The rapid advancement of AI music generation is raising important questions about the fair use of copyrighted works to train these AI models. While AI music startups like Suno are generating significant buzz with their impressive text-to-song capabilities, there are growing concerns that some may be training their models on copyrighted music without obtaining the necessary licenses or consent from rights holders.
Everything you need to know:
✓ Questions raised over whether Suno and others exploit copyrighted music to train AI without licenses
✓ Unfair AI music training potentially undermines musicians, letting tech profit at their expense
✓ Supporting transparent, ethical AI crucial to ensure fairness to human creators as tech advances
Are AI music companies secretly stealing from artists?
The rapid advancement of AI music generation is raising important questions about the fair use of copyrighted works to train these AI models. While AI music startups like Suno are generating significant buzz with their impressive text-to-song capabilities, there are growing concerns that some may be training their models on copyrighted music without obtaining the necessary licenses or consent from rights holders.
In a recent guest column, Ed Newton-Rex, CEO of AI ethics certification company Fairly Trained, raises the alarm about Suno’s training practices. While praising the technical capabilities of Suno’s AI, Newton-Rex points to several clues suggesting the company likely trained its models on a vast corpus of copyrighted songs without appropriate licensing deals in place with major music labels and publishers.
Suno has so far declined to disclose details about its training data sources. In an interview with Rolling Stone, one of Suno’s investors acknowledged the company did not have label deals when it started, invested despite knowing Suno could get sued, and that the founders’ lack of open hostility to the music industry “doesn’t mean we’re not going to get sued.” Suno has also not engaged with Fairly Trained’s offer to get certified for ethical AI training practices, unlike a dozen other AI music startups.
If it’s true that Suno and potentially other AI music companies are exploiting copyrighted works without permission to train competing AI models, Newton-Rex argues this poses an existential threat to human musicians and the music industry. Generative AI music is already starting to eat into streaming revenues as some people listen to AI music instead of human-created songs on services like Spotify. If AI music training isn’t done above board with proper licensing, it will inevitably lead to lost income for musicians as AI unfairly competes with their work.
Two roads for AI music: Exploitation or ethics
There are two paths forward – the exploitative approach of training AI models on copyrighted music without consent, or the ethical approach of only using appropriately licensed or public domain works. Newton-Rex highlights that many AI music startups, around a dozen certified so far by Fairly Trained, are putting in the hard work to train their models fairly without resorting to exploiting musicians’ intellectual property.
However, some of the biggest generative AI companies continue to scrape huge datasets of creative works to train their models with little regard to licensing. This has sparked major public outcry from creators and a wave of lawsuits aimed at protecting their rights and livelihoods. If we as a society condone the unauthorized use of copyrighted material to train AI, we are implicitly supporting the destruction of the creative industries for the commercial benefit of big tech companies.
It’s crucial that the media and public apply more scrutiny to the training data practices of buzzy AI startups like Suno. We can’t let the hype around their capabilities blind us to unfair exploitation happening behind the scenes. Suno should disclose its data sources and prove it is not freeloading off musicians’ work if it wants to claim the ethical high ground. The AI music companies doing things the right way by actually licensing training data or using public domain works exclusively deserve our praise and support.
In the end, it doesn’t matter if you’re a team of musicians building the AI, claim to respect IP rights, or decorate your office with pictures of famous composers. As Newton-Rex argues, “If you train on copyrighted work without a license, you’re not on the side of musicians. You’re unfairly exploiting their work to build something that competes with them. You’re taking from them to your gain — and their cost.” Only by insisting on fair, transparent and ethical AI training practices can we ensure technological progress in music doesn’t come at the expense of human artists.
