[News] The AI Thread!

At first, the YouTube videos look like scenes from Nickelodeon’s popular “SpongeBob SquarePants” cartoon.

SpongeBob, the title cheery yellow character, appears outside his pineapple-shaped home, while Mr. Krabs, SpongeBob’s cranky boss, is at the Krusty Krab restaurant he runs. But unlike in the show, the characters in the videos aren’t singing jolly songs about life in the underwater city of Bikini Bottom. Instead, they’re rapping about drugs and guns.

The mastermind behind the raps is an artist named Glorb. Their music, which has been streamed millions of times on Spotify and YouTube, appears to use artificial intelligence to replicate the iconic characters’ voices.

As AI tools continue to evolve rapidly, it has become easier for artists like Glorb to make music using generative AI — and become successful in their own right. However, experts who focus on AI and music said questions surrounding copyright and ownership still linger as a new era of technology dawns in the music industry.

“It opens up so many more possibilities for someone, you know, to essentially have, like, a fan fiction version of a song because they love the artist,” said Josh Antonuccio, an associate professor and the director of the School of Media Arts & Studies at the Ohio University Scripps College of Communication.

The SpongeBob-inspired tracks have turned Glorb — who keeps their identity anonymous — into an online sensation. On Spotify, Glorb averages just under a million listeners a month — their most popular song, “The Bottom 2,” has amassed more than 11 million streams. The artist’s music videos, which feature character models from the show, have also racked up millions of views on YouTube.

Glorb, who declined to be interviewed, isn’t publicly affiliated with Nickelodeon. A spokesperson for the Paramount-owned network didn’t immediately respond to a request for comment. Representatives for YouTube and Spotify also didn’t immediately respond to requests for comment.The music industry could see an influx of artists who use some kind of AI, especially as technology continues to advance, said Tracy Chan, the CEO of Splash, a generative AI music company. Already, generative AI music programs like Suno, which allows users to enter prompts and generate songs based on the text suggestions, have been hailed as the ChatGPT of music.

“I think it’s important that we figure out how to both, as an industry … how do you kind of balance that we’re creating more and more content, which is ultimately good, but also kind of rewarding the folks that are, you know, kind of the source material, so to speak,” Chan said.

Glorb isn’t the first to use the technology to create original music. In some cases, major artists have been involved with AI renditions of their work. In June, Paul McCartney announced The Beatles would release one final record, “Now and Then,” using AI technology to extract the voice of the late John Lennon. The singer Grimes, a champion of AI, released elf.tech, a platform on which artists can use an AI replication of Grimes’ voice in their music. The terms of the agreement include that Grimes receives part of the royalties earned from any music that includes the AI version of her voice.

But in other instances, AI-generated music using artists’ work has sparked some concerns from those in the music industry.

In April 2023, an artist named Ghostwriter went viral for the track “heart on my sleeve,” which used AI voice replications of the rapper Drake and the singer The Weeknd. The song was quickly removed from several platforms, including YouTube, where a message read: “This video is no longer available due to a copyright claim by Universal Music Group.”

Shortly before the Ghostwriter song circulated online, UMG (which has no relationship to NBCUniversal, the parent company of NBC News) had urged streaming services to prohibit AI programs from using its copyrighted music to train themselves.

“We have a moral and commercial responsibility to our artists to work to prevent the unauthorized use of their music and to stop platforms from ingesting content that violates the rights of artists and other creators,” UMG, which is considered one of the so-called Big Three global music companies, said in a statement to The Financial Times. “We expect our platform partners will want to prevent their services from being used in ways that harm artists.”

Part of the problem stems from the fact that music streaming platforms have few tools to detect and track how much AI music is on their apps, Chan said.

He compared traditionally created music to a fingerprint — streaming platforms can compare other songs against that fingerprint, and when they find a track that matches it, they can assess the upload and remove it if necessary. AI-generated music doesn’t have that hypothetical fingerprint. Therefore, it’s much harder to track and remove. Because there is limited technology to track AI music uploaded to various platforms, it’s hard to know how much of it is out there, Chan said.

“You’ve got to believe that it exists, but, again, is it reaching mass consumption? Probably not yet,” he said. “Because once it kind of hits the culture, so to speak, that’s where I think kind of a lot of the rights holders like labels and such [will] take action against those platforms and ask them to take it down.”

Lawmakers are already beginning to consider how to regulate AI-generated voices in music.

Last month, Tennessee Gov. Bill Lee signed the Ensuring Likeness Voice and Image Security Act — also known as the “ELVIS Act.” The law, which claims to be the first of its kind, “build[s] upon existing state rule protecting against the unauthorized use of someone’s likeness by adding ‘voice’ to the realm it protects,” Lee’s office said in a news release in January.

Many in the industry, including the Recording Academy and Warner Music Group CEO Robert Kyncl, praised the legislation.

Antonuccio, the Ohio University associate professor, said a wave of technology-infused music should both excite and frighten the industry and consumers.Even if more laws are introduced, Antonuccio said, trying to curb the tsunami of content that uses generative AI voices will remain almost impossible.

“There is an extreme generative remix culture that we are just beginning to enter into,” he said. “And I think there are some exciting parts of that, but frankly, I think there’s, there’s many things that should concern all of us with that.”

Spotify says that “tens of millions” of the 100 million tracks in its library have been streamed at least once but fewer than 1,000 times annually, representing 0.5% of the streamer’s stream-share royalty pool. Based on Spotify’s current per-stream rate, 1,000 annual streams generates around $3, often below the minimum that many distributors require before making payouts to artists. Under the status quo, the money Spotify pays out on those songs remains with the distributor until the threshold for payment to the artists is reached. Under the new policy, Spotify will withhold those royalties and roll them into the stream-share pool, now limited to songs with more than 1,000 streams.

The other major policy changes announced today are targeted at practices the company considers fraudulent: streaming bots and short-form “functional noise” content. Spotify currently removes songs from its library when it detects artificial streams generated from bots or scripts; starting in 2024, the company plans to penalize labels and distributors with per-track penalties when “flagrant” artificial streaming is detected, but did not specify what those penalties would be, or how its tools detect such activity.

The company also seeks to target “bad actors” that publish short-form noise tracks such as whale sounds, ASMR, and white noise, then stack them in playlists to generate what it deems “outsized payments.”

In the outgoing system, Spotify has paid the same royalty rate for a five-minute track and a 30-second track, meaning that a 100-track playlist of 30-second tracks could generate significantly more royalties than a 10-track playlist of five-minute tracks, despite being the same 50 minutes of content.

The company plans to increase the minimum length of “functional noise recordings” required to generate royalties from 30 seconds to two minutes, as well as “work with licensors to value noise streams at a fraction of the value of music streams.”

Spotify did not specify what fraction of the current rate it seeks, what criteria it will use to determine which tracks are functional noise, or by what means it would determine which tracks met such criteria.