University of Chicago Researchers Develop Tool to Counter AI Companies’ Impact on Online Art
Researchers from the University of Chicago have created a groundbreaking tool called Nightshade to empower online artists in their fight against AI companies. The software strategically inserts poisonous pixels into digital art to manipulate the interpretation of generative AIs. These toxic pixels can cause the models to learn incorrect associations, such as misidentifying a dog as a cat or a car as a cow. Nightshade takes advantage of the security vulnerability in models like Stable Diffusion, which collect vast amounts of training data from the internet.
During the testing phase, the researchers fed infected content to Stable Diffusion and instructed it to generate images of dogs. Interestingly, after being given 50 samples, the AI produced images of misshapen dogs with six legs. With 100 samples, the output began resembling a cat, and after 300 samples, the AI started generating full-fledged cat images. This experiment demonstrated the effectiveness of Nightshade in manipulating generative AIs.
The impact of Nightshade extends beyond just object recognition. It also affects tangentially related ideas and concepts, as generative AIs excel at connecting words and making associations. By introducing poison pixels, the software disrupts the connections between related concepts, such as different dog breeds or art styles.
While AI companies can potentially remove the toxic pixels, it is an arduous task due to the sheer volume of training data and the complexity of models. Removing corrupted samples requires manually identifying and deleting each one, a challenging process considering there are billions of data samples and typical images have millions of pixels.
Nightshade is currently in the early stages of development and has been submitted for peer review at the computer security conference Usenix. Professor Ben Y. Zhao, the team lead at the University of Chicago, revealed plans to make Nightshade available for public use through Glaze, another tool created by his team. Glaze allows artists to protect their personal style from being adopted by AI. Furthermore, Professor Zhao intends to make Nightshade open source, enabling others to create their own variations of the software.
When asked about expanding Nightshade’s application to video and literature, Professor Zhao pointed out the significant differences in those domains and stated that the team currently has no plans to tackle them. However, the emergence of Nightshade has been met with positive feedback, with suggestions that it could lead to AI developers showing more respect for artists’ rights and potentially even paying royalties.
As the field of digital art continues to grow, aspiring artists interested in illustration can explore TechRadar’s list of the best digital art and drawing software in 2023 for guidance and recommendations.
I have over 10 years of experience in the cryptocurrency industry and I have been on the list of the top authors on LinkedIn for the past 5 years.