At a recent Senate Judiciary Committee hearing on artificial intelligence (AI) and copyright, Ben Brooks, the head of public policy at Stability AI, admitted to using billions of images without obtaining the copyright owners’ consent to train their AI image generator, Stable Diffusion.
This revelation only reaffirms what is already known in the AI image industry, as other leaders such as David Holz from Midjourney have also confessed to utilizing copyrighted works without permission.
During the hearing, artist representative Karla Ortiz took the opportunity to attack Brooks over this practice.
Senator Mazie Hirono confronted Brooks, questioning whether he compensated the creators for using their content to train the AI image generator.
Uncomfortably, Brooks responded that if the image data is freely available on the internet and not subject to an opt-out request in their models, they would use it without payment or consent.
Earlier in the hearing, Senator Chris Crooks inquired about whether Stability AI pays for its data. Brooks skillfully evaded the question, resulting in another uncomfortable moment when Senator Hirono clarified that Stability AI does not pay for the data they use.
When Senator Hirono turned to Ortiz for her input, the artist shared the concerns echoed by many photographers and artists regarding AI image generators.
Ortiz stated that she has never been asked for consent, credited, or compensated for the use of her work, both personal and commercial, to train AI. She emphasized that companies like Stability AI do not provide her with a choice.
Ortiz’s perspective reflects that of countless photographers worldwide. Stability AI relies on the LAION dataset, which is a publicly available data pool that can be freely searched.
The website Have I Been Trained enables copyright holders to determine if their images were included in the data model.