An AI tech business, Musical.AI, demonstrated current technology that’s able to attribute the most influential songs used in the inputs for a piece of music generated by AI, at a SOCAN info session, Artificial Intelligence – Creator-Focused Companies, held Oct. 30, 2024.
Sean Power, co-founder and CEO of Musical.AI, said at the session that such technology defies the popular narrative that attribution is impossible when AI generates new songs.
Generative AI technology uses a very large set of data points – from multiple millions of songs – to create a “latent space,” and then uses statistical data in that space to predict models. Musical.AI’s technology is then able to determine by percentage the songs that were most influential in generating the new AI-created piece of music.
SOCAN Legal Counsel Fraser Turnbull, who moderated the event, talked about “the three Cs” for songwriters and composers whose songs or compositions are being used to generate new pieces of music in AI: credit, consent, and compensation. He stressed that transparency in the process is crucial in determining if SOCAN members’ works have been used for inputs, and the attribution process enabled by Musical.AI’s technology is important for determining when those works influence AI-generated outputs.
Participants agreed that crediting of generative AI inputs seems the likeliest path in the near future, following the precedent of the EU AI Act requiring transparency from general-purpose AI models, including AI music generators, and Bill AB-2013 in California, which mandates every AI business in California, and every company doing business with them, to disclose all of the datasets used to create their generative models. These transparency obligations are important because there can be no compensation to music creators without first being able to credit the music creators of the inputs.
Power recommended that, for the moment, SOCAN members shouldn’t license their work directly to an AI company, because there’s no method developed yet that allows them to track how it’s being used. But he added that tech companies are currently working on innovations to remedy that.