation in biological cognition. This raises important questions about the relationship between functional regulation and phenomenology. While ERT does not claim that artificial systems implementing regulatory mechanisms are conscious, future philosophical and empirical work may investigate how regulatory dynamics relate to experiential structure in living systems, and whether degrees or forms of regulation correlate with phenomenological richness.

Fourth, the present work focuses primarily on individual sense-making systems, even when discussing participatory and interactive contexts. Extending ERT to fully collective or multi-agent systems—where regulation itself may be distributed across agents, artifacts, and environments—remains an important direction for future research. Such extensions may help explain long-term coordination, breakdown, and reorganization in social, cultural, and human–AI systems.

Finally, while Enactive Drift Regulation (EDR) is introduced as one operational instantiation of ERT, it is not intended as the sole or definitive realization. Other architectures may instantiate ERT’s principles differently, and comparative work will be necessary to understand which regulatory designs are most effective under different forms of drift.

In sum, ERT should be understood not as a closed theory, but as a conceptual foundation for studying regulation, drift, and coherence in adaptive systems. Its primary contribution lies in reframing adaptation as an organizational problem rather than an optimization problem, thereby opening new avenues for empirical investigation and system design.

8. Implications for Consciousness, AI, and Co-Creation

Enactive Regulation Theory (ERT) reframes adaptation around regulation under drift rather than optimization under stability. This reframing has implications that extend beyond any single architecture or domain. Rather than offering predictions or prescriptions, this section outlines opening conditions for how consciousness, artificial systems, and co-creative interaction can be rethought through the lens of enactive regulation.

8.1 Implications for Consciousness Studies

ERT does not attempt to define consciousness in phenomenological or metaphysical terms. Instead, it identifies a functional role that consciousness appears to play in biological systems: lived feedback that enables drift-sensitive regulation of sense-making.

From this perspective, consciousness is not an epiphenomenal byproduct of cognition, nor merely a representational workspace. It is the means by which organisms feel their own loss of coherence and modulate behavior accordingly. Tension, effort, confusion, flow, and insight become intelligible as regulatory signals rather than subjective embellishments.

ERT thus offers a bridge between enactive accounts of experience and regulatory dynamics without reducing experience to computation. It clarifies what consciousness does—it participates in regulation—without claiming that this role can or should be replicated in artificial systems.

8.2 Implications for Artificial Intelligence

For artificial systems, ERT suggests a shift in design priorities. Rather than optimizing performance metrics under assumed stability, adaptive systems should be designed to remain coherently organized under continual change.

This reframing has several consequences:

Importantly, ERT does not imply that artificial systems must be conscious. Instead, it shows that systems can enact the regulatory function associated with consciousness—monitoring coherence and reorganizing structure—without phenomenology. This allows AI research to draw inspiration from enactive principles while remaining disciplined about its claims.

8.3 Implications for Co-Creation and Human–AI Interaction

ERT also sheds light on co-creative interaction. Creative collaboration—whether between humans, or between humans and machines—unfolds under drift. Intentions shift, styles evolve, and coordination must be continuously renegotiated.

From an ERT perspective, strong co-creation depends not on responsiveness alone, but on shared regulation of drift. Flow, creative impasse, and breakthrough can be understood as regime dynamics: moments when coherence is sustained, lost, or reorganized across participants.

For human–AI systems, this suggests that effective co-creation requires agents that can:

ERT thus reframes co-creative intelligence as a regulatory achievement rather than a generative one. Creativity emerges not from producing novelty per se, but from sustaining viable sense-making together as conditions change.

8.4. Implications for Conscious Experience

The account of regulation developed here has direct implications for how consciousness is understood. On the present view, consciousness is not identified with the presence of privileged internal representations, global broadcasts, or access to task-relevant information. Nor is it treated as an epiphenomenal accompaniment to learning or control. Instead, consciousness is situated at the level of ongoing sense-making as it is actively regulated under conditions of drift.

From an enactive perspective, cognition is not exhausted by information processing or problem solving, but consists in the organism’s or system’s capacity to maintain coherent engagement with its environment over time. Regulation, in this sense, is not merely a background mechanism that supports cognition; it is constitutive of it. The ability to detect misalignment, modulate participation, and reorganize patterns of activity in response to changing conditions is what allows sense-making to remain viable. Conscious experience, on this account, arises within this regulatory process rather than alongside it.

