Neural networks have no similar art
ARTinvestment.RU   07 июля 2017

And not all spectators were able to distinguish the art of neural networks from the real pieces from the Art Basel fair

the Eternal battle of artificial and human intelligence came to the territory previously subject only to the people — territory of art and creativity. One thing — to calculate all possible combinations and algorithms in chess or to recognize a person on the basis of the millions of individuals, this machine has long been may, and often more successful than us. Another thing — act of creation, what-то fundamentally new, as it was considered all the past century, only the human genius. However, this sector already do not give rest to developers of neural networks.

the Last achievement of the team of Ahmed ElGamal from Rutgers University in new-Джерси (which, incidentally, keeps a magnificent collection of Soviet unofficial art of Norton Dodge) — works of art, created jointly by neural networks of two types. The neural network of the first type is able to recognize what style and direction particular work of art. She's trained to do it on the basis of 80 thousand works from WikiArt, which represents over a thousand artists from the XV to the XX century. There are about 13 thousand works of impressionism, 2 000 cubist paintings, 1000 paintings of the Early Renaissance, etc.

Neural network of the second type generates some images and shows them first, trained to recognize art neural network. The first network or finds their art to a certain style or not. By trial and error the second network learns to create images that fall under the category of "art" (again in the opinion of the person and another neural network).

But this is a team of ElGamal was not enough. Computer image in a certain style — is a repetition of already passed. Create a new style — that was the ultimate goal. Theorists of art believe that an artist who wants to make a breakthrough in his career, should be based on the world creative achievement, and then add to it that-то new. The work needs to include the effects of novelty, surprise, complexity, mystery and uncertainty, but they should not be excessive, otherwise you can cause antipathy in the viewer. About this, the programmers tried to teach their neural network — to do something new, not too drastically different from the old.

to test the reaction of audience on the neural networks generated image, group of subjects was asked to compare them with the works of abstract expressionists from the database WikiArt and works with the Art Basel fair in 2016. Naturally, people did not say what the picture is made by computer, and what — man. Subjects were asked, in-первых, to say whether they like the image in-вторых, to determine whether it's-то new, and-третьих and foremost, decide whom it was created — machine or person.

the test Results were interesting. To distinguish the work of abstract expressionists from the computer most are without problems. But the work with the Basel fair from the work of neural networks to distinguish some couldn't. Moreover, some works created on the computer, the audience rated higher than real paintings with Art Basel. They found themmore inspiring.

However, the creators of the neural network are not inclined to interpret the results of the test in the way that the computer was a talented artist. All the work of the neural network is built on an algorithm that takes into account what pleases the audience, causing an emotional reaction. With the same success in the art could be called a cat video.

Sources: technologyreview.com, artinvestment.ru

*All these images were created by neural networks



Индексы арт-рынка ARTIMX
Индекс
Дата
Знач.
Изм.
ARTIMX
13/07
1502.83
+4,31%
ARTIMX-RUS
13/07
1502.83
+4,31%
Show:

Top 18

Back to Top