REVIEW 3 major objections 6 minor 14 references
Extended Creativity: A Conceptual Framework for Understanding Human-AI Creative Relations
T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash
Pith's one-line read This paper proposes that human-AI creative collaboration takes three distinct relational modes—Support, Synergy, and Symbiosis—defined by technical autonomy and perceived agency, and that each mode affects a different level of creativity.
desk verdict Useful conceptual synthesis with a real internal gap: the two dimensions named actually allow off-diagonal configurations the taxonomy doesn't cover; fixable, but central. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The machinery is a two-dimensional taxonomy built from existing conceptual tools. Technical autonomy is graded in five levels from transparent-deterministic systems to open, co-evolving systems; agency is graded in three levels from causal effect-production to perceived intentionality; and creative outcomes are placed on the Four-C scale (mini-c, little-c, pro-c, big-C). Each of the three modes—Support, Synergy, Symbiosis—is defined by a specific band on the autonomy and agency scales, and the paper derives from that configuration which creativity levels the mode can meaningfully affect. The taxonomy also includes two orthogonal operational dimensions—coordination mechanism (hierarchical vs stigmergic) and relational polarity (cooperative vs competitive)—to cover cases where the same mode feels collaborative or adversarial.
What would settle it
Give two groups of creators the same generative model with identical capabilities, but present one group with a Support-style interface (fixed operations, no visible adaptation, human retains final say) and the other with a Synergy-style interface (visible responsiveness, suggestions that adapt to feedback). If the Support group produces professional-level, peer-recognized creative work as often as the Synergy group, the claim that mode determines creativity level fails. If the Synergy group shows a clear shift from everyday to professional creativity with the underlying model held constant, the perceived-agency axis is doing real work.
Extended reading notes
Core claim
The paper's central claim is that human-AI creative relations are not one phenomenon but three qualitatively different relational modes, and that the differences can be captured along two dimensions: technical autonomy (the system's capacity to operate independently of direct human intervention) and perceived agency (the degree to which the system is experienced as an intentional, influential creative partner). In Support, autonomy is low and agency is causal only; AI executes fixed operations while the human retains full creative control. In Synergy, autonomy reaches adaptive, context-responsive levels and agency becomes contingent, with AI generating alternatives that influence the human's next move. In Symbiosis, autonomy is at the highest, self-modifying levels and agency approaches perceived intentionality, so the human-AI pair functions as a single creative unit. The paper predicts these modes map onto the Four-C levels of creativity—mini-c and little-c for Support, through pro-c for Synergy, and potentially big-C for Symbiosis—and argues that creative value does not scale linearly with autonomy.
Load-bearing premise
The framework assumes that technical autonomy and perceived agency are the two dimensions that adequately distinguish human-AI creative relationships; if trust, task type, or who holds final evaluative authority matters more, the predicted link between mode and creativity level would not hold.
Editorial extensions
If this is right
- If the taxonomy is right, Support systems can democratize everyday creativity by lowering technical barriers, but they will not, by themselves, push creators into professional or paradigm-shifting territory.
- Synergy systems should be the practical target for raising professional creative output, since their adaptive dialogue expands a creator's imaginative space while leaving final authorship and evaluation with the human.
- Genuine big-C creativity would require Symbiosis-level integration, which implies that efforts to achieve transformative AI-assisted breakthroughs must address deeply coupled human-AI integration, not just better prompts.
- Ethical questions change character across modes: authorship and ownership stay clear in Support, become contested in Synergy, and turn into questions of cognitive privacy, identity, and equitable access in Symbiosis.
- The paper's proposed epistemic-surprise measure offers a way to operationalize creative agency as the capacity to induce coherent, unexpected model updates in a human collaborator, independent of output quality ratings.
Reading between the lines
- Editorial inference: The two-axis logic suggests a direct design experiment—hold the underlying AI constant and change only the interface framing, autonomy levers, and feedback responsiveness, then measure which Four-C level users reach; if perceived agency alone shifts creative outcomes with no change in technical autonomy, the second axis carries real causal weight.
- Editorial inference: Because the paper itself concedes that creative value does not scale linearly with autonomy, the mode-to-creativity mapping is likely moderated by variables the taxonomy does not include, such as trust, domain expertise, and who holds final evaluative authority; empirical work should measure those moderators explicitly.