This reframing shifts the explanatory burden in consciousness studies away from identifying privileged internal states and toward understanding how viable forms of organization are sustained over time. Conscious experience, on this view, is not a static property but a temporally extended achievement—one that depends on the system’s capacity to regulate its own participation in a changing world. Experiences such as clarity, confusion, effort, or fluency are not treated as intrinsic features of internal states, but as expressions of how regulatory dynamics succeed or falter in maintaining coherent engagement.

Importantly, this account does not attempt to reduce consciousness to regulation, nor does it claim that regulatory organization alone is sufficient for conscious experience. Rather, it proposes that consciousness is inseparable from the lived dynamics of regulation in systems capable of sustaining autonomous sense-making. To be conscious, on this view, is to be continuously involved in the negotiation of one’s own conditions of intelligibility—conditions that are never fully fixed and must be actively maintained under drift.

By locating consciousness in the regulation of sense-making rather than in isolated mechanisms or representations, Enactive Regulation Theory aligns with biological and phenomenological accounts that treat experience as inherently temporal, relational, and precarious. Consciousness is thus understood not as a discrete component of cognition, but as an emergent dimension of regulated engagement—one that comes into view precisely when coherence must be actively sustained in the face of change.

8.4 Remaining Viable Through Reorganization

Across these domains, a common insight emerges. The central challenge for adaptive systems—biological or artificial—is not prediction, optimization, or control in isolation. It is the capacity to remain meaningfully organized over time in the face of inevitable drift.

Enactive Regulation Theory provides a conceptual framework for understanding this challenge, while Enactive Drift Regulation demonstrates how it can be addressed operationally. Together, they invite a reorientation of cognitive science, AI, and HCI toward regulation as the core of adaptive intelligence.

Rather than asking how systems can perform better under fixed assumptions, ERT encourages a different question:

How can systems remain viable as the very conditions of viability change?

That question, we suggest, will increasingly define the future of adaptive intelligence research.

9. Conclusion: Enaction, Regulated

This paper has argued that enactive accounts of cognition require an explicit theory of regulation to explain how sense-making is sustained under conditions of continual, non-stationary change. While enaction has long emphasized organism–environment coupling, viability-based normativity, and breakdown-driven reorganization, it has left underspecified how the need for reorganization is detected, graded, and coordinated across extended timescales. Enactive Regulation Theory (ERT) addresses this gap by reframing adaptation as the regulation of coherence under drift. By treating drift as a temporally extended signal of declining organizational coherence rather than as noise, error, or task shift, ERT clarifies why learning and optimization alone are insufficient for long-duration adaptability. Regulation, on this view, operates at the level of organizational dynamics, modulating stability and plasticity in response to emerging misalignment between agent and environment, and thereby placing coherence—not accuracy or performance—at the center of adaptive sense-making. ERT also reinterprets lived experience as regulatory feedback that enables anticipatory and graded reorganization, preserving enaction’s non-representational commitments while clarifying how viability is maintained across historical time. Importantly, ERT does not replace enaction or adjudicate its philosophical disputes; rather, it offers a dynamical refinement that renders explicit the regulatory processes already implicit in enactive theory. In doing so, it provides a conceptual bridge across biological cognition, artificial adaptive systems, and co-creative interaction without collapsing their differences. As environments, technologies, and forms of participation become increasingly fluid, the central challenge for adaptive systems is less prediction or optimization than remaining coherently organized over time. By naming regulation under drift as the core mechanism of sustained sense-making, ERT clarifies what adaptation requires when change is not the exception, but the rule.


Acknowledgments

This work was developed through an extended process of human–AI collaboration. The author worked iteratively with a large language model as a dialogical partner to refine conceptual distinctions, clarify theoretical structure, pressure-test arguments, and strengthen the articulation of Enactive Regulation Theory. The AI system contributed to drafting, reorganization, comparative analysis, and the sharpening of explanatory clarity. All theoretical claims, interpretations, and final editorial decisions remain the responsibility of the author.

References

Ashby, W. R. (1956). An introduction to cybernetics. Chapman & Hall.

Åström, K. J., & Murray, R. M. (2008). Feedback systems: An introduction for scientists and engineers. Princeton University Press.

Barandiaran, X. E., & Moreno, A. (2006). On what makes certain dynamical systems cognitive: A minimally cognitive organization program. Adaptive Behavior, 14(2), 171–185.

