Extended Creativity Frameworks

• Extended Creativity Frameworks are theoretical and operational models that integrate multi-level, distributed creative processes across individual, social, and human–AI systems.
• They incorporate modular agent-based architectures and multi-layer abstractions to systematically benchmark and enhance creative outputs in research, design, and the arts.
• These frameworks offer quantitative evaluation metrics and protocol-driven methodologies, enabling robust, scalable, and ethically mindful collective creativity.

Extended creativity frameworks encapsulate a family of theoretical, formal, and operational models for describing, supporting, and analyzing creativity across individual, group, human–AI, and machine-agent systems. These frameworks generalize beyond classical, individual-centered or output-evaluative views, introducing mechanisms for abstraction, social interaction, distributed agency, and structured multi-level search. They provide technical infrastructure for understanding, benchmarking, and enabling machine-augmented and collective creative processes in research, design, and the arts.

1. Theoretical and Taxonomic Foundations

The central concept of extended creativity is rooted in the systematic expansion and operationalization of foundational theories, especially Margaret Boden’s taxonomy (combinational, exploratory, transformational creativity), Rhodes’ 4Ps (Person, Process, Press, Product), and distributed/relational cognition.

Key theoretical elements are:

• Multi-level creativity: Combinational (recombining known elements), exploratory (systematic search within a conceptual space), and transformational (altering the underlying conceptual space) (Gu et al., 2024, Shahhosseini et al., 5 Nov 2025).
• Abstraction gradients: Explicit representation of ideas and creative artifacts at multiple generalization levels, from domain-specific implementations to universal principles (Gu et al., 2024).
• Distributed and relational agency: Creativity is modeled not as an attribute of isolated agents but as an emergent property of hybrid networks—human, artificial, and artifactual—interacting dynamically (Gaggioli et al., 12 Jun 2025, Ahmed et al., 3 Feb 2026).
• Technical autonomy and agency axes: Mapping system roles from low-autonomy, tool-like support (human-in-control), through mixed-initiative synergy, to high-autonomy, symbiotic co-creation, as formalized via operational axes (Gaggioli et al., 12 Jun 2025, Haase et al., 2024).

Extended creativity frameworks thus bridge gaps between descriptive theory and executable practice—embedding creative dynamics into computational, agent-based, or protocol-driven models that can be studied, compared, and evolved systematically.

2. Formal Models and Architectures

Extended creativity frameworks operationalize these principles via precise formalisms and modular system architectures:

• Agent-based structures: Systems are decomposed into specialized agents—retrieval, combinatorial, critic, refiner, and director agents coordinate brainstorming, decomposition/recombination, assessment, and iterative refinement (Gu et al., 2024, Venkatesh et al., 7 Apr 2025).
• Abstraction layers and fields: Innovations and ideas are indexed and recombined along multiple generalization levels—Implementation (L1), Mechanism (L2), Rationale (L3), Universal Principle (L4)—enabling cross-granularity analogies (Gu et al., 2024).
• Socio-cognitive networks: Collective creativity is modeled by agents with tunable semantic networks (modularity parameter $p$ controls ideational breadth), simulating real-world variability in creative potential and networked stimulation effects (Ahmed et al., 3 Feb 2026).
• Interaction protocols: Prescriptive multi-step procedures (e.g., GHOSTY COLLIDER, PRECOG PROTOCOL) explicitly translate theories like bisociation/lateral thinking into action sequences with quality gates, anti-pattern checks, and measurable outcomes (Fujiyoshi, 6 Mar 2026).
• Turn-based and flexible workflow systems: Human–AI creative sessions are structured as alternating or free-form cycles of artifact generation, critique, and non-turn background learning, enabling nuanced agency assignment and learning from interaction (Guzdial et al., 2019, Yang et al., 15 Sep 2025).

These formalizations support both empirical evaluation and practical system design, defining measurable state spaces and pathways for extending, constraining, or transforming creative processes.

3. Evaluation Metrics and Quantitative Assessment

Extended creativity frameworks specify quantitative evaluation protocols that move beyond surface-level novelty/diversity:

Metric/Axis	Description	Framework(s)
Semantic Similarity	Embedding-based cosine similarity for idea fields (problem structure, mechanisms, rationale, principle)	(Gu et al., 2024)
U-O-S (Usefulness-Originality-Surprise)	Triadic scoring inspired by social science and psychology for both convergent and divergent tasks	(He et al., 5 Oct 2025)
Breadth & Modularity	Degree of ideational diversity and network modularity (e.g., via semantic random walks)	(Ahmed et al., 3 Feb 2026)
TTCT-derived	Fluency, flexibility, originality, elaboration (adapted for LLMs and VLMs)	(Chang et al., 2024, Lu et al., 2024)
Retrieval & recombination ablations	Incremental gains via multi-level retrieval and agentic recombination pipelines	(Gu et al., 2024)
Protocol outcome rates	Electric Rate (fraction of high-quality cross-domain collisions), parameter specificity	(Fujiyoshi, 6 Mar 2026)

Evaluation typically involves domain-aligned similarity metrics, LLM- or human-based rubric scoring, and rigorous ablation studies. For example, agent-based combinatorial frameworks show 7–13% improvements in field alignment scores over baselines, while agentic image-editing loops achieve substantial gains in user-rated originality and diversity (Gu et al., 2024, Venkatesh et al., 7 Apr 2025).

