Co-Creativity as Regulated Sense-Making Under Interactional Drift

-Nicholas Davis, PhD


Abstract

Computational creativity research has largely focused on mechanisms for idea generation, novelty, and evaluative criteria such as surprise or value. While these approaches have produced increasingly powerful creative systems, they leave a central phenomenon undertheorized: creative interaction unfolds over time under conditions of drift. Constraints shift, intentions evolve, meanings reconfigure, and alignment can degrade long before visible failure occurs. As a result, systems may continue producing novel outputs while creative engagement stagnates or collapses. This paper reframes creativity not as optimization or output generation, but as regulated sense-making under interactional drift. We introduce the concept of interactional drift to describe gradual misalignment between agent behavior, evolving context, and emerging creative direction, and we distinguish regulation from optimization and learning-as-performance. Regulation, in this account, concerns maintaining the viability and coherence of engagement as conditions change, rather than converging on predefined objectives. We illustrate this framework through a co-creative drawing system, Aether, which demonstrates regulative dynamics such as stabilization, reorganization, and yielding during long-duration interaction. The paper concludes by outlining implications for the design and evaluation of computational creativity systems, emphasizing temporal coherence, drift sensitivity, and sustained engagement over static performance metrics.

1. Introduction: The Missing Problem in Computational Creativity

1.1 Motivation

Computational creativity systems have become increasingly powerful generators. Advances in search, learning, and large-scale generative models have enabled systems to produce artifacts that are novel, stylistically rich, and often indistinguishable from human output at the level of isolated examples (Boden, 2004; Colton, Pease, & Ritchie, 2001; Gatys, Ecker, & Bethge, 2016; Brown et al., 2020). In many domains—visual art, music, text, and design—systems now exhibit impressive local performance (McCormack et al., 2014; Elgammal et al., 2017).

Yet despite these advances, a persistent limitation remains visible in practice: long-duration creative interaction often fails. Over extended engagement, creative systems tend to stagnate, collapse, or lose coherence. Interaction may initially feel promising, but gradually becomes repetitive, directionless, or brittle (Davis, 2013; Karimi et al., 2020). In other cases, novelty continues to appear, yet the interaction itself feels incoherent—outputs no longer seem meaningfully related to what has come before, nor to the evolving creative situation (Kantosalo & Riihiaho, 2019).

Importantly, these failures do not always coincide with obvious technical breakdowns. Systems may continue to generate novel artifacts, satisfy evaluative metrics, or respond correctly to prompts. Nonetheless, the creative process itself deteriorates. This gap between local generative success and global interactional failure suggests that a central problem in computational creativity has been under-theorized (Jordanous, 2012; Edmonds, 2018).

1.2 Diagnosis

Prevailing approaches in computational creativity implicitly assume that creativity is primarily a problem of generation and evaluation. Creative systems are designed to produce ideas or artifacts, while evaluative mechanisms—novelty, value, surprise, or diversity—are used to guide selection or learning (Ritchie, 2007; Wiggins, 2006; Grace et al., 2015). Context, when considered at all, is typically treated as stable, externally specified, or slowly varying relative to the system’s internal dynamics.

Under these assumptions, breakdown is expected to appear as error: a failure to generate novelty, a drop in quality, or an inability to meet predefined criteria. However, many creative failures do not take this form. Instead, creative interaction often degrades before any explicit error occurs. Misalignment accumulates gradually. The system continues to “work,” yet the interaction loses direction, relevance, or meaning (Edmonds & Candy, 2010; Davis, 2013).

We refer to this phenomenon as interactional drift. Drift describes the gradual misalignment between a system’s behavior, the evolving creative context, and the participant’s emerging intentions. Crucially, drift is not equivalent to error. It can occur while outputs remain valid, novel, or even impressive in isolation. Because drift does not immediately violate evaluative criteria, it often goes undetected until coherence collapses (Suchman, 1987; De Jaegher & Di Paolo, 2007).

Existing frameworks lack a way to describe this failure mode. By focusing on outputs rather than trajectories, and optimization rather than regulation, they overlook the dynamics by which creative engagement is sustained—or lost—over time (Jordanous & Keller, 2016; Kantosalo et al., 2015).

