AI spits out images of actual persons and copyrighted content.

When applied properly, the image generators that are so common these days also produce recognisable images of real individuals. It is also possible to make precise replicas of works protected by copyright. This was discovered by researchers from Princeton University, the University of California at Berkeley, ETH Zurich, and the Alphabet companies Google and DeepMind. The study highlights potential legal issues with DALL-E, Midjourney, Stable Diffusion, and Co. potentially violating artists' and businesses' copyrights as well as privacy protection laws. The initial criticisms of the latter point have already started.

By providing captions, such as B. the name of a person, to Stable Diffusion and Google's proprietary image generator Imagen, the researchers were able to acquire their findings. They next checked to see if any of the generated photos corresponded to the actual photographs stored in the model's database. With the training data for the AI, the team was able to extract more than 100 mock photos.

The large datasets of photos and their corresponding text descriptions pulled from the internet are used to train the image-generating AI models. The most recent version of these systems extract an image from the data set and alter it one pixel at a time until the original image is reduced to a collection of random pixels. After that, the AI model reverses the procedure and builds a new image from the pixelated chaos.

Read more of the article here.