4. Systemic Extensions and Emergence

The major contribution of extended creativity frameworks is their capacity to model and scaffold creativity as a systemic, emergent phenomenon:

• Multi-scale modeling: Neural, conceptual, social, and cultural levels are integrated, using mathematical models ranging from associative memory and radial-basis activation to quantum-like Hilbert space representations of concepts and entanglement-based blending (Gabora, 2018, Peng et al., 17 Apr 2025).
• Emergence in bounded domains: Generative creativity is framed as an outcome of dynamic interplay among pattern acquisition, internal world-models, contextual grounding, and stochastic arbitrarity—formulated as $$C_{(I,S)}(t) = \alpha W_I(t) + \beta P_I(t) + \gamma Z_S + \epsilon$$ (Chutaux, 13 Jan 2026).
• Societal and network phenomena: Agent-based models reproduce empirical findings such as diversity-induced stimulation (lower initial overlap yields greater ideational gain), redundancy effects from shared inspiration, and the formation of scale-free creative influence networks (Ahmed et al., 3 Feb 2026, Velardo et al., 2016).

These systemic layers provide both predictive and generative power for cultural evolution, organizational creativity, and large-scale hybrid creative ecosystems.

5. Application Domains and Methodological Best Practices

Extended creativity frameworks are deployed in a variety of practical scenarios:

• Scientific ideation: LLM-based frameworks systematically generalize and recombine ideas across abstraction levels, outperforming direct generation in alignment with published research (7–10% similarity uplift) and enabling automated hypothesis generation (Gu et al., 2024, Shahhosseini et al., 5 Nov 2025).
• Design and brainstorming: Role-structured, prompt-engineered, and strategy-driven protocols (e.g., GPS, multi-agent LLM discussion) achieve higher originality, flexibility, and fluency relative to unstructured or single-agent settings (Chang et al., 2024, Lu et al., 2024).
• Visual and artistic creativity: Multi-dimensional trait modeling (e.g., TraitSpaces 12-trait taxonomy), agentic collaboration frameworks (CREA), and conditionally interpretable assessment tools support richer human–AI co-creation and interpretable evaluation (Luthra, 29 Sep 2025, Venkatesh et al., 7 Apr 2025, Lin et al., 17 Nov 2025).
• Strategic foresight and innovation: Protocol-driven frameworks operationalize weak-signal horizon scanning and innovation emergence via explicit steps, quality gates, and anti-pattern checks, bridging theory and practice (Fujiyoshi, 6 Mar 2026).

Best practices include modular decomposition (abstraction levels, specialized agents), rigorous quantitative evaluation (embedding-based, rubric-aligned), iterative agentic loops (conceptualization, generation, critique, refinement), and explicit mapping to theoretical creativity axes.

6. Limitations, Challenges, and Open Directions

Despite advances, substantial limitations persist:

• Transformational creativity remains rare: Most current systems excel at recombinational and exploratory modes; dynamic alteration of the underlying conceptual space is only partially realized, requiring further research into meta-level and emergent agent interactions (Gu et al., 2024, Shahhosseini et al., 5 Nov 2025).
• Evaluation bottlenecks: Human alignment, domain adaptation, and scalability of expert scoring lag behind algorithmic progress; there is a need for robust, cross-domain, automated evaluators and for metrics that operationalize emergent creative phenomena without reducing them to single-point scores (He et al., 5 Oct 2025, Lin et al., 17 Nov 2025).
• Socio-technical challenges: Issues of authorship, credit, cognitive liberty, privacy, and creative agency grow as AI agents attain higher autonomy and integration with human cognition (Gaggioli et al., 12 Jun 2025, Haase et al., 2024).
• Non-anthropocentric creativity: Formal general creativity frameworks point toward possible forms of machine-native creativity, potentially unrecognizable to humans, raising open theoretical and practical questions about evaluation and societal integration (Velardo et al., 2016).

Future research priorities include developing multi-level, human-AI and agent–agent feedback architectures; advancing methods for operationalizing transformational creativity; refining societal-scale simulation models; and connecting protocol-driven theory to measurable, value-aligned creative outputs.

7. Synthesis and Future Landscape

Extended creativity frameworks rigorously extend classical creativity theory by embedding abstractions, modular agent architectures, distributed interaction models, and systemic emergence principles into creative AI and collective human–AI workflows. Their adoption enables formal intercomparability, systematic evaluation, and principled design of tools and protocols that move creativity research from descriptive taxonomies toward actionable, scalable, and collaboratively executable systems (Gu et al., 2024, Fujiyoshi, 6 Mar 2026, Gaggioli et al., 12 Jun 2025). The future of this field lies in further extending such frameworks to support open-ended discovery, robust co-creativity, domain transfer, and ethical deployment across scientific, design, and artistic domains.