1.3 Thesis

This paper advances a simple but consequential claim:

Co-creativity is not best understood as shared novelty production, but as the capacity to sustain coherent engagement under drift through emergent, interactionally negotiated trajectories rather than fixed objectives or evaluative targets.

From this perspective, co-creativity is a temporal and interactional achievement. It depends not only on the capacity to generate new material, but on the capacity to regulate sense-making as conditions change—when intentions shift, constraints evolve, and meanings are renegotiated (Varela, Thompson, & Rosch, 1991; De Jaegher & Di Paolo, 2007). Creative success, on this account, is measured not solely by what is produced, but by whether engagement remains viable, responsive, and coherent across time (Edmonds, 2018; Davis, 2013).

1.4 Contributions

This paper makes four contributions:

Together, these contributions aim to extend the conceptual foundations of computational creativity toward systems that do not merely generate novelty, but remain meaningfully engaged as creative situations evolve.

2. Related Perspectives

This section situates the present work relative to existing approaches in computational creativity and cognitive science. Rather than surveying the field exhaustively, we highlight a small number of perspectives that illuminate the specific gap this paper addresses. The aim is not to adjudicate between paradigms, but to clarify what remains under-theorized.

2.1 Creativity as Search and Generation

A large body of work in computational creativity treats creativity as a problem of search, generation, or exploration within a defined space. This includes classical search-based models, evolutionary and genetic algorithms, and more recent large-scale generative approaches (Boden, 2004; Wiggins, 2006; Colton et al., 2011). In these systems, creativity emerges through the production of novel candidates and their selection according to evaluative criteria such as novelty, value, surprise, or fitness (Ritchie, 2007; Grace et al., 2015).

These approaches have been highly successful in producing compelling artifacts and scaling creative output (McCormack et al., 2014; Elgammal et al., 2017; Brown et al., 2020). However, they typically presuppose stable or externally defined evaluation criteria. Even when criteria adapt over time, the adaptation itself is usually framed as an optimization problem with respect to a performance signal (Wiggins, 2006; Jordanous, 2012).

As a result, these models have limited resources for addressing temporal coherence in interaction. Drift is either invisible—because outputs remain novel—or treated as noise relative to the optimization objective. The question of how creative engagement remains meaningful across time is largely displaced by the question of how effectively a space is searched (Edmonds, 2018).

2.2 Process-Oriented and Co-Creative Systems

A second line of work emphasizes process, interaction, and co-creation. Mixed-initiative systems, interactive creativity tools, and human–AI collaboration frameworks foreground responsiveness, turn-taking, and shared control (Horvitz, 1999; Davis, 2013; Yannakakis et al., 2014). These systems acknowledge that creativity unfolds through interaction rather than isolated generation.

This shift marks an important advance. By situating creativity in ongoing engagement, process-oriented systems recognize that timing, responsiveness, and context matter (Edmonds & Candy, 2010; Kantosalo et al., 2015). However, many such systems remain fundamentally reactive. They respond to user input, adapt locally, or mirror detected patterns, but lack mechanisms for regulating interaction over longer horizons.

In practice, this often leads to familiar failure modes: repetition masked as responsiveness, escalation without direction, or collapse into incoherence as interaction history accumulates (Davis, 2013; Kantosalo & Riihiaho, 2019). Without a regulative account of how engagement should be maintained under changing conditions, responsiveness alone proves insufficient for sustaining creative trajectories.

2.3 Sense-Making and Enactive Accounts

The present work is closely aligned with enactive and dynamical approaches to cognition, which treat sense-making as an active, embodied, and relational process. In these accounts, meaning does not arise from internal representations alone, but from ongoing engagement between agent and environment (Varela, Thompson, & Rosch, 1991; Thompson, 2007). Perception, action, and affect are understood as dynamically coupled rather than sequentially staged (Di Paolo, 2005; Kelso, 1995).

These perspectives provide crucial conceptual resources for understanding creativity as something enacted over time rather than computed instantaneously. In particular, they foreground the historical and situated nature of sense-making—an emphasis that directly informs our treatment of creative drift (De Jaegher & Di Paolo, 2007; Di Paolo & De Jaegher, 2012).

