Illustrators, photographers, sketch artists, and designers have been grappling with the intrusion of artificial intelligence (AI) into their creative space, but a team of researchers from the University of Chicago claims to have devised a new solution. Glaze, a program developed by these researchers and downloaded nearly a million times, hides an image behind a layer of inaccurate data to prevent it from feeding into machine learning models.
The challenge facing artists lies in the need to publish their work online to gain visibility and commissions. However, this inadvertently fuels the same AI systems that could undermine their careers. Glaze leverages the significant difference between how AI and humans perceive art.
Ben Zhao, a Computer Science professor at the University of Chicago, explains that AI interprets visual information differently from humans. The Glaze program, released in 2023, can alter an image in a way that deceives AI machine learning models by making minimal changes imperceptible to the human eye.
Glaze uses the artist’s own computer to re-render an image by adding visual noise. While this image might still contribute to AI learning models, the extracted data becomes inaccurate, according to Zhao.
This seemingly subtle distortion changes the way an AI model perceives the photo or artwork, capitalizing on a fundamental disparity between AI’s image interpretation and human perception. This discrepancy has existed for a decade and has been resistant to mitigation attempts, allowing attacks on these models to still be possible, Zhao points out.
The primary goal of Glaze is to protect an artist’s individual style. The research team designed the program in response to concerns raised by artists who feared that AI models might focus on their work.
As Zhao emphasizes, the goal is not perfection but to address the threat that artists face in the dilemma of losing their income or realizing that an AI model could replace them. Glaze steps into this void until legal protections or guidelines catch up with the technology.
While Glaze could trigger a back-and-forth between artists and AI developers, it’s not about battling AI. The tool’s effectiveness might lie in forcing AI developers to divert resources to decipher Glaze, ultimately achieving its purpose of safeguarding artists’ creative output.