What Musicians Are Actually Doing With AI (According to New Research)

I've been thinking about AI's potential in music since I wrote about it in 2023, and using these tools in my own practice, but I'll be honest, I wasn't sure if what I was experiencing matched what other musicians were doing. The marketing around these platforms is all hype and possibility, while the skeptical takes focus on replacement fears. What's actually happening in between?

Two research papers landed on my desk recently that finally gave me some real data. Not marketing claims or hot takes, but actual research into how artists are using AI music tools and what the technology landscape looks like right now. As someone navigating this space personally, these findings hit different than the usual discourse.

Here's what the research actually shows.

The first paper, "Artistic Trends in AI Music," dug into how musicians are actually incorporating AI into their creative practice. The findings surprised me, mostly because they validated some things I'd been feeling but couldn't quite articulate.

Artists aren't using AI to replace themselves. The research found a clear hierarchy in how musicians approach AI tools: Sound Design ranks highest, followed by Co-Composition, with full AI Composition at the bottom (though it’s rising in 2025). This mostly matches my experience, though I have explored that full AI composition realm with some really fascinating results. It actually allowed me to test drive my ideas in different styles, which I thought was pretty cool. But primarily, I use AI to generate interesting textures or explore sonic possibilities I wouldn't have thought of on my own.

New performance categories are emerging. The paper identifies "Prompt Jockeys" as a legitimate new type of performer: artists who use AI models live, creating music on the fly through real-time prompting. Think of it like DJing, but instead of mixing existing tracks, you're generating new material in the moment. Dadabots has been pioneering this approach with their 24/7 AI livestreams and live performances that react to human input.

AI is becoming an artistic medium, not just a tool. This is where things get really interesting. Some artists are using AI to create what the researchers call "generative AI as an artistic medium"—allowing others to create music using their vocal likenesses. Holly Herndon's Holly+ project lets anyone use an AI clone of her voice, while Grimes launched Elf.Tech with a transparent revenue-sharing model for AI-generated content using her voice. I highlighted both of these projects in my 2023 essay as examples of where things might be heading, and it's fascinating to see them now validated as significant trends in the research.

This creates new degrees of interaction between artists and audiences and transcends traditional creative boundaries. But it also raises fascinating questions about ownership and authenticity that we're still figuring out.

The sound of AI music is more nuanced than expected. While AI-generated music is often nearly indistinguishable from non-AI music, some artists are deliberately embracing AI's flaws and aesthetic quirks. They're exploring what the researchers call the uncanny—that familiar yet strange quality that AI sometimes produces. Though it's worth noting that this uncanny quality is becoming increasingly subtle as the technology improves. Early music AI was scratchy and low resolution, but nowadays we have high fidelity outputs that make these quirks much less obvious to the human ear.

This resonates with my own experience. Sometimes the most interesting moments come from AI's "mistakes" or unexpected combinations. Additionally, artists are curating custom datasets to train models that produce unique sounds (e.g. Neutone models), essentially using AI's pattern-matching abilities to create entirely new sonic territories.

Share

The second paper, "An Overview of Music AI and its Potential," maps out how dramatically the technology has changed. We've moved from AI that generates symbolic music (MIDI files, sheet music) to systems that create complete audio songs with vocals directly from text prompts.

Several platforms can now generate full songs in minutes that are difficult to distinguish from human-created music. This isn't incremental improvement, it's a categorical change in what's possible.

The research breaks down music AI into three categories:

Music Generation AI includes everything from audio generation (Suno, Udio, Eleven Music, etc.) to MIDI generation (more efficient for background music) to sample combination services like Mubert that assemble pre-recorded loops.

AI Production Tools covers the expanding world of AI-powered mixing, mastering, vocal synthesis, and neural audio effects that simulate guitar amps and other gear. This is where I see the most immediate practical value for working musicians.

Music Understanding AI encompasses recommendation systems, automatic transcription, and audio fingerprinting; the behind-the-scenes intelligence that powers platforms like Spotify and Shazam.

What strikes me about this framework is how it shows AI touching every part of the music creation and consumption pipeline, not just the "creative" parts we tend to focus on.

Both papers grapple seriously with the copyright and ethical issues that anyone using these tools needs to understand. The legal landscape is complex and evolving rapidly.

The major music labels have filed lawsuits against several AI music companies for training their models on copyrighted music without permission. But here's the twist: those same labels are now in licensing talks with some of these companies. Meanwhile, over 200 artists including Billie Eilish and Katy Perry have signed letters opposing AI training on copyrighted works without consent.

The second paper raises a crucial philosophical question: should current music generation AI be viewed as "replication technology" rather than creative tools? The authors argue it's more like record digging or Spotify than musical instruments: systems that recombine existing patterns rather than truly creating something new.

This hits at something I've been wrestling with personally. When I use AI tools, am I creating or am I curating? The answer probably depends on how I use them, but the question itself matters for how we think about authorship and creativity.

As someone navigating this space both as a musician and someone interested in technology, here's what I'm taking from this research:

The tools are powerful but the creative decisions still matter. AI can generate impressive results, but the research shows that artists are using it most effectively as a collaborator, not a replacement. The creative vision, curation, and artistic intent remain human.

Legal uncertainty creates real risks. If you're using these tools commercially, you need to understand the evolving legal landscape. The ethical AI companies that licensed music from the start are feeling disadvantaged by the major labels' licensing talks with companies that initially didn't ask permission.

Artist agency models are emerging. Holly Herndon and Grimes established early examples with their projects from a couple years ago, showing different approaches to maintaining creative control while embracing AI collaboration. These weren't just interesting experiments, they've become potential templates for how artists can navigate this technology on their own terms, and we're likely to see more experiments like this.

Quality vs. accessibility trade-offs are real. The research confirms what many users have experienced: these tools can produce impressive results, but they often need additional work to reach professional standards. Understanding these limitations helps set realistic expectations.

The research suggests we're at a pivotal moment where the technology has achieved remarkable capabilities but faces fundamental questions about its role in music culture and artist sustainability. The tools are here, they're powerful, and they're not going away. The question is how we choose to use them.

I'm curious about your experiences navigating AI tools in your creative practice. Are you seeing similar patterns in your work? Different approaches? The research gives us a foundation, but the real learning happens in the community of people actually using these tools thoughtfully.

What's your take on where this is all heading?

Pons, J., Zukowski, Z., Parker, J. D., Carr, C. J., Taylor, J., & Evans, Z. (2025). Music and artificial intelligence: Artistic trends. arXiv preprint arXiv:2508.11694. https://arxiv.org/abs/2508.11694

Masuda, N., Baker, T., & Tokui, N. (2025). An overview of music AI and its potential (2025 edition) [White paper]. Qosmo Inc. https://qosmo.jp/en/publication/musicai-whitepaper-2025