Desire and Data: Metabolism of AI

Abstract Generative artificial intelligence (AI) is reshaping audio-visual media, yet its long-term creative sustainability remains uncertain. This article proposes a metabolic framework to explain how AI systems consume, transform, and renew cultural material, and how these cycles depend on human creativity. Drawing on cybernetics, general systems theory, and autopoiesis, the study situates AI within a coupled human-machine ecosystem. Human creativity is both a fuel that sustains AI and the antidote to its stagnation, providing unpredictable variation that prevents recursive degradation while remaining vulnerable to commodification that limits regeneration. A qualitative, practice-based approach combines conceptual analysis with case studies of independent and marginal AI art/research. Four thematic dynamics are examined: (1) aesthetic convergence, with synthetic folklore as algorithm-native form; (2) creative entropy, including Model Autophagy Disorder (MAD), with desire as a counterforce to homogenisation; (3) creative metabolism, a cycle of cultural input and transformation addressing erosion of desire under algorithmic optimisation; and (4) creative countercurrent practices resisting corporate logics and highlighting marginality’s role in sustaining diversity. Case studies – from CLIP-VQGAN and Aphantasia to ConceptLab, Active Divergence, and Roope Rainisto’s post-photography – show how low-resource, critical engagement with AI fosters cultural variation. Findings suggest that supporting marginal creative practices is essential not only for aesthetic diversity but for the viability of AI-mediated creative ecosystems. Future research considers emerging agentic AI systems and autonomous artists.