djfrosty

When Drake dismissively told Metro Boomin to go and “make some drums” in one of his recent diss tracks during his beef with Kendrick Lamar, the superproducer went off and did just that — and the result marked a turning point for the use of AI in music production. 

The beat, titled “BBL Drizzy,” pairs a vintage-sounding soul vocalist over some 808 drums. The producer released it to SoundCloud on May 5, encouraging his fans to record their own bars over it for the chance to win a free beat, and it swiftly went viral.

But soon after, it was revealed that the singer from the “BBL Drizzy” beat didn’t exist — the voice was AI-generated, as was the song itself. The vocals, melody and instrumental of the sample were generated by Udio, an AI music startup founded by former Google Deep Mind engineers. Though Metro was not aware of the source of the track when he used it, his tongue-in-cheek diss became the first notable use case of AI-generated sampling, proving the potential for AI to impact music production. (A representative for Metro Boomin did not respond to Billboard’s request for comment).

As with all AI tracks, however, a human being prompted it. King Willonius, a comedian, musician and content creator, had put together the Udio-generated song on April 14, pulling inspiration from a recent Rick Ross tweet — in which the rapper joked that Drake looks like he got a Brazilian Butt Lift — to write the lyrics. “I think it’s a misconception that people think AI wrote ‘BBL Drizzy,’” Willonius told Billboard in an interview about the track. “There’s no way AI could write lyrics like ‘I’m thicker than a Snicker and I got the best BBL in history,’” he adds, laughing. 

There are a lot of issues — legal, philosophical, cultural and technical — that are still to be sorted out before this kind of sampling hits the mainstream, but it’s not hard to imagine a future where producers turn to AI to create vintage-sounding samples to chop up and use in beats given that sample clearances are notoriously complicated and can drag on for months or years, even for big name producers like Metro Boomin. 

“If people on the other side [of sample clearance negotiations] know they’re probably going to make money on the new song, like with a Metro Boomin-level artist, they will make it a priority to clear a sample quickly, but that’s not how it is for everyone,” says Todd Rubenstein, a music attorney and founder of Todd Rubenstein Law. Grammy-winning writer/producer Oak Felder says clearing a sample for even a high-profile track is still a challenge for him. “I’ll be honest, I’m dealing with a tough clearance right now, and I’ve dealt with it before,” he says. “I had trouble clearing an Annie Lennox sample for a Nicki Minaj record once… It’s hard.”

Many smaller producers are not able to sample established songs because they know that it could get them into legal trouble. Others go ahead without permission, causing massive legal headaches, like when bedroom producer Young Kio sampled an undisclosed Nine Inch Nails song in an instrumental he licensed out on BeatStars. The beat was used by then-unknown Lil Nas X and resulted in the Billboard Hot 100 No. 1 “Old Town Road.” When the sample was discovered, Nas was forced to give up a large portion of his publishing and master royalties to the band. 

Udio’s co-founder, David Ding, tells Billboard that he believes AI samples “could simplify a lot of the rights management” issues inherent to sampling and explains that Udio’s model is particularly adept at making realistic songs in the vein of “Motown ‘70s soul,” perhaps the most common style of music sampled in hip-hop today, as well as classical, electronic and more. “It’s a wide-ranging model,” Ding says.

Willonius believes AI samples also offer a solution for musicians in today’s relentless online news cycle. While he has made plenty of songs from scratch before, Willonius says AI offered him the chance to respond in real-time to the breakneck pace of the feud between Drake and Kendrick. “I never could’ve done that without AI tools,” he says. Evan Bogart, a Grammy-winning songwriter and founder of Seeker Music, likens it to a form of digital crate digging. “I think it’s super cool to use AI in this way,” he says. “It’s good for when you dig and can’t find the right fit. Now, you can also try to just generate new ideas that sound like old soul samples.”

There’s a significant financial impact incurred from traditional sampling that could also be avoided with AI. To use the melody of “My Favorite Things” in her hit song “7 Rings,” for example, Ariana Grande famously had to cede 90% of her publishing income for the song to “My Favorite Things” writers Rodgers and Hammerstein — and that was just an interpolation rather than a full sample, which entails both the use of compositional elements, like melody, and a portion of the sound recording.

“It certainly could help you having to avoid paying other people and avoid the hassle,” says Rubenstein, who has often dealt with the complications of clearing songs that use samples and beats from marketplaces like BeatStars. But he adds that any user of these AI models must use caution, saying it won’t always make clearances easier: “You really need to know what the terms of service are whenever you use an AI model, and you should know how they train their AI.”

Often, music-making AI models train on copyrighted material without the consent or compensation of its rights holders, a practice that is largely condemned by the music business — even those who are excited about the future of AI tools. Though these AI companies argue this is “fair use,” the legality of this practice is still being determined in the United States. The New York Times has launched a lawsuit against OpenAI for training on its copyrighted archives without consent, credit or compensation, and UMG, Concord, ABKCO and other music publishers have also filed a lawsuit against Anthropic for using their lyrics to train the company’s large language model. Rep. Adam Schiff (D-CA) has also introduced a new bill called the Generative AI Copyright Disclosure Act to require transparency on this matter. 

Udio’s terms of service puts the risk of sharing its AI songs on users, saying that users “shall defend, indemnify, and hold the company entities harmless from and against any and all claims, costs, damages, losses, liabilities and expenses” that come from using whatever works are generated on the platform. In an interview with Billboard, Udio co-founder Ding was unable to answer what works were specifically used in its training data. “We can’t reveal the exact source of our training data. We train our model on publicly available data that we obtained from the internet. It’s basically, like, we train the model on good music just like how human musicians would listen to music,” says Ding. When pressed about copyrights in particular, he replies, “We can’t really comment on that.”

“I think if it’s done right, AI could make things so much easier in this area. It’s extremely fun and exciting but only with the proper license,” says Diaa El All, CEO/founder of Soundful, another AI music company that generates instrumentals specifically. His company is certified by Fairly Trained, a non-profit that ensures certified companies do not use copyrighted materials in training data without consent. El All says that creating novel forms of AI sampling “is a huge focus” for his company, adding that Soundful is working with an artist right now to develop a fine-tuned model to create AI samples based on pre-existing works. 

“I can’t tell you who it is, but it’s a big rapper,” he says. “His favorite producer passed away. The rapper wants to leverage a specific album from that producer to sample. So we got a clearance from the producer’s team to now build a private generative AI model for the rapper to use to come up with beats that are inspired by that producer’s specific album.”

While this will certainly have an impact on the way producers work in the future, Felder and Bogart say that AI sampling will never totally replace the original practice. “People love nostalgia; that’s what a sample can bring,” says Felder. With the success of sample-driven pop songs at the top of the Hot 100 and the number of movie sequels hitting box office highs, it’s clear that there is an appetite for familiarity, and AI originals cannot feed that same craving.

“BBL Drizzy” might’ve been made as a joke, but Felder believes the beat has serious consequences. “I think this is very important,” he says. “This is one of the first successful uses [AI sampling] on a commercial level, but in a year’s time, there’s going to be 1,000 of these. Well, I bet there’s already a thousand of these now.”

This story is included in Billboard‘s new music technology newsletter, Machine Learnings. To subscribe to this and other Billboard newsletters, click here.