AI Briefing: The music industry raises the decibel for the fight over generative AI

Last month was marked by a flurry of announcements from industry giants and startups alike. The same week Universal Music Group and SoundLabs revealed an AI vocal plug-in for artists using their own voices, the tech startup Futureverse began an alpha launch for Jen, an AI music model trained on dozens of licensed music catalogs. Other recent generative AI updates include Google’s DeepMind announcing a new tool for making video soundtracks, and ElevenLabs debuting a new text-to-audio app and Stability AI releasing a new AI sound generator.

Amid these advancements, tension is brewing as popular AI music platforms face heightened scrutiny and legal challenges. Startups like Suno and Udio were recently sued by major record labels for alleged copyright violations. But the recording industry’s lawsuits could also help prioritize the use of generative AI tools trained with licensing datasets. 

Last week, a group of content-licensing providers announced the formation Dataset Providers Alliance (DPA), of a trade group that aims to promote ethically sourced data. One of the founding members is Rightsify, an AI music startup that offers licensed music datasets to developers and uses its Hydra II model to create AI-generated music for hotels and videos. 

According to Rightsify CEO Alex Bestall, one of the DPA’s goals is to help advocate for ethical datasets across content music, voice, text, video and image content. He thinks the new record label lawsuits could also be a “net plus” for raising awareness about ethical datasets and pushing companies towards proactive licensing. He added that the new complaints feel different than earlier lawsuits related to AI and copyright: “This one seems a lot more specific and I think that’s probably the most impactful thing recently.” 

“If you or I did as an individual [sampled songs without permission], we’d be sued because it’s an unlicensed sample,” Bestall told Digiday. “Just because it was done through a [AI] model, why should it be exempt? So it’s unlicensed sampling, it’s publicly performing the lyrics, which also are unlicensed. There are multiple copyright infringements there as far as we’re concerned.”

To create its Hydra II model, Rightsify licensed music through direct agreements with musicians. It also obtained likeness and biometric releases — particularly important for privacy implications when datasets involve identifiable voices — and worked with the organization Fairly Trained to help ensure the ethical integrity of Hydra II.

Another startup with a new AI music model trained on only licensed content is Futureverse, which earlier this month debuted Jen, which was trained on 40 fully licensed music catalogs. Everyone involved with each song had to sign off, said Shara Senderoff, co-founder and CEO of Futureverse. The training data also went through a database of 150 million songs to make sure nothing violated copyright protections, with every AI song made then goes through the database a second time to make sure nothing gets flagged. 

For each song Jen creates, Futureverse also generates a cryptographic hash on a blockchain to verify both the source material and that Jen created the song. When asked if Futureverse plans to publish its full training set, Senderoff said the reason for not releasing it is “this has to be done right and we want to control our own destiny.” She also hopes to show a licensing approach to training an AI music model can work. (Futureverse also debuted a new platform called R3CORD that lets users “record,” share and sell tracks through a marketplace for creators.)

According to Senderoff, the plan is to add a “considerable number” of new Jen-enabled features in the next six months ranging from user-generated content to an enterprise-grade API to musicians experimenting with AI music. For example, users might use Jen to remove parts of a song or upload individual stems for generating entirely new songs. She hopes Jen can help with everything from user-generated content to tools for pro-consumers and A-list producers.

“At the end of the day, the industry understands that this has the opportunity to increase creativity, not decrease it,” Senderoff said. “…We believe we’re making an absolute co-producer.”

Consent, credit, and compensation are crucial for ethical AI, says Mike Huppe, CEO of SoundExchange. He suggests tech giants could negotiate revenue sharing from AI-generated content with material providers, weighted by usage, artist notoriety, and influence. Huppe emphasizes that this is a hypothetical solution, not an official stance, and that both AI datasets and outputs need separate consideration.

“There’s very complex pricing and payment models that exist on the internet or not on the internet for all sorts of things,” Huppe said. “This is doable…“How do we operationalize [AI] on the back end and how do we make sure that creators of all types fairly participate in the wealth that is being created?”

Prompts and Products: Other AI news and announcements

• Toys”R”Us used OpenAI’s Sora platform to create an AI-generated video depicting an AI-generated version of a young Charles Lazarus. Others brands to release new campaigns with generative AI included Motorola and Snickers.
• NBC announced it will use an AI voice of Al Michaels for narration summaries during the 2024 Paris Olympics.
• The Recording Industry Association of America and several major record labels sued two popular AI music startups citing copyright infringement.
• The design platform Figma released new generative AI tools for creating apps, and other content.
• Notion released a new way to create entire websites using generative AI.
• Google Deepmind released research about how generative AI is being misused for political misinformation.
• OpenAI said its ChatGPT desktop app for MacOS is now available to all users. The company also delayed its release of a “Voice Mode” feature announced earlier this spring.
• Time announced a new licensing deal with OpenAI, making it the latest media company to strike a deal with the AI startup. Meanwhile, the center for investigative reporting filed a lawsuit against OpenAI and Microsoft alleging copyright infringement.
• A new survey from Salesforce finds only 10% of global workers currently trust AI to operate autonomously, but a total of 77% say they eventually will.