AI-based surveillance is quite a thing, and the tech is being adopted at high speeds. However, not everyone loves the idea of being monitored in public, and a new trick could help such people from being detected.
A group of engineers from the University of KU Leuven in Belgium has come up with a new trick which involves the use of colorful patches to fool the AI surveillance systems.
The AI technology which is made to detect people in public won’t see the people who have used the patches, making them invisible to the technology.
It is further suggested that if the specially-designed patches are printed on a T-shirt, people can get away from the AI surveillance with ease.
While the new method of befooling AI sounds pretty interesting, it is not a foolproof plan to fool around with AI as the colored patches work with just one algorithm called YOLOv2.
Therefore, the whole “fooling” system needs to be improved in order to dupe the AI surveillance system.
As a reminder, this is not the first trick where an AI system was befooled. With the use of adversarial examples with similar patterns, Tesla’s autopilot mode was also tricked into driving in the wrong lane which involved the use of colored patches.
In addition to this, even facial recognition on smartphones can easily be duped with the use of 3D masks or 3D fingerprints.