What is the world like for computers?

We know that computers can look at the world and determine what they see using software that recognizes images. They can perceive objects and people and even distinguish between individuals and emotions. But what do they actually see?

This question was raised by researchers of artificial intelligence and pattern recognition. Scientists from the University of Wyoming and Cornell University have found that computers see the world completely different than we do. Perhaps this is not surprising.

The team began working with the latest system of image recognition algorithms, built on the basis of the so-called deep neural network (GNS). After reviewing the visual data of millions of images, the STS learned to distinguish a dog from a dolphin.

Then the team combined this process with another genetic algorithm that creates new images from the old ones. Using a technique that is sometimes called evolutionary art, the genetic algorithm turns the original image into a new one under the supervision of a person. Starting with any image, for example, depicting dolphins, the algorithm creates something relatively "dolphin". However, scientists replaced the choice of the image of a man to choose by the SOT system and came to strange results.

"We thought that we would get something similar, a lot of high-quality recognizable images," scientist Jeff Klyun told the New Scientist resource. "But instead we got pretty strange pictures: a cheetah that has nothing to do with a cheetah."

In the process, images that are similar to abstract art or visual white noise, but which the STS with 99 percent probability defines as specific objects - a ball or an electric guitar, for example. And although this is a certain optical illusion for a computer, it can create certain problems for face recognition systems.

The article is based on materials https://hi-news.ru/computers/na-chto-poxozh-mir-dlya-kompyuterov.html.