DoNotPay introduces a new feature for confusing image recognition: making AI unable to recognize users' photos

Donotpay, a robot legal application, has just launched a new feature called photo ninja, which aims to make the artificial intelligence application unable to recognize the user's identity by slightly modifying the user's photos. Donotpay claims that the feature can prevent photos uploaded online from being used for malicious purposes, and that everything can be done automatically. For a monthly fee of $3, you can upload photos to donotpay and secretly incorporate information that can confuse facial recognition tools.

As a "counterexample", it uses special means to destroy AI behavior. With the continuous development of the role of artificial intelligence, as well as the continuous development of potential threats and privacy destruction technology, this field has been paid more and more attention by researchers.
"Photo Ninja uses a series of novel hidden writing, disturbance detection, visible overlay, and several other AI based enhancement processes to protect the user's image from reverse search, while not damaging the image appearance," donotpay said.
It is reported that the trained AI system can carry out pixel level data analysis on images. However, the appearance of adversarial examples can deceive them based on sufficiently subtle pixel color changes.
Although the human eye can't see the difference, the computer can't analyze it correctly as usual, so as to understand it as two completely different images.
For example, when submitting personal data photos to online dating platforms, users can use photo Ninja's preprocessing to prevent others from retrieving relevant pictures and social media information through tineye, Clearview AI and Google's image search function.