Artificial Intelligence is playing an increasing role in every aspect of our daily lives. It's not easy to get away with the technology, therefore, AI-powered video systems are widely used to track our faces in stores, offices and other public places. In countries like China, AI also provides an extra layer of security in policing and Government surveillance. Here a question arises, in today's world, is it possible to dodge AI?
The answer is yes.
Who exposed the flaw in AI surveillance?
As reported by MIT Technology Review, researchers from Belgian University KU Leuven demonstrated that you can actually hide from an AI-powered video system with the assistance of a simple color printout.
The researchers from KU Leuven demonstrated on YoLo(v2), an open source-object recognition system, that the image they produced could actually hide a person from an AI-powered computer vision system. It simply means any crook could actually bypass AI surveillance by using adversarial patches.
Researcher's view about AI surveillance
One of the researchers Wiebe Van Ranst told MIT Technology review team that their work has proved AI-based software isn't as error-free as once thought to be.
By exploiting these flaws, one can hide from security cameras or dodge government scrutiny. He also told MIT Technology review that they are looking to develop a patch that could work on multiple detectors. Right now, they aren't aware which of the detector is currently in use within the Belgian government surveillance system.
How did they Exploit AI surveillance
The KU Leuven researchers exploited what is known as adversarial machine learning in which deep neural networks are trained to recognize different things by feeding in different variables and tweaking parameters. By feeding manipulated examples and monitoring the output, it is possible to figure out what type of images confuse the system.
Experts opinion about the error detection
This work is being considered groundbreaking by the experts as AI is increasingly used in surveillance software and cameras. Stores like Amazon even used AI video system to eliminate checkout lines. Adding to its importance, China is catching criminals and tracking certain ethnic groups with assistance provided by Artificial Intelligence. This type of error could prove fatal and cause a big dent in the Chinese Government's reputation around the world.
In certain circumstances, this research has raised a lot of questions regarding the use of AI. Surely, there is a need to remove this and other errors from surveillance software. Effective steps must be taken in order to restore consumer confidence in AI technology.