- Editorial inference: The epistemic-surprise proposal could be turned into a quantitative falsification target: compute the prediction-error reduction a collaborator experiences and test whether Support, Synergy, and Symbiosis configurations produce measurably different surprise profiles, giving the framework an operational handle beyond self-report.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a conceptual framework for understanding human-AI creative relations, introducing the notion of 'Extended Creativity' systems. It defines three relational modes—Support, Synergy, and Symbiosis—along two dimensions: technical autonomy (using Simmler and Frischknecht's five levels) and perceived agency (using Rammert's causal/contingent/intentional gradations). Each mode is mapped to levels of Kaufman and Beghetto's Four-C creativity model, and the paper discusses ethical, research, and design implications, including a proposed 'epistemic surprise' metric. The paper is explicitly a conceptual contribution; it presents no empirical data and acknowledges that Symbiosis is largely theoretical.
Significance. If the framework holds, it offers a useful organizing vocabulary for human-AI co-creativity research, integrating two established taxonomies and generating falsifiable predictions about which relational modes affect which creativity levels. The paper is commendably grounded in prior literature, explicitly flags its speculative components, and proposes a novel operationalization of creative agency via epistemic surprise. However, the central dimensional basis of the taxonomy has an internal gap (see major comments), and the creativity-impact mapping is asserted rather than demonstrated. These issues are fixable within the manuscript's scope, making the paper a promising but not yet fully coherent conceptual contribution.
major comments (3)
- [§4.2, §5, Table 1] The taxonomy does not follow from the two stated dimensions. Section 4.2 explicitly states that 'a system may possess a high degree of autonomy—operating independently—yet still lack agency,' which under the paper's own definitions admits four combinations of autonomy and agency. However, Section 5 and Table 1 assign modes only to the ascending diagonal: Support (low autonomy + causal agency), Synergy (mid autonomy + contingent agency), Symbiosis (high autonomy + intentional agency). High-autonomy/low-agency systems (e.g., an autonomous generator whose outputs are routinely ignored) and low-autonomy/high-agency systems (e.g., a deterministic tool users believe is an intentional partner) are left unnamed and unanalyzed. This is an internal inconsistency between the dimensional definitions and the mode table, and it means the modes are not derived from the dimensions but are stipulated prototypes. Please either explicitly frame the modes as diagonal prototypes or extend the taxonomy to cover off-diagonal configurations.
- [Table 1; §5.1–5.3, §8] The mapping from mode to creativity level (Support→mini-c/little-c, Synergy→pro-c, Symbiosis→all levels including big-C) is asserted rather than derived or evidenced. The sections provide qualitative vignettes and claims, but no mechanism linking autonomy and agency to the Four-C levels is specified, and no empirical evidence is presented. Since the paper's central claim is that these modes 'differentially affect' creativity levels, the reader needs at least a principled argument for the mapping. Section 8's concession that creative value does not scale linearly with autonomy or agency further undermines any implicit assumption that the diagonal is a value axis. The authors should state the assumed causal mechanisms and, if these are intended as predictions, specify how each could be tested.
- [§4] The selection of technical autonomy and perceived agency as the two key dimensions is not justified comparatively. Other candidate dimensions—such as trust, task type, or who holds final evaluative authority—are mentioned in the surrounding literature but are not systematically compared. Because the entire taxonomy rests on these two axes, a brief argument for why these dimensions are the most decisive for creative outcomes would strengthen the framework. At minimum, the paper should acknowledge that this choice is a design decision and discuss how different choices might alter the taxonomy.
minor comments (6)
- [§3.2, §5.4] Section numbering is inconsistent: Section 3.2 appears twice (once for 'The evolution of human-AI creativity' and once for 'Distributed creativity'), and Section 5.4 refers to 'Section 3.4,' which does not exist.
- [§3.2, §6.2] There are typographical errors: 'has lead to' in Section 3.2 should be 'has led to'; 'Extended Creativity e systems' in Section 6.2 should be 'Extended Creativity systems.'
- [§3.2 references] The reference to Kahneman (2002) for System 1 and System 2 is imprecise; the standard citation is Kahneman (2011), 'Thinking, Fast and Slow.'
