How Musicians Can Get Paid for Training AI
Summary
Warner Music Group's acquisition of Sureel and its partnership with STIM highlight emerging strategies for compensating musicians whose work trains generative AI. Sureel's software labels music files with owner-defined usage instructions, tracking AI companies' training data use to determine licensing fees. Concurrently, SoundVerse advocates for ongoing artist participation, proposing differential royalty payments based on how much specific training data influences a generated output. The core challenge involves accurately attributing causality between training data and AI output, rather than mere similarity, to prevent gaming the system. While complex, these attribution systems aim to sustain a vibrant creative sector, potentially encouraging musical experimentation and diversity. The industry is also seeing a shift towards smaller, customized AI models and privately negotiated agreements between major labels and AI companies, alongside discussions about national policies, taxation, and redistribution to support cultural infrastructure.
Key takeaway
For policy makers and legal professionals debating AI training data compensation, relying solely on complex attribution algorithms risks creating easily gamed systems. You should prioritize establishing multi-layered, auditable attribution frameworks that integrate computer science, musicology, law, and economics expertise. Additionally, consider broader policy tools like taxation and redistribution to support cultural infrastructure and ensure fair compensation for creative workers.
Key insights
New economic models are emerging to compensate musicians for AI training data, moving beyond traditional "use" definitions.
Principles
- AI attribution requires measuring causal influence, not just similarity.
- Attribution systems must be auditable and multi-layered.
- Economic structures should avoid being easily gamed.
Method
Sureel's method involves labeling media with owner instructions, tracking AI training usage, and setting licensing fees. SoundVerse proposes differentially rewarding training data based on its influence on specific AI outputs.
In practice
- Implement media labeling with AI usage instructions.
- Explore differential royalty models for training data.
- Form creator alliances for custom AI model training.
Topics
- AI Training Data
- Music Royalties
- Copyright Attribution
- Generative AI Ethics
- Licensing Agreements
- Cultural Policy
Best for: Investor, CTO, VP of Engineering/Data, Legal Professional, Policy Maker, AI Ethicist
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by IEEE Spectrum.