Lately my deep learning focus has been on understanding artificial intelligence — not just how it works in terms of training data, random forests, etc., but its legal, ethics, and policy implications. Hence, I’m not terribly surprised a lawsuit’s been launched related to the use of artists’ work in training AI algorithms.
In short, a group of artists have sued Stability AI, Midjourney, and DeviantArt, claiming their work was used without their permission to train generative AI systems in a manner that infringed on their copyrights. The Verge has some coverage of the case:
At risk of spinning up the outrage machine, there are some questions worth posing and which I’m guessing will play out in court. For example, if you accept the premise that every artistic endeavor derives or is at least influenced by other things, how much of any creative work is truly original? My wife, who’s an educator, writer, and photographer, has a great line — that one’s originality is a condition of the obscurity of one’s sources. So where does the line get drawn?
If AI algorithms are trained on databases of existing work, how does that differ from artists who take influences from or are inspired by other works of art? This opens up a fair use argument. Music is an obvious example, where there is a rich tradition of mixing beats from other artists. On the other hand, is it different if existing works are scanned and remixed by an AI algorithm as opposed to a human being? Regardless of your position on the subject, this argument is going to be interesting to watch.