NEW YORK: OkSeven, a content licensing seller of music, images, videos and other datasets used to train artificial intelligence systems, said on Wednesday it has formed the industry's first trade association.
The Dataset Provider Alliance (DPA) will promote “ethical data sourcing” in training AI systems, including protecting the rights of people depicted in datasets and the intellectual property rights of content owners, the companies said in a statement.
Founding members include US music dataset company Rightsify, image licensing service vAIsual, Japanese stock photo provider Pixta and Germany-based data marketplace Datarade.
In recent years, the emergence of generative AI techniques that can mimic human creativity has sparked protests from content creators and a series of copyright lawsuits against tech companies such as Google, Meta and Microsoft-backed ChatGPT-maker OpenAI.
Developers have trained their models with vast amounts of content, much of it collected for free from the internet, without the consent of those who created the works or own the rights to them.
Tech companies claim their use is legal, but they secretly pay for access to private collections of content to fulfill certain types of data needs and to avoid legal and regulatory risks.
The prospect of increased demand for licensed data if copyright holders win their legal battles has spurred the emergence of a new industry of companies packaging content and selling access to it for use in AI systems.
As a result, groups have been formed to establish ethical standards in the industry, such as Fairly Trained, a nonprofit group founded this year that certifies models who do not use copyrighted material without permission.
The DPA covers the content of these transactions and, for example, requires members to agree not to sell text data obtained from crawling the web, or audio recordings of people's voices, without their explicit consent.
Alex Vestal, CEO of Rightsify and its licensing subsidiary GCX, who led the group's founding, said the group would focus on pushing for legislation like the NO FAKES Act, a U.S. bill introduced last year that would make it penalties to make unauthorized digital copies of people's voices and likenesses.
“Advocacy will be a big part of it as everyone has their own position on AI and copyright, but many of these battles have yet to be resolved and will take some time to be resolved,” Vestal said.
The DPA added that it will also push for stronger training data transparency requirements, such as those in European Union AI law and a similar U.S. bill introduced in April, the Generative AI Copyright Disclosure Act.
The group plans to release a white paper outlining its position in July, he said.
