How Musicians Can Get Paid for Training AI

Overview

As generative AI companies train models on musicians' work, new startups like Sureel and SoundVerse are developing systems to ensure artists get paid each time their music influences AI outputs. Rather than accepting one-time buyouts, these platforms use attribution technology to track how training data contributes to generated outputs and calculate ongoing royalties, potentially reshaping how creative compensation works in the AI era.

Key Takeaways

Generative AI has blurred the definition of 'use' in music licensing-whether payment should occur at training time or each time the model produces output remains contested.
Sureel, recently acquired by Warner Music Group, uses software to label music files with licensing instructions and track how AI companies use them in training to set dynamic fees.
SoundVerse advocates for ongoing artist participation and differential rewards based on how much each piece of training data influences each AI-generated output, rather than flat one-time payments.
Attribution technology must measure causality between training data and AI outputs, a complex challenge that goes beyond simple similarity matching.
Well-designed attribution systems could incentivize musical experimentation and diversity by rewarding unusual or unpolished works, not just commercially popular tracks.

Stats & Key Facts

#Warner Music Group acquired Sureel, signaling major label investment in AI attribution solutions

How Musicians Can Get Paid for Training AI

The Core Problem: Redefining 'Use' in the AI Age

Musicians have long been compensated based on how much their work is used across multiple revenue streams.

›Traditional music economics cover vinyl/CD sales, streams, radio play, cover versions, kakaoke, and other licensing categories, each with clear payment agreements.
›Generative AI disrupts this model because the definition of 'use' is ambiguous: does payment trigger once at training, or repeatedly as the model generates outputs?
›Some argue training data is 'used' only once during model development, while creators reasonably contend that their work's creative essence persists in the model's outputs.
›The industry has been accused of 'the biggest act of copyright theft in history,' making fair compensation mechanisms urgent.

Sureel's Labeling and Tracking Approach

Sureel, now backed by Warner Music Group after acquisition, proposes a metadata-based solution to music licensing for AI.

›The platform labels online media files with owner-determined instructions specifying whether AI companies can use the music freely, with limitations, or not at all.
›After labeling, Sureel tracks how AI companies actually use the media during training and calculates licensing fees based on actual usage patterns.
›Sureel has partnered with Swedish copyright agency STIM to explore how these licensing frameworks could be formalized in agreements between musicians and AI developers.
›Co-President Benji Rogers frames attribution as 'measuring, for the first time, the thing the old economics only approximated'-moving from approximation to precision.

SoundVerse's Differential Reward Model

Competing with Sureel's approach, SoundVerse advocates for continuous compensation tied to each AI-generated output.

›SoundVerse founders reject one-time royalty buyouts as fundamentally insufficient compensation for ongoing use of training data.
›Their model assumes that different pieces of training data contribute unequally to each generated output-jazz training data contributes more when the AI generates jazz-sounding music.
›Rather than fixed fees, the platform proposes differential rewards: each piece of training data receives payment proportional to its estimated influence on each specific AI output.
›This approach aligns compensation with actual creative contribution, theoretically providing ongoing income rather than a single upfront payment.

The Attribution Challenge: Measuring Causality

While the concept of differential rewards is appealing, implementing it requires solving a complex technical problem.

›Attribution must go beyond surface-level similarity matching to identify causal relationships-how a specific training data point actually shaped the trained model.
›Sureel CEO Tamay Aykut emphasizes that measuring influence requires understanding the genuine relationship between training data and the resulting AI system.
›Information theoretic principles and modeling historical impact of individual works may be necessary to achieve meaningful attribution.
›Even with perfect attribution, the system could incentivize bad behavior: people might create music designed specifically to maximize training-data royalties rather than to express genuine creativity.

The Risk of Gaming the System

New economic incentives inevitably reshape creative output, sometimes in undesirable ways.

›Music streaming led to shorter song intros as artists adapted to platform economics; AI attribution could similarly distort creative decision-making.
›Without safeguards, reverse-engineered pastiches and commercially calculated works could siphon royalties away from original, experimental music.
›The industry must design systems that discourage gaming while maintaining genuine incentives for quality and innovation.
›Aykut suggests well-designed systems could actually reward unusual and unpolished works more highly than radio-friendly standards, potentially promoting diversity instead of homogenization.

Creating Incentives for Experimentation and Diversity

Beyond solving the immediate compensation problem, attribution systems could reshape the entire creative landscape.

›Current streaming economics favor popularity, which can discourage artistic risk-taking and experimental work.
›If attribution systems value novelty and originality over commercial appeal, they could motivate musicians to push creative boundaries.
›The Swedish copyright agency STIM is actively exploring how Sureel's attribution reports could underlie future licensing agreements, signaling industry momentum toward formal adoption.
›The compelling possibility is that AI attribution could simultaneously sustain economic incentives for creativity while encouraging more diverse and experimental musical output.

Moving Toward Industry Harmony

These emerging platforms represent a shift from confrontation toward sustainable coexistence between AI companies and musicians.

›Rather than one-time buyouts that feel like capitulation, ongoing attribution-based payment models create a shared economic interest in long-term collaboration.
›Major label acquisition of Sureel by Warner Music Group signals that established music industry players see AI attribution as a necessary evolution, not a threat.
›STIM's involvement indicates that copyright organizations are actively working to adapt existing frameworks to the AI era.
›Success depends on technical precision, industry adoption, and regulatory acceptance-all still in development.

Frequently Asked Questions

How does Sureel's system work?

Sureel labels music files with owner-specified instructions about AI use (free, limited, or prohibited), then tracks how AI companies actually use the music during training and calculates licensing fees based on the actual usage patterns.

What's the difference between Sureel and SoundVerse's approaches?

Sureel focuses on tracking usage at training time and setting fees accordingly, while SoundVerse advocates for differential rewards based on how much each piece of training data influences each AI-generated output, creating ongoing payments rather than one-time fees.

Why is attributing influence from training data so difficult?

Attribution must measure causality-the genuine relationship between specific training data and the trained model-rather than just similarity. This requires advanced information theoretic principles and careful system design to avoid both inaccuracy and incentivizing artists to create music designed to game the royalty system.

Could attribution systems encourage musical creativity?

Yes, if properly designed to reward unusual and unpolished works highly, attribution could incentivize experimentation and diversity rather than just commercially popular standards, potentially reshaping the entire creative landscape.

What does Warner Music Group's acquisition of Sureel mean?

It signals that major music industry players see AI attribution technology as essential to the future of music licensing, validating that Sureel's approach represents a legitimate path toward sustainable artist compensation in the AI era.

As these attribution technologies mature, they could transform generative AI from an existential threat to musicians into a sustainable economic model that rewards creativity and diversity.