Still searching for an (un)stable equilibrium: visualising the process of training generative neural networks without data

AI systems can fail in aesthetically interesting ways: The paper demonstrates that training generative neural networks without data produces visualizations of systems "caught in an endless, unresolvable search for equilibrium," revealing that AI doesn't always converge to meaningful solutions—a failure mode that becomes conceptually visible through artistic visualization.

Art can communicate AI concepts beyond technical jargon: The work explicitly aims to guide viewers toward understanding AI "through aesthetic experience rather than technical exposition," suggesting that artistic practices may help non-experts grasp how AI systems actually behave without requiring mathematical or programming knowledge.

Critical perspectives on AI need alternative explanations: The paper positions this artistic practice as "an alternative and critical mode of explainability for AI," implying that traditional technical explanations may be insufficient for developing a comprehensive, critical understanding of AI systems and their limitations.

Visualizing AI processes reveals conceptual truths: By making the training process visible through video, the work aims to provide "a conceptual understanding of what AI could be, rather than a restatement of what it currently is"—suggesting that how we represent AI systems shapes how we understand their potential and constraints.