“Fair training” becomes a thing to protect copyrighted audio in AI training

A unique type of content ownership certification is in the works from a non-profit group called Fairly Trained. As the name implies, the focus is on media content — including music — used to train artificial intelligence services.

 

“Generative AI systems can produce images, video, text and audio in seconds, based on little more than a text prompt. These systems are technically impressive, in part because of the training data that goes into them: the creative output of millions of creators around the world.” –Fairly Trained

 

The problem there is that the creators are sometimes unaware of how their work is being ingested into AI systems for training. In the audio realm, such content includes music and voice. That combustible situation can lead to AI output which is legally actionable.

The new firm will add a certification label to companies which prove the asked for permission to use copyrighted training material. According to The Verge, fair use arguments are not enough, and certifications will be refused if they rely on that slippery slope.

Fairly Trained says it is launching with a portfolio of nine companies already certified. They are Beatoven.AI, Boomy, BRIA AI, Endel, LifeScore, Rightsify, Somms.ai, Soundful, and Tuney.

Fairly Trained was founded by former VP for Audio at Stability AI, Ed Newton-Rex. The UK native has spent time at Voisey and Adreessen Horowitz.

(H/T: Michael Robertson)


Brad Hill