Just change 1 pixel to deceive AI to identify dogs into cars

Understanding artificial intelligence has long been a very popular topic for researchers.Recently, by changing a pixel of an image, one has fooled the AI ​​in identifying and classifying images.

As an example here, the AI ​​was deceived and misidentified the dog instead of an airplane, a frog instead of a truck .

Understanding artificial intelligence has long been a very popular topic for researchers, and many studies have tricked AI. In a recent test by Kyushu University research team, it was shown that AI is very easy to deceive in classifying images.

A new algorithm has been introduced on arVix, in which replacing only one pixel in a 1024 Pixel resolution image can make the AI ​​recognize the wrong image. The experiments have fooled AI over 74% of the total number of tests, this number also increased to 87% higher with 5 pixels edited.

In fact, 1024 Pixel photos are very small, and we need to change more pixels in larger images to deceive AI. Being able to fool AI with just a few small changes also raises fears about attacks. It is hard to improve AI, because we still have not fully mastered the technologies that make artificial intelligence.

1. What is artificial intelligence? What is AI (artificial intelligence)?