At the same time, it is important to delimit scope. This paper builds on enactive and dynamical perspectives, but does not offer a theory of perception or cognition per se. Our concern is not to extend enactive theory, but to adapt its regulatory insights to a specific problem in computational creativity: sustaining coherent engagement under interactional drift.

2.4 Gap Summary

Taken together, existing perspectives leave a critical gap unaddressed. The computational creativity literature lacks: 1) an explicit account of interactional drift as a primary failure mode, and 2) a concept of regulation that is distinct from optimization, learning, or belief updating. As a result, creative systems are often evaluated in terms of what they produce, rather than how engagement unfolds across time (Jordanous & Keller, 2016; Edmonds, 2018). The framework introduced in this paper aims to address this gap by reframing creativity as regulated sense-making—an ongoing achievement rather than a static capability. This reframing does not replace existing approaches. It complements them by making visible a dimension of creative interaction that current models implicitly rely on, but rarely articulate.

3. Creativity Under Interactional Drift

Creative interaction does not unfold in a static context. Over time, intentions shift, constraints are reinterpreted, and what “counts” as a promising direction evolves (Dewey, 1934; Schön, 1983). Yet most computational creativity systems implicitly assume that the conditions under which creativity operates remain stable, or at least externally specified. This section introduces interactional drift as the missing phenomenon that explains why creative systems often fail over extended interaction—even when local outputs remain successful (Edmonds, 2018; Davis, 2013).

3.1 What Is Interactional Drift?

Interactional drift refers to the gradual misalignment that accumulates over time between: an agent’s behavior, the evolving context of interaction, and the emerging direction of creative activity. Drift is not a sudden rupture. It unfolds incrementally as creative interaction proceeds. Each contribution may be locally appropriate, yet increasingly out of sync with the trajectory that the interaction itself is taking on (Suchman, 1987; De Jaegher & Di Paolo, 2007).

Crucially, drift is:

In co-creative settings, neither the human nor the system fully determines the direction of creativity in advance. Direction emerges through interaction (Sawyer, 2012). Drift arises when this emergent direction changes faster than the agent’s behavior adapts to it.

3.2 Drift Is Not Error

A central reason drift is under-theorized is that it does not register as error in conventional terms. Drift can occur while: generated artifacts remain syntactically valid, outputs continue to satisfy novelty or quality metrics, user input is being responded to correctly. From the perspective of local evaluation, nothing appears to be wrong. Yet at the level of interaction, coherence begins to degrade. Contributions feel less aligned, timing becomes awkward, and the interaction loses a sense of direction (Edmonds, 2018; Kantosalo & Riihiaho, 2019).

Drift therefore precedes breakdown. By the time failure becomes visible—through user disengagement, repetition, or collapse into incoherence—the system has already been compensating for misalignment for some time (Davis, 2013). Error-based or outcome-based signals arrive too late to guide meaningful reorganization.

This temporal gap explains why systems optimized for correctness, novelty, or responsiveness often fail to sustain creative engagement. They detect failure only after coherence has already eroded (Jordanous, 2012; Ritchie, 2007).

3.3 Drift in Creative Practice

In human creative practice, drift is not an anomaly—it is a feature. Artists routinely revise intentions mid-process, reinterpret constraints, and follow directions that were not initially anticipated (Dewey, 1934; Schön, 1983). A sketch suggests a new form; a musical phrase reframes a composition; a narrative thread opens an unexpected thematic space. Creative direction, in these cases, is emergent rather than fixed. Coherence is maintained not by adhering to an initial plan, but by remaining sensitive to how the work is unfolding (Sawyer, 2012; Ingold, 2013).

Computational systems, by contrast, typically assume that: objectives are stable, constraints are fixed or explicitly updated, evaluation criteria are known in advance (Wiggins, 2006; Grace et al., 2015). When creative direction shifts implicitly—through interaction history or evolving context—these systems lack mechanisms to notice, let alone regulate, the resulting misalignment. They continue to act as if the creative situation were unchanged (Edmonds & Candy, 2010).

