AI-powered image and text generation has turned the Web upside down. In recent weeks we've seen everything from dedicated search engines to first prizes in art contests. However, it's also necessary to explore the gray area behind the generators: millions of illustrations, paintings and photographs that were collected without the corresponding permission or authorization. The Have I Been Trained portal aims to help users and artists alike to locate their works and photographs within training models and, if possible, request their removal.
"Nanos gigantum humeris insidentes". In its modern version, standing on the shoulders of giants. The phrase points to the discovery of truths inspired by previous works and discoveries, but we can apply it to many other things, including artificial intelligences. An algorithm is as good as its training, and if we focus on projects at the level of DALL-E 2, Stable Diffusion and Midjourney, that training is based on billions of images... made by someone else.
Of course, images are not enough: metadata is essential so that the prompts mechanism doesn't produce absurd results, but the ethical debate ahead is immense, especially with Midjourney and Stable Diffusion selling "credits" for the use of their online versions. The proposal by Mat Dryhurst, Holly Herndon, Jordan Meyer and P_Hoep is to give artists an option with the Have I Been Trained? tool.
Have I Been Trained? Locate your art in AI training models
Basically, the idea is that artists can enter specific parameters or perform a reverse search by uploading images to detect if their work is part of the LAION-5B set. At the same time, users can perform the same search applied to photos or personal works that have been shared on different platforms.
If the result is positive, the next step is to join the Spawning project, currently in beta, to develop a standard that enables artists to define an opt-in/opt-out in training models, specify permissions or limits when adopting their style, and offer their own models to the public.
How well will this work? Honestly we don't know, but one thing is certain: the collection of public material will continue. The concept of scraping is legal in the United States (for now), and copyright laws around the world have entire articles dedicated to "derivative works". Any tool that helps resolve conflicts is welcome, and we hope that both Have I Been Trained? and Spawning continue to advance.
(Editor's note: The site uses NSFW filters, but does not manage to block everything. Be careful.)
Official site: Click here
