Creativity in AI as Emergence from Domain-Limited Generative Models

TL;DR

The paper reframes AI creativity as an emergent property of domain-limited generative models embedded in bounded informational environments, moving beyond posthoc evaluation toward explicit computational modeling. It introduces a four-component decomposition—Patternism, Weltanschauung, Zeitgeist, and Arbitrarity—and a mathematical formalization, including the combined equation $C_{I,S}(t) = \alpha\, W_I(t) + \beta\, P_I(t) + \gamma\, Z_S + \varepsilon$, to describe how creative behavior can arise from structure, history, and contingent variation. A multimodal CGAN trained on an 18th-century European corpus demonstrates emergent patterns beyond direct corpus replication, driven by cross-modal constraints and historical grounding. The work discusses extensions to larger architectures and embodied systems, emphasizing a balance between constraints and exploratory arbitrarity for autonomous creative capacity. Overall, it offers a framework to study creativity as a computational, domain-bound phenomenon with practical implications for designing more adaptable, generalizable AI systems.

Abstract

Creativity in artificial intelligence is most often addressed through evaluative frameworks that aim to measure novelty, diversity, or usefulness in generated outputs. While such approaches have provided valuable insights into the behavior of modern generative models, they largely treat creativity as a property to be assessed rather than as a phenomenon to be explicitly modeled. In parallel, recent advances in large-scale generative systems, particularly multimodal architectures, have demonstrated increasingly sophisticated forms of pattern recombination, raising questions about the nature and limits of machine creativity. This paper proposes a generative perspective on creativity in AI, framing it as an emergent property of domain-limited generative models embedded within bounded informational environments. Rather than introducing new evaluative criteria, we focus on the structural and contextual conditions under which creative behaviors arise. We introduce a conceptual decomposition of creativity into four interacting components-pattern-based generation, induced world models, contextual grounding, and arbitrarity, and examine how these components manifest in multimodal generative systems. By grounding creativity in the interaction between generative dynamics and domain-specific representations, this work aims to provide a technical framework for studying creativity as an emergent phenomenon in AI systems, rather than as a post hoc evaluative label.

Figures (4)

• Figure 1: DCGAN (early epoch) - Representative outputs generated by the DCGAN during early training epochs. Generated images exhibit blurred reconstructions and low-frequency combinations of dominant corpus features, reflecting initial convergence toward dataset-level visual statistics.
• Figure 2: DCGAN (late epoch) - Representative outputs generated by the DCGAN at later training epochs within the same iteration. While visual coherence and textural sharpness improve, generated samples remain strongly aligned with the stylistic and compositional structures of the 18th-century corpus.
• Figure 3: DCGAN (final epoch, same iteration - DCGAN outputs at the final epochs of a single training iteration. Despite increased local consistency, the generative behavior remains conservative, with outputs largely confined to interpolations within the learned visual manifold.)
• Figure 4: CGAN (emergent samples) - Batch of samples generated by the multimodal CGAN across different training iterations. Outputs exhibit significant formal divergence from the original corpus, including abstract configurations and novel structural patterns, illustrating the impact of cross-modal text–image conditioning on generative dynamics.