3.4 Consequence: Structural Drift Without Awareness

The consequence of ignoring interactional drift is that systems optimized for local success can drift structurally without noticing. They continue to generate valid outputs while losing their grip on the creative situation they are embedded in (Suchman, 1987; Davis, 2013).

This failure mode is especially pronounced in long-duration interaction. Short exchanges can mask drift; extended collaboration exposes it (Kantosalo et al., 2015). Without a concept of regulation distinct from optimization or learning, systems have no way to respond to the slow accumulation of misalignment that defines creative breakdown.

Recognizing interactional drift reframes the problem of computational creativity. The central challenge is no longer how to generate novel artifacts, but how to maintain coherent engagement over time as creative conditions change. The next section introduces regulated sense-making as a response to this challenge.

4. Co-Creativity as Regulated Sense-Making

The central claim of this paper is that creativity is not best understood as the production of novel artifacts or ideas, but as the capacity to sustain coherent engagement under conditions of interactional drift. This claim aligns with growing critiques of output-centric models of creativity and intelligence, which argue that novelty and quality alone are insufficient to explain sustained creative practice (Dewey, 1934; Sawyer, 2012; Edmonds, 2018). Instead, creativity must be understood as a form of regulated sense-making that unfolds over time.

This shift requires rethinking both creativity and intelligence: away from generation and evaluation, and toward the regulation of engagement as conditions change (Di Paolo et al., 2017; Davis, 2013).

Definition of Co-Creativity

In this paper, co-creativity is not defined by shared authorship, role symmetry, or collaborative intent. Nor is it equated with responsiveness, turn-taking, or mutual influence alone (cf. Grace et al., 2015). Instead, co-creativity is understood as a form of regulated sense-making that unfolds at the level of interaction.

Under this account, creative activity becomes co-creative when coherence is maintained not by any single agent, but by the ongoing regulation of salience, relevance, and direction across participants as conditions change (De Jaegher & Di Paolo, 2007; Sawyer, 2012). What is shared is not a goal, representation, or evaluative criterion, but the responsibility for sustaining viable engagement under drift.

Co-creativity, on this view, is therefore neither reducible to individual creativity multiplied nor to coordination toward predefined outcomes. It is an emergent property of interactional regulation: a trajectory of sense-making that persists only insofar as drift is detected, negotiated, and reorganized within the interaction itself (Davis, 2013; Edmonds & Candy, 2010).

4.1 Sense-Making: A Minimal Definition

Sense-making refers to the ongoing organization of engagement with a situation (Varela et al., 1991; Thompson, 2007). In creative contexts, this organization concerns:

Importantly, sense-making is not a representational process. It does not involve constructing an internal model of the world or selecting among predefined interpretations (Hutto & Myin, 2017). Nor is it primarily a matter of belief, evaluation, or judgment applied after the fact. Instead, sense-making operates prior to these reflective activities, shaping what becomes thinkable, evaluable, or actionable in the first place.

Creativity, on this view, is not the generation of ideas within a fixed space, but the continuous reorganization of the space itself as interaction unfolds (Ingold, 2013; Sawyer, 2012). What counts as an “idea,” a “constraint,” or a “promising direction” is not given in advance; it emerges through engagement over time.

4.2 Regulation: A Crucial Distinction

If creativity is a form of sense-making, then the key question becomes how this sense-making is sustained as conditions change. This is where regulation becomes central. Regulation must be distinguished from several familiar concepts:

In this sense, regulation refers to the maintenance of viability under change. A system is regulated when it can continue to participate meaningfully in an evolving interaction without collapsing into rigidity or incoherence. This distinction is critical. Many creative systems adapt locally while drifting structurally. They improve according to internal metrics while losing alignment with the creative situation they are embedded in (Edmonds, 2018; Davis, 2013).

