Fairly Trained, a non-profit that certifies AI developers who ethically train their technology (for instance, by not using copyrighted materials without permission) is rolling out tougher, more specific rules for its certification badges.
Many AI developers use architectures that include more than one model. These models serve different functions in the AI technology the company is building. So far, Fairly Trained has offered developers certification badges based on specific models they use.
Now, however, the non-profit will be issuing new badges to certified companies, and those companies “will be required to be transparent with users about which elements of their architecture are and are not certified,” Fairly Trained said in a statement issued on Thursday (August 1).
“We think this transparency is critical to let users make informed decisions about which AI products to use,” Fairly Trained said.
So far, two AI developers have received the new certification badge that requires transparency about all models used: Generative music platform Jen and audio tool company Kits AI.
“It’s been our top priority at Jen to architect an approach that leads the way for ethically-trained models to establish a dominant presence at the forefront of their industries,” Jen Co-Founder and CEO Shara Senderoff said in a statement.
“We believe that this prioritization not only adheres to responsible practices but also positions models to be trusted and widely adopted. We’re incredibly proud to have our audio models certified by Fairly Trained, a meaningful alignment in our commitment to transparency.”
“We think this transparency is critical to let users make informed decisions about which AI products to use.”
Fairly Trained
Jen’s omnidirectional diffusion models have all been certified, Fairly Trained said. Meanwhile, Kits AI’s voice and instrument models have been verified, including its voice library, which pays artists per download if they have trained their voices on the Kits platform.
“At Kits, a big part of our mission is to empower vocalists and creators with AI tools that respect artistry, fairly compensate artists, and promote a sustainable future for music — something we believe Fairly Trained certifications move us all closer to,” said Pippin Bongiovanni, Kits’ Director of Marketing.
Fairly Trained was founded earlier this year by Ed Newton-Rex, a California-based entrepreneur who founded the AI music-making platform Jukedeck a decade ago.
“It’s been our top priority at Jen to architect an approach that leads the way for ethically-trained models to establish a dominant presence at the forefront of their industries.”
Shara Senderoff, Jen
In a widely-read column at MBW, Newton-Rex explained why he left his job at Stability AI, maker of the popular image generator Stable Diffusion.
“I don’t agree with the company’s opinion that training generative AI models on copyrighted works is ‘fair use’,” Newton-Rex wrote, referring to the carve-out in US copyright law that allows unauthorized use of copyrighted works for certain limited uses, such as news and education.
Some AI developers, facing lawsuits from copyright holders, have argued that training AI should be considered fair use – a stance most rights holders disagree with.
Some AI developers have argued that certain AI models – particularly large language models (LLMs) require such large data sets to be trained that it’s practically impossible to train them without using large quantities of copyrighted materials.
Fairly Trained has disputed this assertion, and backed up its argument when it certified its first LLM in March of this year – KL3M, from 273 Ventures.
Fairly Trained also announced it had certified Voicemod, the first company offering AI speech and singing models to receive the certification.
Additionally, two music companies that use generative AI – Infinite Album and Lemonaide – have also gained certification, as has an “AI band” called Frostbite Orckings.
As Fairly Trained hasn’t yet announced whether these companies have received its updated new certification badge, it’s unknown whether all of their models comply with Fairly Trained’s new standards for certification.Music Business Worldwide