Melbourne musicians treat AI's mistakes as raw creative material

The tool moment

An article titled “Emerging paradigms in music technology: valuing mistakes, glitches and uncertainty in the age of generative AI and automation” was published 12 February 2025 in AI & Society. The author, Miguel Loor Paredes of Monash University, conducted a multi-sited ethnography of music technologists in Melbourne, Australia, recruiting 24 participants through a snowballing technique across researchers, developers, and professional and aspiring artists. The published analysis focuses on ten of those participants, drawn from semi-structured interviews, conducted one to three times with most, and observation of their studios, offices and creative routines.

What the documents show

The paper states that “most participants valued the element of creative uncertainty introduced by unanticipated results and events in their practice,” treating mistakes and glitches as a resource for conveying emotion and making an artistic statement, in the author's words standing “in opposition to the controlled processes within generative AI models, based on the management of uncertainty as an input value.” The article also reports that opinion was not uniform: participants held varied and sometimes skeptical views of AI, with “not a single narrative or interpretation among the participants regarding AI.” The journal's own listing confirms this is peer-reviewed, open-access research, not an industry commentary.

What stays with the musician

The finding describes a specific creative community's stated values, not a universal aesthetic rule for how error should be treated in every genre or practice. A musician deciding whether to keep an unplanned glitch, a mistimed hit, a filter artifact, in a finished track is still making a judgment call the paper does not make for them: whether that moment reads to their own audience as expressive or as a flaw to fix. The study documents that Melbourne's music technologists lean toward keeping such moments; it does not establish that every listener or every genre shares that preference.

Judge it by listening

There is no listening test in this ethnography; its evidence is interview and observational, about what practitioners say and do, not about what a general audience perceives when they hear the result. A reader applying its argument to their own work can still ask, by ear, whether an unplanned artifact adds tension or texture that serves the track, or simply reads as a technical error, which is closer to the editorial judgment the paper describes its participants making than to any measurable standard.

• Does this glitch or artifact serve the track's emotional arc, or does it just sound like an error to an unfamiliar listener?
• Would a different genre or audience treat the same moment as a mistake rather than a creative choice?
• Is the appeal of an unplanned result about the sound itself, or about the story of how it happened?

A ten-participant ethnography in one city is a documented account of a specific creative culture, not a general law about audio quality. Treating its finding as evidence for one scene's values, rather than for all musicians, is the distinction the paper itself draws.