4.3 Regulative Dynamics: Clamping and Unclamping

To make regulation concrete without over-formalizing it, we introduce two complementary regulative dynamics: clamping and unclamping, drawing explicitly on Davis et al. (2017). Clamping refers to the temporary stabilization of perceptual or interactional organization. In creative interaction, clamping occurs when a particular pattern, motif, or direction is held long enough to allow coherence to form. Clamping is not fixation; it is provisional stability. It enables local continuity without freezing the system permanently (Davis et al., 2017). Unclamping refers to the reopening of degrees of freedom when coherence degrades. As interactional drift accumulates, previously successful organizations may no longer fit. Unclamping loosens constraints, allowing new patterns, directions, or interpretations to emerge (Davis et al., 2017; Kelso, 1995).

These dynamics operate continuously and reciprocally. Excessive clamping leads to rigidity and stagnation; excessive unclamping leads to fragmentation and noise. Regulation consists in modulating between these extremes in response to drift, rather than eliminating drift altogether. Crucially, neither clamping nor unclamping is triggered by error in the conventional sense. They are responses to changes in coherence, not to violations of correctness or performance thresholds (Davis, 2013).


4.4 Figure 1: Perceptual Regulation Under Drift

Figure 1 illustrates perceptual regulation as a temporal process unfolding under drift. Over time, perceptual or interactional organization gradually misaligns with evolving conditions (Suchman, 1987). Clamping provides temporary stabilization, allowing engagement to remain coherent. Unclamping reopens possibilities when stabilization becomes counterproductive.

The figure emphasizes three key points:

By making these dynamics visible, the figure reframes creative failure as a regulatory problem rather than a generative one (Edmonds & Candy, 2010).

4.5 Implications for Creativity

Under this framework, creativity succeeds not by eliminating drift, but by regulating it. Drift is unavoidable in any extended creative interaction (Dewey, 1934). What distinguishes robust creative systems—human or artificial—is their ability to remain responsive to drift without collapsing or freezing.

This reframing shifts the evaluation of creative systems. The key question is no longer “How novel or valuable are the outputs?” but “Can the system sustain coherent sense-making as creative direction evolves?” (Sawyer, 2012; Edmonds, 2018).

In the next section, we illustrate how this account can be instantiated in a co-creative system, demonstrating how regulated sense-making enables long-duration creative interaction where optimization-based approaches fail.


5. System Illustration: A Co-Creative Drawing Agent

This section illustrates the proposed framework through a co-creative drawing system, Aether. The purpose of this illustration is not to claim superior performance, benchmark results, or task completion. Rather, the system functions as a conceptual probe: a concrete instantiation that makes regulative sense-making dynamics observable in practice.

5.1 System Overview (High-Level)

Aether is a co-creative drawing agent designed to participate in turn-by-turn interaction with a human collaborator on a shared canvas. Interaction unfolds over extended sessions rather than isolated prompts. The system does not aim to complete drawings, optimize aesthetic metrics, or converge on predefined styles.

At a high level, Aether:

Crucially, learning and organization occur online. There is no offline training phase, fixed dataset, or reset between interactions. The system’s behavior reflects accumulated engagement history rather than static model parameters.


Figure X. Interactional regulation in Aether. Sequential frames from a co-creative drawing session illustrating motif emergence and cross-scale regulatory dynamics. White strokes represent human input; blue strokes represent Aether’s contributions. Across iterations, recurring structures (e.g., circular loops, angular probes, and zigzag oscillations) become stabilized as shared motifs. Rather than generating outputs independently, Aether modulates its behavior in response to emerging field tensions, reinforcing, extending, and orthogonally probing existing structures. The sequence demonstrates drift-aware regulation, motif crystallization, and multi-scale coherence in interactive creative practice.


5.2 Interactional Regulation in Practice

Aether’s behavior is best understood in terms of regulative dynamics, not output generation.

Motif emergence.

When a human introduces a recurring structure—such as circular forms, angular oscillations, or repeated spatial rhythms—the system abstracts these into motifs. These motifs are not copied stroke-by-stroke. Instead, they function as organizational tendencies that shape Aether’s contributions: scale, curvature, placement, and repetition vary while preserving structural coherence.

Motif fading.

Importantly, motifs are not accumulated indefinitely. When human behavior shifts, previously dominant motifs gradually lose influence. This fading is not triggered by explicit error signals; it reflects a loss of interactional coherence as the motif no longer aligns with emerging direction.

