A GROUP of scientists recently invented an “invisibility cloak’, capable of tricking AI software from detecting the wearer.
To the naked eye, the cloak looks like an ugly sweater.
Designed in partnership between researchers from the University of Maryland and Facebook Artificial Intelligence, the sweater leaves the wearer untraceable in the face of artificial intelligence.
A report from the university explains that the sweater’s cluttered design is what breaks AI systems.
“This stylish sweater is a great way to stay warm this winter,” write the researchers.
“It features a waterproof microfleece lining, a modern cut, and anti-AI patterns — which will help hide from object detectors.”
A video of the sweater in action was shared by the researchers, and it shows how it functions in front of AI.
The clips show the software recognizing people in varying distances and positions yet being unable to pick up the sweater.
As the wearer walks toward the camera, the details of the sweater become visible.
It appears to be made up of an enlarged image of a crowded market, which likely confuses the AI.
Most read in Tech
“We load images from the COCO detection dataset, and pass them through a detector,” write the researchers.
“When a person is detected, and pattern is rendered over that person with random perspective, brightness, and contrast deformations.”
The sweater was initially designed to test the AI’s algorithm for flaws and can elude most object detectors.
Tricking AI detectors is not an easy feat.
These systems have a vast pool of resources that they use when spotting people and objects, making the invisibility sweater a significant achievement.
While the sweater is capable of tricking software, it’s still unable to make people completely invisible and has about a 50 percent success rate.