Creators Are Now Trying to ‘Poison’ AI Into Submission
The war against the machines is in full swing.
What started as general disdain for generative AI has now turned into a full-scale assault.
While artificial intelligence platforms continue to train themselves on artists’ images and styles without their consent, the humans are fighting back. More specifically, some new products have been launched that are meant to “poison” the data, therefore confusing the AI into making mistakes.
Nightshade is a free tool from researchers at the University of Chicago. It’s clearly something creators have been waiting for: Since its release in mid-January, more than 250,000 people have downloaded the software to combat AI models training on their work.
“The tool essentially works by turning AI against AI,” notes VentureBeat in an article on the topic.
As a primer, DALL-E and Midjourney AI work by scraping data from the web, including commercial art. Midjourney reportedly has trained on more than 100 million images without permission from the original makers.
