In William Gibson’s novel Zero History, a key character dons the ugliest T-shirt in the world – a ridiculous-looking garment that magically renders the wearer invisible to CCTV.
Now, as states across the world deploy artificially intelligent surveillance systems to track, trace and monitor citizens, we may find ourselves wearing ugly T-shirts of our own. Researchers at Northeastern University, MIT and IBM have designed a top printed with a kaleidoscopic patch of colour that renders the wearer undetectable to AI. It’s part of a growing number of "adversarial examples" – physical objects designed to counteract the creep of digital surveillance.
“The adversarial T-shirt works on the neural networks used for object detection,” explains Xue Lin, an assistant professor of electrical and computer engineering at Northeastern, and co-author of a recent paper on the subject. Normally, a neural network recognises someone or something in an image, draws a "bounding box" around it, and assigns a label to that object.
By finding the boundary points of a neural network – the thresholds at which it decides whether something is an object or not – Lin and colleagues have been able to work backwards to create a design that can confuse the AI network’s classification and labelling system.
Looking specifically at two object-recognition neural networks commonly used for training purposes – YOLOv2 and Faster R-CNN – the team were able to identify the areas of the body where adding pixel noise could confuse the AI, and in effect turn the wearer invisible.
This isn’t the first time objects have been designed to try and trick artificial intelligence. In 2016, researchers from the US universities Carnegie Mellon and North Carolina at Chapel Hill created glasses that could fool facial recognition technology into misclassifying the wearer. In 2017, US researchers duped deep neural networks into thinking a stop sign was actually a 45mph speed limit sign, with a few subtle graffiti-like additions.
But all these previous adversarial attacks have been created on static materials. Doing it for video surveillance is much trickier. “For the physical attacks, the real challenge is to remain undetected during the whole video duration,” says Battista Biggio, assistant professor at the University of Cagliari, and creator of the first adversarial example, which was successful in fooling spam email detection. “When detection is running in every frame, remaining consistently undetected is much harder.”
Unlike a stop sign, T-shirts wrinkle and crease when the wearer moves, so the team had to take this into account. The T-shirt researchers are the first to succeed in creating an adversarial example designed to be printed onto a moving material. To do it, they used what Lin calls a “transformer” – a method for measuring the way a T-shirt moves, and then mapping that on to the design.
The researchers recorded a person walking while wearing a checkerboard pattern and tracked the corners of each of the board’s squares in order to accurately map out how it wrinkles when the person moves. Using this technique improved the ability to evade detection from 27 per cent to 63 per cent against YOLOv2, and from 11 per cent to 52 per cent against Faster R-CNN.
However, Lin says it’s unlikely that we’ll see these T-shirts in the real world. “We still have difficulties in making it work in the real world because there’s that strong assumption that we know everything about the detection algorithm,” she explains. “It’s not perfect, so there may be problems here or there.”
In fact, the researchers don’t actually want to help people evade surveillance technology at all. Instead, Lin says that the team’s ultimate goal is to find holes in neural networks so that surveillance firms can fix them, rather than to assist people in avoiding detection. “In the future, hopefully we can fix these problems, so the deep learning systems can’t be tricked.”
😓 How did coronavirus start and what happens next?
❓ The UK's job retention furlough scheme, explained
💲 Can Universal Basic Income help fight coronavirus?
🎲 Best video and board games for self-isolating couples
👉 Follow WIRED on Twitter, Instagram, Facebook and LinkedIn
This article was originally published by WIRED UK
