One of the bedrock elements to decentralization is ownership. Proof of ownership, owning a piece of digital art, supporting an artist through the purchase of nonfungible tokens (NFTs)—there are numerous ways in which the different facets of ownership impact the concept of decentralization as a whole.

The impact can also be felt within the burgeoning industry of artificial intelligence (AI), where questions surrounding ownership of content swirl regularly across all sectors—most notably in music and entertainment.

Enter Jen, the ethically-trained generative AI music platform powered by Futureverse and developed by a team of PhDs and scientists that introduces a new standard for copyright compliance in text-to-music generation. The platform’s goals are to adhere to transparency, compensation and copyright identification for the music it trains on and the tracks it helps users create.

With over 40 licensed catalogs in its initial training set, its prompt-based flow simplifies the creative process for artists, while also providing assurance that all outputs respect both artists and industry copyright standards. Jen’s music research approach has led its work to be cited by Meta’s AI Lab, Meta FAIR, Sony AI and Bytedance AI.

“When you type something and you hear a note and it just affects you, and then you want to go down the process of being able to iterate and evolve something—that’s how prolific some of these AI tools will end up being on the world when they’re helping people do things they’ve never done,” Shara Senderoff, Jen co-founder and music-tech entrepreneur, said to me in a recent interview. “It still requires a human element,” he said. “You have to sit down to work with Jen—the human still has to do something.”

Read more: New Music Streaming Platform Sona Aims to Disrupt Without Ruffling Industry Feathers

Much like how AI within the entertainment industry can be a tool used to help screenwriters and filmmakers—not replace them—AI within the music industry can be a tool to aid artists and producers with their music creation, not replace them.

“A huge part of my mission with Jen was to prove it could be done right, that you could license music and actually launch a product with licensed music,” Senderoff said. “We proved you can put out a model—and we put out a model that is very Alpha. It doesn’t know a lot of genres or song structures because we didn’t steal 10 million songs that other people own. How can [the AI] know song structure if it hasn’t been taught song structure?”

Senderoff acknowledges that Jen’s outputs would have come out the gate well ahead of its competition—namely Udio and Suno—if it, too, had trained on millions of commercial copyrighted tracks.

“We would have better models if we’d done that, but my approach was to prove that we can create a fair framework for the industry,” she said. “We had to architect a licensing model where creators—some of whom have made music their entire lives—felt valued, where their copyrights weren’t completely copied.”

To prove the model could be achieved, Senderoff and her team set up a training process with licensed and owned material—both master and publishing—and then brought in the owner of the material to ensure it hadn’t infringed on any copyrights.

“AI training is dangerous,” Senderoff said. “When something goes in, you can’t just be like, ‘Let me pull that out, let’s unlearn it.’ That’s not how it works. You have to completely retrain models from scratch, and models can’t be retrained in five minutes. It takes many weeks and excessive graphical processing units (GPUs) to train models on music catalogs. It’s not overnight.”

If the wrong content or if any copyright-infringing content gets through the system and poisons the pool, the entire system is at risk.

“We had to implement every precautionary gate possible,” Senderoff said. “We put a gate at the very beginning where all of the training data filters through 150 million copyrights where it’s being checked against a significant database of music.”

The Jen team also uses blockchain to generate a cryptographic hash for each track, which is then recorded on The Root Network blockchain. The process acts as an advanced form of verification, ensuring the integrity and timestamp of each track's creation.

“The reason it’s being hashed is because if you take the output and want to add a music sample on top of it and release it and tell people Jen made it, we’ll be able to provide proof of the original cryptographic hash and show how the files do not match one another because one was altered after the hash,” Senderoff said. “I built in those gates not because I wanted to be fancy with web3, but because that’s what the blockchain is for—it’s the technology. It’s advanced watermarking.”

Archaic licensing models

The technology could disrupt the entire music industry in a way we haven’t seen since Napster in the early aughts—a shift that for some executives is not only welcomed—but encouraged.

“I think archaic music licensing models are f*cking stupid,” said Les Borsai, former music manager and current co-founder of Wave Digital Assets. “I don’t believe intellectual property should be used as a weapon—as leverage—to get a better deal. That’s someone’s creative everything. For some business guy to say they’re going to hold it hostage or not license it to someone else because the numbers aren’t right—it’s bullsh*t. We don’t need to be so greedy that we take it all—the reason we take it all is because these companies are unnecessarily overstaffed. If we’re looking into a world that has AI and automation, all of these things need to be disrupted in the future.”

But we still might have yards to gain before that disruption takes hold. For Senderoff, AI music has so far been misunderstood.  

She said AI music has so far been misunderstood.  “You ask a lot of people what AI music is and a lot of people have a misconception that it’s illegal and is ruining the music industry. We have a large community in Futureverse, and a lot of people didn’t understand the ethical training detail. They didn’t realize—until the big lawsuits came out—how important it is to train AI using source material that does not harm artists.”

Anything an artist creates on Jen is theirs to own.  Later this fall, Jen will roll out R3CORD, a feature where artists can “record” tracks for sale, adding a marketplace layer to their personal creations.

“To ensure artistry maintains the value it deserves, we must commit to honor the creativity and copyrights of the past, while embracing the tools that will shape the next generation of music creation,” Senderoff said. “As we bring Jen to market, we are partnering with music rights holders and aligning with the industry at large to deliver what we believe is the most ethical approach to generative AI.”