# AI as Noise: Merzbow and the Aesthetics of Productive Failure
The new AI-generated visualisations accompanying Merzbow’s recent concerts appear, at first glance, to resolve a problem that has long surrounded noise music: how can sound that resists representation be made visible without being reduced to illustration? The answer proposed by these works is neither synchronisation nor spectacle. Instead, the visuals treat artificial intelligence as a system of unstable translation—a machine that receives sonic excess and returns visual excess.
This may be one of the most convincing places for AI within contemporary audiovisual practice.Artificial intelligence is not inherently suited to art or design. Its outputs are often too smooth, too coherent, too statistically plausible. Trained on enormous collections of existing images, AI tends to reproduce recognisable visual conventions even when prompted to generate novelty. In design, this can result in generic solutions: images without material responsibility, formal decisions without necessity, and aesthetics detached from authorship, context, or consequence. In art, the same mechanisms can turn complexity into style and ambiguity into decoration.Yet these limitations do not make AI artistically irrelevant. On the contrary, they may become productive when the system is used not as an autonomous creator but as a conceptual instrument. AI becomes compelling when its failures, biases, hallucinations, repetitions, and unstable associations are not concealed but incorporated into the work. Its value may lie less in producing finished images than in exposing the processes through which images are generated, classified, predicted, and made culturally legible.Merzbow’s visualisations operate precisely within this territory.Noise has always challenged the distinction between signal and error. It is not simply unwanted sound; it is the material that reveals the instability of communication itself. In Merzbow’s music, saturation, feedback, distortion, repetition, and density do not function as obstacles to meaning. They produce another form of meaning—one that emerges through intensity, duration, physical pressure, and perceptual overload.AI is structurally close to this understanding of noise. Its outputs are generated through probabilistic inference: patterns are predicted, transformed, and continuously negotiated within vast latent spaces. What appears as an image is the temporary stabilisation of innumerable possibilities. Beneath every coherent result lies a field of uncertainty, statistical variation, and computational excess.The AI visualisation therefore does not need to “represent” Merzbow’s music. It can behave like the music. It can become a visual noise process.During the concerts, the generated images appear to fluctuate between recognisable forms and abstraction. Figures emerge only to dissolve. Architectural structures become biological textures; surfaces mutate into impossible materials; fragments of landscapes collapse into unstable fields of colour and motion. The system seems unable to decide what it is seeing—or what it is becoming. This indeterminacy is not a technical weakness. It is the conceptual centre of the work.The visual field does not illustrate the sound. It undergoes pressure from it.This distinction is crucial. Conventional concert visuals often translate music into spectacle through rhythm, colour, or symbolic imagery. They provide a visual equivalent of the sound and therefore make the experience easier to consume. Merzbow’s AI visualisations move in the opposite direction. They preserve friction. The images do not clarify the music; they introduce another unstable layer of perception.The result is an audiovisual environment in which neither medium dominates. Sound destabilises the image, while the image alters the listener’s interpretation of sound. Perception becomes recursive. The audience is not positioned outside the work as a passive observer but becomes part of a feedback system involving acoustic intensity, machine-generated imagery, technological mediation, and embodied attention.In this sense, the project belongs to a broader history of AI-based art, although its conceptual position differs from many of the field’s most celebrated examples.Harold Cohen’s *AARON*, developed from the early 1970s onward, remains one of the foundational works of computational art. Rather than using AI as an image generator in the contemporary sense, Cohen constructed a rule-based system capable of producing drawings according to evolving formal principles. *AARON* raised questions about authorship and artistic agency: if a machine can generate images independently, where does the artist’s authorship reside? In the code, the rules, the selection process, or the system’s autonomous behaviour?More recent works have shifted attention from autonomy to datasets and cultural representation. Refik Anadol’s large-scale data sculptures transform archives and collections into immersive visual environments. His works demonstrate AI’s capacity to make information spatial and atmospheric. However, their aesthetic power often depends on a form of technological sublimity: vast datasets become flowing, luminous, monumental abstractions. The machine appears visionary.Trevor Paglen and Kate Crawford’s *ImageNet Roulette* takes a sharply critical position. The work exposes the violence of machine classification by allowing viewers to experience how an AI system categorises human faces. Its significance lies not in visual beauty but in institutional critique. AI is revealed as a political technology shaped by historical prejudices, labour systems, and taxonomic assumptions.Anna Ridler’s *Mosaic Virus* approaches AI through material labour and data construction. The artist created a dataset of thousands of hand-labelled tulip images and used machine learning to generate new floral forms. The work connects contemporary AI to the history of tulip speculation, demonstrating that datasets are neither neutral nor immaterial. They are constructed through selection, labour, and economic value.Sofia Crespo’s work investigates the relationship between artificial intelligence, biological form, and ecological imagination. Her generated organisms appear both familiar and impossible, occupying a space between scientific visualisation and speculative life. Here, AI becomes a tool for examining how humans imagine nature through technological systems.These works remain essential because they identify different conceptual uses of AI: autonomous procedure, data visualisation, institutional critique, dataset authorship, and speculative biology. Yet Merzbow’s visualisations suggest another possibility. They treat AI not primarily as intelligence but as noise.This shift may be more important than it initially appears.The dominant cultural narrative presents AI as a system that reduces uncertainty. It predicts, optimises, classifies, automates, and produces answers. Its commercial value is associated with efficiency and control. Noise, by contrast, is associated with interference, unpredictability, error, and excess.Merzbow reverses this relationship. AI is not used to eliminate noise but to generate it. The system becomes valuable precisely because it cannot maintain perfect semantic stability. Its tendency to produce visual anomalies, false connections, unstable morphologies, and unpredictable transitions is allowed to remain visible.The equation could therefore be expressed as:
**AI = noise**
Not because AI is meaningless, but because its meaning is probabilistic, unstable, and continuously reconstructed.Within this framework, the imperfections of AI become aesthetically relevant. A hallucination is not simply an error; it is an unexpected semantic event. A distorted form is not necessarily a failed representation; it may reveal the limits of the system’s visual categories. Repetition becomes rhythm. Dataset bias becomes material for critique. Instability becomes movement.This does not mean that every AI-generated image is art. Nor does it justify the uncritical use of generative systems in design. AI cannot replace artistic intention, historical knowledge, material sensitivity, or ethical responsibility. It cannot determine why a form should exist, what it means within a specific context, or how it affects the social conditions of its production.AI is not perfect for art. It is not perfect for design. It may not be perfect for anything.But perfection is rarely the relevant criterion.Art often emerges from constraints, contradictions, and failures. A tool becomes artistically significant when its technical logic can be transformed into a conceptual problem. In Merzbow’s visualisations, AI is necessary not because it produces superior images but because its instability corresponds to the aesthetic and philosophical structure of noise.The machine does not visualise the music as an external object. It enters into the same condition of excess.The concert becomes a temporary system in which sound, image, algorithm, and audience continuously affect one another. The visual output is neither illustration nor decoration. It is a metastable field: always approaching form, never fully arriving.Perhaps this is the strongest conceptual role AI can currently occupy. Not as an artificial artist, not as a replacement for human creativity, and not as a machine for producing seamless visual content—but as a technology capable of making uncertainty perceptible.Merzbow’s AI visualisations do not demonstrate what artificial intelligence can create.They demonstrate what happens when artificial intelligence is allowed to fail loudly.