If you create any kind of visual art in the twenty-first century, you have probably resigned yourself to being mimicked by AI. These intelligent bots scour the web and scoop up your style, ready to replicate it for anyone who types in the right description. But it doesn’t have to be this way. Now artists have their own software to fight back. Glaze is a tool developed in the University of Chicago SAND lab that fools the bots, allowing it to protect your art from AI mimicry.
AI Sees the World Differently
A human’s eyes and brain perceive the world much differently than a computer intelligence. Machines don’t see objects in pictures. Instead, the image is broken down into pixels.
A pair of neural networks work together to analyze what it sees. First it breaks the image into pixels and gives these labels. It uses these labels (funneled through a mathematical algorithm) to make a prediction about the image. A second neural network then judges the prediction against real-life examples.
Because AI examines images at a pixel level, this means it can be fooled.
Fooling AI
A 2014 team from Cornell University and the University of Wyoming figured out a way to “fool” AI. They generated a picture embedded with strong visual features associated with a different object. The AI thought it was seeing something else.
Now the software Glaze adds tiny changes to an artist’s work, so subtle they are hardly detectable by the human eye. But AI doesn’t look the way we do. At the pixel level, things look very different. The AI will pick up on these changes and think it is “seeing” something else. This protects the artist’s style from being copied by learning algorithms.
Not a Perfect Defense
There are, of course, some limitations to Glaze. It relies on manipulating the pixels of the artwork. If you have a minimal and clean style of art, there is less “noise” they can insert to fool the AI.
Additionally, AI is continually evolving and learning and this may only be a stopgap protection. Still, it offers artists a way to fight about against having their style scrubbed from the web and used by AI with no compensation.
Tim Kane