Barandiaran, X. E., Di Paolo, E. A., & Rohde, M. (2009). Defining agency: Individuality, normativity, asymmetry, and spatio-temporality in action. Adaptive Behavior, 17(5), 367–386. https://doi.org/10.1177/1059712309343819

Beer, R. D., & Di Paolo, E. A. (2023). The theoretical foundations of enaction: Precariousness. BioSystems, 223, 104823. https://doi.org/10.1016/j.biosystems.2022.104823

Beer, S. (1972). Brain of the firm: The managerial cybernetics of organization. Allen Lane.

Bich, L., Mossio, M., Ruiz-Mirazo, K., & Moreno, A. (2016). Biological regulation: Controlling the system from within. Biology & Philosophy, 31(2), 237–265.

Bruineberg, J., Kiverstein, J., & Rietveld, E. (2018). The anticipating brain is not a scientist: The free-energy principle from an ecological-enactive perspective. Synthese, 195, 2417–2444. https://doi.org/10.1007/s11229-016-1239-1

Clark, A. (1997). Being there: Putting brain, body, and world together again. MIT Press.

Clark, A. (2008). Supersizing the mind: Embodiment, action, and cognitive extension. Oxford University Press.

Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences, 36(3), 181–204. https://doi.org/10.1017/S0140525X12000477

Conant, R. C., & Ashby, W. R. (1970). Every good regulator of a system must be a model of that system. International Journal of Systems Science, 1(2), 89–97. https://doi.org/10.1080/00207727008920220

De Jaegher, H., & Di Paolo, E. A. (2007). Participatory sense-making. Phenomenology and the Cognitive Sciences, 6(4), 485–507. https://doi.org/10.1007/s11097-007-9076-9

De Jaegher, H., Di Paolo, E. A., & Gallagher, S. (2010). Can social interaction constitute social cognition? Trends in Cognitive Sciences, 14(10), 441–447. https://doi.org/10.1016/j.tics.2010.06.009

Di Paolo, E. A. (2005). Autopoiesis, adaptivity, teleology, agency. Phenomenology and the Cognitive Sciences, 4(4), 429–452. https://doi.org/10.1007/s11097-005-9002-y

Di Paolo, E. A. (2015). Interactive time-travel: On the intersubjective retro-modulation of intentions. Journal of Consciousness Studies, 22(1–2), 49–74.

Di Paolo, E. A. (2021). Enactive becoming. Phenomenology and the Cognitive Sciences, 20, 783–809. https://doi.org/10.1007/s11097-019-09654-1

Di Paolo, E. A., Buhrmann, T., & Barandiaran, X. (2017). Sensorimotor life: An enactive proposal. Oxford University Press.

Di Paolo, E. A., Rohde, M., & De Jaegher, H. (2010). Horizons for the enactive mind: Values, social interaction, and play. In J. Stewart, O. Gapenne, & E. A. Di Paolo (Eds.), Enaction: Toward a new paradigm for cognitive science (pp. 33–87). MIT Press.

Di Paolo, E. A., & De Jaegher, H. (2012). The interactive brain hypothesis. Frontiers in Human Neuroscience, 6, 163. https://doi.org/10.3389/fnhum.2012.00163

Di Paolo, E. A., Thompson, E., & Beer, R. D. (2022). Laying down a forking path: Tensions between enaction and the free energy principle. Philosophy and the Mind Sciences, 3. https://doi.org/10.33735/phimisci.2022.9187

French, R. M. (1999). Catastrophic forgetting in connectionist networks. Trends in Cognitive Sciences, 3(4), 128–135. https://doi.org/10.1016/S1364-6613(99)01294-2

Froese, T., & Di Paolo, E. A. (2011). The enactive approach. Wiley Interdisciplinary Reviews: Cognitive Science, 2(1), 1–11.

Fuchs, T., & De Jaegher, H. (2009). Enactive intersubjectivity. Phenomenology and the Cognitive Sciences, 8(4), 465–486.

Gama, J., Žliobaitė, I., Bifet, A., Pechenizkiy, M., & Bouchachia, A. (2014). A survey on concept drift adaptation. ACM Computing Surveys, 46(4), 44. https://doi.org/10.1145/2523813

Haken, H. (1983). Synergetics: An introduction. Springer.

Hinder, F., et al. (2024). One or two things we know about concept drift—A survey on unsupervised drift detection. Frontiers in Artificial Intelligence.

Hutto, D. D., & Myin, E. (2017). Evolving enactivism: Basic minds meet content. MIT Press.

Juarrero, A. (1999). Dynamics in action: Intentional behavior as a complex system. MIT Press.

