In my previous article, I posed nine questions to music industry leadership about AI and music. The first question was: What does "AI music" mean?
AI Music Is Not One Thing
Who remembers the early days of "cloud computing"? At the outset, there were many confused conversations about "the cloud," largely because people used the same term to describe very different technologies and services. This disconnect only became apparent when experts[1] took the trouble to develop an early taxonomy of different types of "cloud computing," including Software as a Service (SaaS), Platform as a Service (PaaS), and Infrastructure as a Service (IaaS), allowing clearer language for discussion, cooperation, and agreement.
Similarly, the music technology ecosystem now needs a shared taxonomy to ensure we are talking about the same thing when we say "AI music." A common vocabulary will be an important step to building mutual understanding and shared trust.
"AI music" currently collapses a diverse ecosystem into a single label: an ecosystem that covers production tools, full-generation platforms, fan products, sync tools, video soundtracks, voice replicas, sound-alike systems, discovery systems, coding assistants, and metadata tools. These AI-assisted and AI-enhanced tools should not be treated as a single moral, legal, or product category.
A stem separator, a mastering assistant, a lyric brainstorming tool, a licensed video-music generator, and a clone of a living performer's voice all involve AI, but they each give rise to very different rights, permissions, and risk profiles.
The same problem exists with the phrase "generative AI," which often collapses use cases that should be governed differently. A musician who rejects fully automated song generation may still value a tool that cleans unwanted noise from rehearsal audio. A platform that accepts AI-assisted post-production may still reject unauthorized vocal impersonation. A publisher might license training for one category of outputs while prohibiting artist-style prompting or consumer-facing derivatives.
Without precise language, policy becomes blunt, product design becomes defensive, and creators are asked to consent to categories they may not fully understand.
A Functional Taxonomy for "AI Music"
A useful first distinction is between assistance, generation, substitution, simulation, and infrastructure:
- Assistive tools strengthen an identifiable human process: editing, restoration, mixing, accessibility, search, translation, transcription, sound design, or ideation.
- Generative tools create new musical material, but even this category ranges from a drum fill to a complete track.
- Substitution tools seek to replace an existing market function, such as background library music or low-budget production.
- Simulation tools imitate identifiable voices, styles, catalogs, or performers.
- Infrastructure tools support the entire ecosystem—they do not necessarily create music at all but detect, disclose, label, track, identify, license, or pay for it.
While these categories will not resolve every difficult question, nor are they always mutually exclusive, they provide a more effective starting point to discuss governance.
Rights and Disclosures Should Follow Function
This taxonomy matters because rights and disclosures should follow function. For example, copyright questions turn heavily on the degree of human authorship, selection, arrangement, and modification. The U.S. Copyright Office has emphasized that copyright can protect human-authored expression in works that include AI material, but not purely AI-generated material standing alone.[2]
That legal distinction is also a product distinction. Tools should help users preserve evidence of human contribution where that matters for authorship, credit, and professional credibility.
The same principle applies to transparency. A binary "AI" or "Not AI" label is too crude for music because music is layered. AI may have been used for ideation, lyrics, composition, vocals, instrumentation, mixing, mastering, artwork, translation, workflow management, or metadata. A more useful disclosure system would describe how AI was involved rather than merely announcing its involvement. Supporting industry-standard AI disclosures reflects a more granular crediting of AI involvement in the creative process, without turning a generic "AI" label into a warning sign or stigma.
The deeper principle is interoperability. A disclosure made when a piece of music is created should accompany that piece through labels, distributors, DSPs, rights societies, analytics dashboards, and listener-facing credits without being re-entered or reinterpreted at each step.
An Opportunity for Collaboration
For the Music Technology Coalition, the taxonomy problem should become a convening problem. The Coalition can serve as a neutral forum where creators, technologists, labels, publishers, distributors, educators, and platforms agree on shared use-case language. That language could support product reviews, creator education, model contract terms, metadata schemas, disclosure prompts, and research benchmarks. The objective should not be to freeze innovation, but to prevent a vague vocabulary from becoming a substitute for governance.
A mature AI music ecosystem will not ask whether AI is good or bad in the abstract. It will ask what the tool does, whose work it relies on, what rights are implicated, how outputs are used, what information travels with the work, and what choices remain with the human creators and listeners involved.
The first act of trust is naming the thing accurately.
[1] Specifically, the US National Institute for Standards and Technology, NIST
[2] U.S. Copyright Office Copyright and Artificial Intelligence, Part 2: Copyrightability