Melbourne producers value glitches AI systems are built to remove

The audio task

A producer deciding whether to let a generative tool “clean up” an unexpected sound, or a plugin designer deciding how much randomness to leave in a patch, is making a choice about whether imperfection is a bug to remove or a resource to keep.

What the documents show

An article in AI & Society, published online 12 February 2025 by Miguel Loor Paredes, reports a multi-sited ethnography of Melbourne's music-technology scene conducted between July 2022 and November 2023, covering home studios, co-working spaces, venues, research labs, and educational settings, with roughly ten music technologists studied in depth among a broader group of 24 participants across research, sound engineering, songwriting, production, and education. The article documents practitioners describing “happy accidents,” unexpected sounds from experimentation that became the most interesting part of a recording, and one artist deliberately choosing an imperfect AI voice-synthesis result because its inaccuracy made it “more interesting.” Participants distinguished technical perfection from what one musician called mistakes that “carry emotion,” and the author reports this sits in tension with generative AI systems built to minimise unpredictability and maximise control. The article also documents participants prioritising direct human collaboration over AI-mediated workflows for conveying artistic intention, while still using AI tools alongside that preference. The author is explicit that “there was not a single narrative” among the people studied.

Rights status

This is a qualitative, ethnographic study, confirmed by its own DOI record as a peer-reviewed AI & Society article rather than a preprint; it makes no copyright, licensing, or consent determination, and states no legal position on AI-generated music. Its contribution is descriptive: an account of how one scene's practitioners talk about and use unpredictability, not a rights or compliance document.

What to check before you use it

This is an editorial checklist. The fieldwork covers one city's scene over a specific seventeen-month window that predates many 2025 and 2026 tool releases, so treat its account as a snapshot of that period and place rather than a current global survey. The detailed sample is small, roughly ten practitioners examined closely, which supports a cultural account but not a statistical claim about how common any given attitude is. Because the author reports contradictory views within the same group, avoid quoting a single participant's line as if it represented a scene-wide consensus.

• Is a quoted attitude toward glitches being treated as one practitioner's view or the scene's consensus?
• Does the 2022 to 2023 fieldwork window predate the specific tool or feature under discussion?
• Is “AI can't replicate this” being read as the author's finding, or as one participant's stated belief?

The study's use for a production decision is more atmospheric than prescriptive: it documents that a meaningful slice of working music technologists treat imperfection as expressive material, a cultural data point worth weighing against, not replacing, a plugin's own stated capabilities.