Regime shifts.

As motifs reorganize, the system undergoes regime shifts—qualitative changes in how it structures engagement. A regime may favor dense local elaboration, sparse probing marks, or rhythmic repetition. These regimes are not labeled or predefined; they emerge through sustained interaction.

Pauses and yielding.

Aether does not respond to every human action with maximal output. At times, it yields space, produces minimal marks, or pauses entirely. These behaviors are not failures or latency artifacts; they function as regulatory acts, allowing interactional coherence to stabilize or reorganize.

Non-accumulative behavior.

Unlike systems that continuously layer outputs, Aether avoids compounding misalignment. When coherence degrades, it does not intensify production to compensate. Instead, it reduces commitment—an instance of unclamping—before re-engaging.

Together, these behaviors reflect regulation at the level of interactional organization rather than goal pursuit.

5.3 Drift Sensitivity

The system’s defining property is its sensitivity to interactional drift.

When human drawing behavior changes—shifting aesthetic direction, altering pacing, or introducing qualitatively new structures—Aether does not treat this as noise or anomaly. It reorganizes its participation accordingly.

Three features are central:

This sensitivity allows interaction to remain viable over extended durations without relying on externally imposed resets or evaluation thresholds.

5.4 What the System Demonstrates

The system illustration supports three core claims of the paper:

While space constraints limit the number of examples presented here, interaction snapshots (Figure X) illustrate motif emergence, regime shifts, and yielding behavior. These visual traces make regulative dynamics legible without requiring technical exposition.

The value of this system lies not in what it produces, but in how it remains in relation to a changing creative situation—demonstrating creativity as regulated sense-making under interactional drift.

6. Implications for Evaluation and Design

Reframing creativity as regulated sense-making under interactional drift has direct consequences for how computational creativity systems are evaluated and designed. Many existing evaluation practices implicitly assume static tasks, stable criteria, and short interaction horizons. These assumptions obscure the very phenomena that determine whether creative systems remain viable over time.

6.1 Why Current Metrics Fail

Most computational creativity metrics are output-centric. They assess creativity by measuring properties of generated artifacts—novelty, surprise, value, or quality—often in isolation from the interaction that produced them.

From a regulative perspective, these metrics are insufficient.

Novelty does not imply coherence.

A system can continuously generate novel outputs while drifting structurally—producing artifacts that no longer relate meaningfully to prior interaction or emerging direction. Novelty may increase even as engagement degrades.

Surprise does not imply sense-making.

Surprising outputs can signal productive deviation, but they can also indicate loss of coupling. Without regulation, surprise becomes indistinguishable from noise.

Output quality does not imply interaction quality.

High-quality individual artifacts can be produced within interactions that feel brittle, exhausting, or directionless. Conversely, interactions that sustain engagement may include imperfect or unfinished outputs.

These failures are not flaws in metric design per se; they reflect a mismatch between what is measured and what actually matters for long-duration creative interaction.

6.2 Alternative Evaluation Questions

If creativity is understood as the ability to maintain coherent engagement under drift, evaluation must shift from artifact assessment to interactional dynamics. Rather than asking whether a system produces creative outputs, evaluation should ask whether it supports creative trajectories. Key questions include:

These questions are not easily reducible to scalar metrics. They require longitudinal observation, qualitative analysis, and attention to temporal structure—precisely the dimensions often excluded from evaluation.

6.3 Design Implications

Viewing creativity as regulated sense-making suggests concrete design principles for future systems. First, drift detection should be treated as a first-class concern. Systems must be able to sense gradual misalignment before breakdown occurs, rather than reacting only to explicit failure signals. Second, systems require regulative mechanisms distinct from learning or optimization. These include the ability to temporarily stabilize organization (clamping) and to release or reorganize it when coherence degrades (unclamping). Third, creative systems must exhibit temporal sensitivity. Short-horizon responsiveness is insufficient. Systems must track interactional history, allow patterns to fade, and respect the pacing of reorganization. Together, these implications suggest a shift in how creativity systems are conceived. Rather than designing generators that optimize outputs, we can design participants that regulate their engagement—remaining responsive, coherent, and viable under change.

