MEASURING CREATIVITY IN CONTEMPORARY ART VIA AI MODELS

Abstract [English]

The process of measuring creativity in modern art has mostly been based on subjective expert opinion, cultural background and qualitative interpretations, which although they may be valuable, are usually not scalable, consistent and repeatable. As digital art practices and art datasets continue to expand at a very high rate, there is an increasing demand to have computational systems that can enable the systematic evaluation of the features of creativity without compromising artistic subtleties. In the present paper, the author suggests an AI-enabled creativity measurement framework which is a combination of computer vision, natural language processing, and multimodal learning to measure creativity in modern artworks. The framework conceptualizes creativity as a multidimensional construct that entails visual novelty, stylistic deviance, conceptual richness, narrative novelty and contextual topicality. Deep convolutional and transformer-based vision models are used to extract features of visual analysis that include color harmony, compositional complexity, variation of texture and deviation of style. The conceptual and semantic levels are represented by the use of NLP models on the texts of artists, exhibition, and critical descriptions, which allows analyzing originality, metaphor density, and coherence of the theme. Multimodal visionlanguage models also match a visual and textual representation to generate an overall creativity score, one that is holistic that captures the perceptual as well as interpretative elements of art. The suggested approach is compared to the baseline statistical and single-modality models in terms of the quantitative indicators of novelty indices, semantic divergence scores, cross-modal coherence, and accuracy in the classification of creativity.