Kenne Michael 
Researchers at the University of Chicago have developed a tool enabling digital artists to “poison” their work to prevent developers from training artificial intelligence (AI) systems on it.
The tool, named “Nightshade,” after the family of plants whose members are known for their poisonous berries, modifies images in such a way that their integration contaminates the data sets used to train artificial intelligence with false information.
According to MIT’s Technology Review, Nightshade modifies the pixels of a digital image to mislead an artificial intelligence system.
For example, Tech Review mentions convincing an artificial intelligence that an image of a cat is actually a dog and vice versa.
Theoretically, the AI’s capacity to generate accurate and sensible outputs would be compromised.
If a user requested an image of a “cat” from a tainted AI, they might receive a dog labeled as a cat or an amalgamation of all the “cats” in the AI’s training set, including those that are actually transformed images of dogs.
Vitaly Shmatikov, a Cornell University professor who reviewed the work, opined that researchers “don’t yet know of robust defenses against these attacks” — implying that even robust models such as OpenAI’s ChatGPT could be vulnerable.
Ben Zhao, a professor at the University of Chicago, manages the Nightshade research team.
The new utility is an expansion of their existing artist protection software known as Glaze.
In their previous work, they devised a technique for artists to obscure, or “glaze,” their artistic style.
For instance, an artist who created a charcoal portrait could be glazed to appear as modern art to an AI system.
According to Technology Review, Nightshade will eventually be integrated into Glaze, which is presently available for free download or web use.