7. Limits, Scope, and Misuse Risks

The account developed in this paper is intentionally scoped. By framing creativity as regulated sense-making under interactional drift, we aim to illuminate a specific failure mode in computational creativity systems—not to offer a universal theory of creativity or a replacement for existing approaches. Clarifying these boundaries is essential to prevent overextension and misapplication.

7.1 What This Account Does Not Claim

First, not all creativity is interactive. Many creative processes—both human and computational—are solitary, offline, or artifact-centric. The present account is concerned specifically with interactive and co-creative systems, where engagement unfolds across time and mutual influence.

Second, not all drift is perceptual or interactional. Drift can arise from hardware degradation, data corruption, task redefinition, or external constraints that lie outside the scope of perceptual or sense-making organization. Regulation at the interactional level cannot compensate for all sources of instability.

Third, regulation does not guarantee success. Regulating drift does not ensure creative outcomes, insight, or value. Interactions may still fail, terminate, or lose coherence due to factors beyond the system’s regulatory capacity. The claim is not that regulation prevents failure, but that it allows failure to occur without catastrophic collapse or premature stagnation.

These clarifications are not weaknesses of the account. They reflect its commitment to realism about creative systems operating under open-ended conditions.

7.2 Misuse Risks

The concept of regulation carries its own risks. One risk is treating regulation as control. Regulation, as described here, concerns maintaining viability under change—not enforcing stability or steering interaction toward predefined outcomes. When regulation is implemented as constraint enforcement or goal maintenance, it undermines the very flexibility it is meant to support. A second risk is instrumentalizing attunement. If regulation is used solely to optimize engagement metrics, productivity, or novelty rates, it collapses back into performance optimization. Attunement becomes a means to an end rather than a condition for coherent interaction. A third risk is over-monitoring interaction. Excessive tracking, intervention, or meta-control can destabilize engagement by interrupting the dynamics it seeks to regulate. Regulation must remain lightweight and responsive rather than supervisory. These risks highlight that regulation is not a technique to be maximized, but a capacity to be carefully integrated.

7.3 Open Questions

The framework also raises important open questions. One concerns scaling to collectives. How do regulative dynamics operate in multi-agent or group creative systems where coherence must be negotiated across many participants? Another concerns formalization. While this paper introduces clamping and unclamping conceptually, further work is needed to formalize these dynamics across different architectures and modalities without reducing them to optimization heuristics. Finally, the relation between regulation and learning remains an open area. Regulation operates at the level of organization rather than parameter optimization, but how these layers interact over long time horizons warrants deeper investigation. Addressing these questions is essential for extending the account beyond individual systems and for integrating it responsibly into broader computational creativity research.

8. Conclusion: Toward Creativity Without Optimization

This paper has argued for a reframing of computational creativity. Rather than understanding creativity primarily as the production of novel outputs, we have characterized it as the capacity to sustain coherent engagement under interactional drift. The central challenge for creative systems is not generating novelty, but remaining oriented as contexts, intentions, and interactional dynamics change over time.

By introducing interactional drift as a core phenomenon, we highlight a mode of misalignment that precedes breakdown and often remains invisible to standard metrics. Addressing this requires a notion of regulation distinct from optimization or learning-as-performance. Regulation, in this account, concerns maintaining the viability of sense-making under change rather than maximizing outcomes against fixed criteria.

The contribution of this work is therefore a new problem framing and set of design criteria for computational creativity. Through the example of a co-creative drawing agent, we show how regulative dynamics—such as stabilization and reorganization—can support long-duration creative interaction without explicit optimization.

Creativity persists not by optimizing outcomes, but by remaining oriented under change.

Acknowledgements

The author acknowledges the use of an AI language model (ChatGPT, developed by OpenAI) as a conversational and editorial aid during the development of this paper. The system was used to support brainstorming, clarification of concepts, and iterative refinement of the manuscript. All theoretical framing, design decisions, interpretations, and conclusions are solely those of the author. The AI system did not conduct experiments, determine research claims, or function as an author or co-author.


