Nick Bryan-Kinns

Executive Summary

Nick Bryan-Kinns is Professor of Creative Computing at UAL's Creative Computing Institute, where his research focuses on human-centred and explainable AI for the arts. He is also a practising bass player, and the two roles are not separate. His practice informs his research, and his research sharpens his practice. The result is a perspective that is simultaneously inside the technology and critically outside it.

His response to AI is surgical differentiation. He embraces experimental tools — Combobulator for glitchy textures, Rave for surreal interpolations, custom variational autoencoders for folk melody exploration — while actively rejecting tools like Suno and Udio that generate finished songs and eliminate human agency. He is more interested in the ideation stage than the polished stage, more interested in provocation than production. His practice is process-driven; outcome and publication are beside the point.

As a researcher, he commissions artists to work with small-dataset models, teaches students to critique AI's IP exploitation and cultural flattening, and has developed a precise vision for what genuine AI collaboration would actually require. None of today's tools come close.

The Academic Lens: Human-Centred AI for the Arts

Nick's research agenda is built around a core conviction: that AI tools for creative practice must be designed around the artist's process, not around the technology's capabilities. His work on explainable AI for the arts asks not just whether a system produces interesting outputs, but whether the artist can understand, interrogate, and meaningfully direct what the system is doing.

This shapes how he evaluates every tool he encounters. The question is never "what can this generate?" but "where does the human remain in control, and where does the system take over without explanation?" Most commercial AI tools fail this test. They optimise for impressive outputs at the expense of legible process, which is precisely what makes them useful for demonstration and problematic for practice.

His commissioned research reinforces this. Working with artists on small-dataset models, he observed that the tools which worked best offered both surprise and a sense of control.

What I learned from commissioning artists is that AI systems need both opportunity for surprise and a sense of control. Without both, it's not creative partnership.

Surgical AI Integration: What He Embraces and What He Rejects

When Suno first appeared, Nick was excited by the prospect of having a band again. The reality was alienating. The tool generated a complete finished song with no room for his input, producing something professionally bland but creatively empty. That wasn't collaboration; it was displacement. He now uses these tools only pedagogically, to demonstrate AI's problems to students.

He shows equal disinterest in AI mastering or automated mixing. His aesthetic favours rough edges, glitchy textures, jarring transitions, the sounds that reveal human intention. The polished stage already has plenty of tools; what interests him is the ideation stage, where the work is still alive and unresolved.

His actual toolkit is more experimental:

- **Combobulator** transforms old band recordings into glitchy, unrecognisable versions shared with ex-bandmates as provocations, not finished works but conversation starters.

- **Rave** creates surreal interpolations between tracks. "The outputs are totally surprising madness. No idea what's coming, but you find these interesting points that sound fun and different."

- **Natural Drums AI** generates rhythms no preset can produce, syncopations more unusual than any standard jazz pattern.

- **Custom variational autoencoder** trained on Irish folk tunes generates melodic fragments for bass lines, with small consensual datasets ensuring the AI speaks a language he curates.

- **ChatGPT** serves as music theory tutor. He writes a chord sequence, asks what a jarring key change might be, then works the suggestion into a song.

I take what it gives me and work it into a song. There's back-and-forth, but I'm learning, not copying.

Provocation as Collaboration

Nick's sharing model redefines what collaboration means. He sends AI-transformed sketches of old band recordings to musician friends as provocations.

'Hey, listen to this crazy thing I made with AI, which songs do you think this came from?' It's a game. The transformation is so extreme they can't tell. That prompts them to send something back, and we have a creative conversation.

Distribution is never about audience building; it's about stimulus-response loops that fuel his own creative cycle. His live performance interest follows similar logic. Working with UAL colleagues who live-code electronic music, he is exploring how his bass playing can be triggered in unusual ways by AI, not to make performance easier, but to recreate the felt sense of shared anticipation that defines playing with other humans.

What AI Still Can't Do: Memory and Shared Anticipation

Nick's vision for ideal AI collaboration is precise. He doesn't ask for better generation; he asks for dialogue, memory, and synchronicity.

The problem, as he sees it, is that the exchange is one-directional. The AI generates, he responds, but when he generates again, the system has no awareness of what he made or where he is taking the work.

The problem is it's one-way. It generates, I respond, but when I press generate again, does it notice what I made? I want AI that remembers this is version 10 and responds to the direction I'm taking.

What he values most about playing with humans is the felt sense of being in sync, the ability to anticipate a key change before it happens and respond to it in real time.

What I love about human collaboration is shared anticipation — feeling when someone's about to change key. You'd need much more interactive dialogue for AI to achieve that. It would need to listen and remember.

Current tools, even those integrated directly into digital audio workstations, remain firmly in the tool category. They either generate ideas you work with, or make what you are playing sound more professional. Neither constitutes collaboration.

Authorship and Ethical Boundaries

Nick's authorship stance is unusually clear, shaped in part by his professional involvement in UK copyright policy. When Rave interpolates between two tracks to create something new, he asserts full ownership: Rave is not a legal entity, and the idea and execution are his.

His ethical boundaries follow directly from his research position. He refuses Suno and Udio for anything beyond teaching because the ownership of training material is unresolved. For commissioned work, he trains models on small, consensual datasets to avoid appropriation.

My concern is scraping content without permission. If creators used my material to train models they profit from, I get no recognition. That's exploitation.

This is not just personal ethics; it is a research principle. Ethical AI use requires curated, consensual data. That position runs through his practice, his teaching, and his vision for where the field needs to go.