In 2023, the conversation around AI's impact on creative industries has intensified, particularly regarding its potential to infringe on artists' rights. A notable development in this area is Nightshade, a tool designed to protect artists by introducing undetectable pixels into their work, effectively disrupting AI training data. This initiative aligns with ongoing lawsuits against major companies like OpenAI and Meta for copyright infringement.
Created by University of Chicago professor Ben Zhao and his team, Nightshade aims to restore control to artists over their creations. Currently undergoing peer review, the tool was tested on recent Stable Diffusion models and a custom-built AI. Nightshade functions like a "poison," altering how machine-learning models interpret prompts, resulting in unexpected outputs. For instance, a request for a handbag might be misinterpreted as a toaster, or a cat might appear instead of a dog.
Nightshade builds on the team's earlier tool, Glaze, which subtly modifies an artwork's pixels to make it appear different to AI systems. Artists can upload their work to Glaze and activate Nightshade to enhance protection.
By leveraging technologies like Nightshade, artists can encourage leading AI companies to properly seek permissions and compensate creators for their contributions. However, companies attempting to remove these protective measures will face the daunting task of identifying every piece of corrupted data. While Zhao acknowledges the potential for misuse of such tools, he emphasizes that causing significant harm would necessitate thousands of modified artworks.