EVALUATING ORIGINALITY IN AI-GENERATED CONTEMPORARY WORKS

Abstract [English]

The fast development of the generative artificial intelligence has considerably altered the modern artistic operations, posing the essential concerns about the matter of originality, authorship, and artistic worth of the AI-generated products. Although AI systems can create visually attractive and stylistically varied results, it is a more pressing problem how to judge whether these were original creations or just recombinations of acquired information. This analysis suggests a holistic analysis of originality in AI modern art by joining the computational evaluation with human analysis. The study constructs originality as a multidimensional phenomenon that covers novelty, non-traditionality, intention to create something new, and relevance to its contexts in terms of cultural and historical reference space. The proposed framework is based on the theories of computational creativity and human-AI co-creativity, but it also considers the shared authorship models where originality is created through the interaction between artists, datasets, algorithms, and curatorial choices. The originality assessment model based on AI is presented and is a combination of visual, semantic, stylistic, and contextual feature extraction with embedding-based similarity and divergence analysis. The quantitative measure of originality in terms of novelty scores, stylistic distance measures and entropy based diversity measures are used to represent the structural and statistical aspects of originality. These calculation tests are then complemented by qualitative tests such as the art experts, curators and audience perception studies in order to cover the subjective and interpretive aspects which most automated programs fail to cover. A comparative study of AI-based evaluation and traditional originality assessment methods shows the advantages and the constraints of the former.