- [Figure 4] Figure 4 is not explicitly described or called out in the text beyond its caption; please add a sentence explaining what the figure illustrates and how it relates to the taxonomy.
- [§5.4, Table 1] The operational modes in Section 5.4 (coordination mechanisms, relational polarity) are listed in Table 1 but their relationship to the three relational modes is not systematically explained; a brief discussion of how they combine would improve clarity.
- [§7] The epistemic-surprise metric in Section 7 is only a sketch; if it is to be a contribution, please provide at least a formal definition or reference to a specific method for computing it, or explicitly label it as a direction for future work.
Circularity Check
No significant circularity: the taxonomy is constructed from external frameworks and prior author work appears only as background, not as load-bearing derivation.
full rationale
No circularity found. This paper is a conceptual taxonomy rather than an empirical derivation: the three modes (Support, Synergy, Symbiosis) are constructed by adopting external frameworks — Simmler and Frischknecht's five levels of technical autonomy, Rammert's graded agency levels, and Kaufman and Beghetto's Four-C creativity model — and then combining them in a table. The modes are not used to define those input frameworks, and the creativity-impact mapping in Table 1 is a stipulated interpretive proposal, not a prediction recovered from fitted data. There are no fitted parameters anywhere in the paper, and no quantity is renamed as a result after being fit to that same quantity. The authors' own prior work appears only as background context (e.g., Gaggioli et al. 2013 on creative networks; Vinchon et al. 2023 on generative AI), and it is not invoked as the justification for the taxonomy's central claims. The term 'Extended Creativity' is introduced descriptively, as the object of study, and does not circularly define the relational modes. The cited frameworks are external, independently published sources rather than author-derived uniqueness theorems. The skeptical observation that the autonomy/agency axes admit off-diagonal configurations not named by the taxonomy is a completeness or construct-validity concern about the framework, not a circularity concern, because the taxonomy does not claim to derive its mode labels from those axes by a formal equivalence. Accordingly, the derivation chain is self-contained with respect to the paper's stated aims, and the appropriate score is 0.
Assumptions & free parameters
assumptions (5)
- domain assumption The Standard Definition of Creativity (novel + appropriate) is a valid basis for defining creative products.
- domain assumption Kaufman and Beghetto's Four-C model adequately captures the range of creative achievement from mini-c to big-C and applies to hybrid human-AI systems.
- domain assumption Simmler and Frischknecht's five-level autonomy taxonomy is valid for contemporary AI systems.
- domain assumption Rammert's graded agency model (causal, contingent, intentional) applies to AI systems.
- domain assumption Distributed creativity (Glaveanu, 2014) is an appropriate foundation for understanding human-AI creative relations.
invented entities (1)
-
Extended Creativity system
Cite this review
Pith. "Pith review of Extended Creativity: A Conceptual Framework for Understanding Human-AI Creative Relations." pith.science (2026). https://pith.science/paper/XUSAOCUL
@misc{pith2026250610249,
author = {Pith},
title = {Pith review of: Extended Creativity: A Conceptual Framework for Understanding Human-AI Creative Relations},
year = {2026},
howpublished = {\url{https://pith.science/paper/XUSAOCUL}},
note = {Machine review of arXiv:2506.10249}
}
read the original abstract
Artificial Intelligence holds significant potential to enhance human creativity. However, achieving this vision requires a clearer understanding of how such enhancement can be effectively realized. Drawing on a relational and distributed cognition perspective, we identify three fundamental modes by which AI can support and shape creative processes: Support, where AI acts as a tool; Synergy, where AI and humans collaborate in complementary ways; and Symbiosis, where human and AI cognition become so integrated that they form a unified creative system. These modes are defined along two key dimensions: the level of technical autonomy exhibited by the AI system (i.e., its ability to operate independently and make decisions without human intervention), and the degree of perceived agency attributed to it (i.e., the extent to which the AI is experienced as an intentional or creative partner). We examine how each configuration influences different levels of creativity from everyday problem solving to paradigm shifting innovation and discuss the implications for ethics, research, and the design of future human AI creative systems.