References

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

Boden, M. A. (2004). The creative mind: Myths and mechanisms (2nd ed.). Routledge.

Brown, T. B., et al. (2020). Language models are few-shot learners. Advances in Neural Information Processing Systems, 33, 1877–1901.

Colton, S., Pease, A., & Ritchie, G. (2001). AI and creativity: A survey. Artificial Intelligence Review, 16(1), 215–229.

Colton, S., Wiggins, G. A., & McCormack, J. (2011). Computational creativity: The final frontier? Proceedings of ECAI.

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.

Davis, N. (2013). Human–computer co-creativity: Blending human and computational creativity. Springer.

Davis, N., Hsiao, C.-P., Singh, K., & Magerko, B. (2017). Building creative systems by regulating internal and external pressures. Proceedings of Creativity & Cognition (C&C).

De Jaegher, H., & Di Paolo, E. A. (2007). Participatory sense-making. Phenomenology and the Cognitive Sciences, 6(4), 485–507.

Dewey, J. (1934). Art as experience. Perigee.

Di Paolo, E. A. (2005). Autopoiesis, adaptivity, teleology, agency. Phenomenology and the Cognitive Sciences, 4(4), 429–452.

Di Paolo, E. A., & De Jaegher, H. (2012). The interactive brain hypothesis. Frontiers in Human Neuroscience, 6, 163.

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

Edmonds, E. (2018). The role of the artist in generative and computer-based art. Leonardo, 51(3), 235–239.

Edmonds, E., & Candy, L. (2010). Interactive art and the challenge of sustained engagement. Proceedings of Creativity & Cognition.

Edmonds, E., & Candy, L. (2010). Relating theory, practice and evaluation in interactive art. Leonardo, 43(5), 470–476.

Elgammal, A., Liu, B., Elhoseiny, M., & Mazzone, M. (2017). CAN: Creative adversarial networks. Proceedings of the International Conference on Computational Creativity (ICCC).

Gatys, L. A., Ecker, A. S., & Bethge, M. (2016). Image style transfer using convolutional neural networks. Proceedings of CVPR.

Grace, K., et al. (2015). Evaluating creative systems. Proceedings of ICCC.

Grace, K., Maher, M. L., Fisher, D., & Brady, K. (2015). Modeling expectation for evaluating creative systems. Proceedings of ICCC.

Horvitz, E. (1999). Principles of mixed-initiative user interfaces. Proceedings of CHI.

Hutto, D. D., & Myin, E. (2017). Evolving enactivism. MIT Press.

Ingold, T. (2013). Making: Anthropology, archaeology, art and architecture. Routledge.

Jordanous, A. (2012). A standardised procedure for evaluating creative systems. Cognitive Computation, 4(3), 246–279.

Jordanous, A., & Keller, B. (2016). Modelling creativity: Identifying key components through a corpus-based approach. PLOS ONE, 11(3).

Kantosalo, A., & Riihiaho, S. (2019). Creative systems in extended interaction. Proceedings of ICCC.

Kantosalo, A., Toivonen, H., & Grobler, J. (2015). Two approaches to creative interaction. Proceedings of ICCC.

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

McCormack, J., et al. (2014). Ten questions concerning generative computer art. Leonardo, 47(2), 135–141.

Ritchie, G. (2007). Some empirical criteria for attributing creativity to a computer program. Minds and Machines, 17(1), 67–99.

Sawyer, R. K. (2012). Explaining creativity: The science of human innovation (2nd ed.). Oxford University Press.

Schön, D. A. (1983). The reflective practitioner. Basic Books.

Suchman, L. A. (1987). Plans and situated actions. Cambridge University Press.

Thompson, E. (2007). Mind in life. Harvard University Press.

Varela, F. J., Thompson, E., & Rosch, E. (1991). The embodied mind. MIT Press.

Wiggins, G. A. (2006). A preliminary framework for description, analysis and comparison of creative systems. Knowledge-Based Systems, 19(7), 449–458.

Yannakakis, G. N., Liapis, A., & Alexopoulos, C. (2014). Mixed-initiative co-creativity. Proceedings of the Foundations of Digital Games (FDG).