YouTube’s ‘Dream Track’ Heralded as Groundbreaking — But is Google Infringing Songs to Build AI Platforms?

YouTube’s ‘Dream Track’ is heralded as a modern marvel of generative AI — but is Google infringing on artists’ copyright to build its AI platforms?

Google’s generative artificial intelligence platform Dream Track, designed around creating music for use in YouTube Shorts, has been well received by many prominent voices in favor of the already shifting world of music and AI. But not everyone in the music industry is thrilled about the idea of YouTube as an important player in the world of generative AI — particularly because Google trained its model on a large amount of copyrighted recordings without first securing permission.

Dream Track allows certain YouTube creators to generate a soundtrack for clips on Shorts based on text prompts that can include replicas of artists’ voices. Several artists knowingly participated, including Charli XCX and Demi Lovato. Still, negotiations for broader licensing deals on a label-wide scale continue to drag on, adding pressure on YouTube to progress its technology ahead of competitors.

Generative artificial intelligence models, or genAI, require machine learning to start properly generating output. Companies like Google utilizing existing copyrighted works in order to train their AI models isn’t an inherent problem should they gain permission from the rights holders from the outset. But seeking forgiveness rather than permission for the use of existent work has led to several high-profile lawsuits, including the Authors Guild v. OpenAI, and Getty Images v. Stability AI.

Universal Music Group (UMG), among a selection of other companies, sued AI startup Anthropic in October, claiming that “in the process of building and operating AI models, (Anthropic) unlawfully copies and disseminates vast amounts of copyrighted works.”

Cases like these are expected to set a precedent for AI training, but the process could take years — and the technology is developing at a breakneck pace.

Tech companies would like to believe their activities fall under fair use, as “innovation in AI fundamentally depends on the ability of (language models) to learn in the computational sense from the widest possible variety of publicly available material,” says Google.

“If a generative AI model is trained on music for the purpose of creating new musical works that compete in the music market, then the training is not a fair use,” says Dennis Kooker, president of global digital business and US sales for Sony Music Entertainment. “Training in that case cannot be without consent, credit, and compensation to the artists and rights holders.”

Getting major labels on board with genAI is an uphill battle. Securing the rights for the material to train the model only after doing so, but before releasing it, is less than ideal. But for some voices in the industry, that may be better than nothing at all.

And it’s hardly the first time Google has asked for forgiveness rather than permission. In 2004, the company began scanning books en masse without permission from rights holders to create Google Books. That led to a lawsuit from the Authors Guild, which was ultimately dismissed in 2013.

“What we in the AI world think of as ‘training data’ is what the rest of the world has thought of for a long time as creative output,” admits Ed Newton-Rex, former VP of Audio at Stability AI. “In that community, where you need a huge amount of data, you don’t see many people talking about the concerns of rights holders.”

But YouTube professes to give a damn, despite its track record of transgressions. A statement from a YouTube representative concludes:

“We remain committed to working collaboratively with our partners across the music industry to develop AI responsibly and in a way that rewards participants with long-term opportunities for monetization, controls, and attribution for potential genAI tools and content down the road.”