How does STIM ensure that compensation from the AI license reaches the right music creators and publishers?

Since the licensee compensates the rightsholders both for training the AI model and for the music that is later generated, two different distribution principles are needed. 

Compensation for training is distributed proportionally and equally among all works included in the license. 

For generated music (output), a different method is required: attribution, in which the compensation is linked to the specific works that influenced the output. Where attribution does not provide a reliable basis, analogy-based distribution will be used instead, based on how the works were used in the AI service’s training data. 

STIM is closely following the development of attribution technology. It is a relatively new and untested method, but if proven reliable, it can enable distribution of income based on actual influence. 

As part of the license, STIM is introducing attribution. This means that relevant data is analyzed and structured before it is provided to the AI service. The service is then required to use an attribution provider (such as Sureel) to analyze the generated music. Initially, STIM will require licensees to submit such analyses monthly, to learn more, monitor developments, and evaluate the technology.