Reference graph
Works this paper leans on
-
[1]
Support: AI enhances technical efficiency while preserving human creative control. 14 Technology The technological foundation of the Support mode consists primarily of specialized automation systems built on rule-based algorithms and task-specific computational processes. These applications are engineered to streamline creative workflows by executing well...
work page 2015
-
[2]
Synergy: collaborative creative dialogue between human and AI. 17 Technology The technological foundation of the Synergy model encompasses interactive systems designed for adaptive engagement with human creators. These include advanced language models like GPT-4, DALL-E, and Midjourney, which can generate contextually relevant creative content in response...
work page 2008
-
[3]
Extended Creativity: conceptual foundations 3.1 Defining creativity Creativity is a multifaceted concept that has intrigued scholars across disciplines, yet it remains challenging to define with precision. Encompassing activities from routine problem-solving to revolutionary innovations, creativity manifests in personal, social, and historical contexts. O...
work page 1953
-
[4]
Examples of AI-human relational modes across Extended Creativity domains. 8 Conclusions Examining the evolving relationship between AI and creativity reveals a key insight: designing AI systems based solely on traditional models of human creativity risks constraining their potential to explore new forms of innovation and expression—forms that may not alig...
-
[5]
By mapping different configurations of human–AI interaction along these axes, we aim to clarify the varying degrees and forms of creative involvement that characterize these systems. 4.1 Autonomy 10 Autonomy can be broadly defined as a system's capacity for self-governance: the ability to act according to internal mechanisms, goals, or principles, without...
work page 2020
-
[6]
A taxonomy of Extended Creativity systems Having established the foundational conceptual dimensions, we now propose a three-tier taxonomy of human–AI creative relationships: Support, Synergy, and Symbiosis. These relational modes are not meant as rigid categories, but rather as flexible patterns of interaction that evolve alongside the development of Exte...
work page 2007
-
[9]
Symbiosis: unified human-AI creative agency. Technology The technological foundation enabling the Symbiosis model represents the convergence of several cutting-edge domains: advanced neural interfaces, real-time adaptive systems, and generative AI architectures. At the input level, high-resolution BCI capture neural signatures associated with creative ide...
work page 2025
-
[10]
of Simmler and Frischknecht's taxonomy. They demonstrate not only contextual adaptivity and output indeterminacy but also the capacity to recursively reconfigure their own parameters based on evolving interaction patterns. These systems integrate real-time (neuro)feedback, continuous model updates, and advanced Bayesian mechanisms that anticipate not just...
work page 2004
Show all 14 references
-
[11]
Taxonomy of Extended Creativity systems. Dimension Support Synergy Symbiosis Example AI-powered image editing tools automating technical adjustments while human maintains aesthetic control Writer collaborating with language model to explore narrative possibilities through iter...
2010
-
[77]
Neither are we forced to claim that the activities of humans, machines and programs are substantially the same kind of behaviour
explains: "This gradual and multi-level model of agency gives us the possibility to escape the dilemma of having to either reserve agency up to the humans or to flatten the concept of agency unnecessarily. Neither are we forced to claim that the activities of humans, machines ...
2008
-
[2009]
System 0
established a distinction between big-C creativity—historically significant achievements that transform fields—and little-c creativity, seen in everyday problem-solving and personal expression. Kaufman and Beghetto expanded this binary model by adding mini-c (personally meanin...
2015
-
[2019]
They characterize it as a gradual property, emerging from networks of interdependence involving both human and non-human entities
offer a compelling framework to conceptualize agency. They characterize it as a gradual property, emerging from networks of interdependence involving both human and non-human entities. Within these networks, complex activities can be broken down into subtasks or operations per...
2008
-
[2021]
Rhodes, M. (1961). An analysis of creativity. The Phi Delta Kappan, 42(7), 305–310. Runco, M. A., & Jaeger, G. J. (2012). The standard definition of creativity. Creativity Research Journal, 24(1), 92–96. Schmutz, J. B., Outland, N., Kerstan, S., Georganta, E., & Ulfert, A. S. ...
1961 arXiv
-
[2024]
(Article No. 252). ACM. https://doi.org/10.1145/3613905.3650929 Sternberg, R. J., & Lubart, T. I. (1995). Defying the crowd: Cultivating creativity in a culture of conformity. Free Press. Lubart, T. I. (2001). Models of the creative process: Past, present and future. Creativit...
1995
Reviewed August 7, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.