Ed Newton-Rex pointed out some interesting similarities between copyrighted tracks and pieces generated by the "ChatGPT for music".
With the continued growth of generative AIs in both their capabilities and numbers, it was only a matter of time before they extended their influence beyond the realm of digital art, causing concern among representatives of other industries about the possibility of being replaced by machines. One such example that the internet has been all the rage recently is Suno, an AI that allows users to generate music tracks using text prompts, capable of mimicking various musical styles, creating melodies, and even vocalizing the provided lyrics.
Apart from enjoying the recent spike in popularity, this "ChatGPT for music" has also drawn the attention of those opposed to AI, who understandably raised the question that AI developers really don't like to answer: how was Suno trained? As users experiment with the model, some have started noticing similarities between Suno-generated tracks and well-known pieces of music, spanning classical compositions to 20th-century hits and contemporary tunes.
One enthusiast in particular was Ed Newton-Rex, who, being a Composer, former VP Audio at Stability AI, and advocate for fair AI training, possesses expertise in both the AI and music fields. Recently, Newton-Rex joined Music Business Worldwide to share a comprehensive analysis of Suno, which led to a depressing conclusion that there's a high likelihood Suno was trained on copyrighted music, without the necessary licenses held by its creators.
In the article, Newton-Rex thoroughly detailed his experiments with Suno, revealing that the AI is capable of producing music in the style of numerous artists and bands, including Eminem, Ed Sheeran, ABBA, Oasis, Queen, and Blink-182, just to name a few. What's more alarming, the author discovered that the AI can easily mimic melodies and chord sequences from popular songs, providing music sheets that highlight these similarities.
With the low probability of these occurrences being coincidental, Newton-Rex concluded that Suno was likely trained on copyrighted music. While it's plausible that the developers obtained all the required licenses, Newton-Rex pointed out that "one of Suno's investors indicated they didn't have licenses when they invested, and Billboard's sources recently confirmed they don't have licenses in place with the three major labels." Not surprisingly, Suno's creators have remained silent on the matter, leaving the question of "how was Suno trained?" without a clear answer.
Additionally, the author shared a TL;DR version of the analysis over on Twitter:
So, what do you think? Is it all just a big coincidence or was Suno AI actually trained on copyrighted materials? Leave your thoughts in the comments!
Read Ed Newton-Rex's original breakdown on Music Business Worldwide and don't forget to join our 80 Level Talent platform and our Telegram channel, follow us on Instagram, Twitter, and LinkedIn, where we share breakdowns, the latest news, awesome artworks, and more.