Kelso, J. A. S. (1995). Dynamic patterns: The self-organization of brain and behavior. MIT Press.

Khamassi, M., Velentzas, G., Tsitsiklis, J., & Tzafestas, S. (2015). On the role of non-stationarity in lifelong reinforcement learning. Adaptive Behavior, 23(3), 170–190.

Kirsh, D. (2013). Embodied cognition and the magical future of interaction design. ACM Transactions on Computer-Human Interaction, 20(1), 1–30.

Kirkpatrick, J., Pascanu, R., Rabinowitz, N., … Hadsell, R. (2017). Overcoming catastrophic forgetting in neural networks. Proceedings of the National Academy of Sciences, 114(13), 3521–3526. https://doi.org/10.1073/pnas.1611835114

Licklider, J. C. R. (1960). Man–computer symbiosis. IRE Transactions on Human Factors in Electronics, HFE-1(1), 4–11. https://doi.org/10.1109/THFE2.1960.4503259

Lu, J., et al. (2020). Learning under concept drift: A review. arXiv.

Maturana, H. R., & Varela, F. J. (1980). Autopoiesis and cognition: The realization of the living. D. Reidel.

Montévil, M., & Mossio, M. (2015). Biological organisation as closure of constraints. Journal of Theoretical Biology, 372, 179–191.

Moreno, A., & Mossio, M. (2015). Biological autonomy: A philosophical and theoretical enquiry. Springer.

Mossio, M., & Bich, L. (2017). What makes biological organisation teleological? Synthese, 194(4), 1089–1114. https://doi.org/10.1007/s11229-014-0594-z

Noë, A. (2004). Action in perception. MIT Press.

Parisi, G. I., Kemker, R., Part, J. L., Kanan, C., & Wermter, S. (2019). Continual lifelong learning with neural networks: A review. Neural Networks, 113, 54–71. https://doi.org/10.1016/j.neunet.2019.01.012

Quiñonero-Candela, J., Sugiyama, M., Schwaighofer, A., & Lawrence, N. D. (2009). Dataset shift in machine learning. MIT Press.

Ramírez-Vizcaya, S., & Froese, T. (2019). The enactive approach to habits. Frontiers in Psychology, 10, 301. https://doi.org/10.3389/fpsyg.2019.00301

Rietveld, E., & Kiverstein, J. (2014). A rich landscape of affordances. Ecological Psychology, 26(4), 325–352. https://doi.org/10.1080/10407413.2014.958035

Shalev-Shwartz, S. (2012). Online learning and online convex optimization. Foundations and Trends in Machine Learning, 4(2), 107–194. https://doi.org/10.1561/2200000018

Smith, L. B., & Thelen, E. (2003). Development as a dynamic system. Trends in Cognitive Sciences, 7(8), 343–348.

Suchman, L. A. (1987). Plans and situated actions: The problem of human–machine communication. Cambridge University Press.

Sutton, R. S., & Barto, A. G. (2018). Reinforcement learning: An introduction (2nd ed.). MIT Press.

Thelen, E., & Smith, L. B. (1994). A dynamic systems approach to the development of cognition and action. MIT Press.

Thompson, E. (2007). Mind in life: Biology, phenomenology, and the sciences of mind. Harvard University Press.

Thompson, E., & Stapleton, M. (2009). Making sense of sense-making. Topoi, 28(1), 23–30.

Turrigiano, G. (2012). Homeostatic synaptic plasticity: Local and global mechanisms for stabilizing neuronal function. Cold Spring Harbor Perspectives in Biology.

Varela, F. J. (1979). Principles of biological autonomy. North-Holland.

Varela, F. J., Thompson, E., & Rosch, E. (1991). The embodied mind: Cognitive science and human experience. MIT Press.

Weber, A., & Varela, F. J. (2002). Life after Kant. Phenomenology and the Cognitive Sciences, 1(2), 97–125.

Widmer, G., & Kubat, M. (1996). Learning in the presence of concept drift and hidden contexts. Machine Learning, 23(1), 69–101. https://doi.org/10.1007/BF00116900

Žliobaitė, I., Bifet, A., Pfahringer, B., & Holmes, G. (2015). Active learning with drifting streaming data. IEEE Transactions on Neural Networks and Learning Systems, 26(10), 2412–2426. https://doi.org/10.1109/TNNLS.2014.2346671

Žliobaitė, I., Pechenizkiy, M., & Gama, J. (2016). An overview of concept drift applications. Big Data Analysis, 1(1), 1–31.