Researchers have developed a system of creative artificial intelligence that can invent new styles of painting. The system – which is still under preparation – has already produced works that have sparked public interest, thus opening the way for the new form of artists that will escape from the narrow frames of the flesh.
The aim of researchers at the American universities of Rutgers and South Carolina, as well as Facebook’s artificial intelligence workshop, is the automatic creation of computer art that is “groundbreaking but not overly innovative”. For this purpose, an algorithm is used that uses two neural networks, one of which downloads ideas and the other evaluates them.
One network creates images, and the other network – trained through the analysis of 81,500 different artworks from classical and rococo to cubist and surrealistic – judges what from the ones proposed by the first network can be considered a work of art and what not. The important thing is that the second network has the “wisdom” to recognize an original work of art, even if it differs from anything else in the past, being something unique.
In this way, and through continuous feedback, an innovative work of art emerges, but not so innovative, that it is considered out of place and time. “We want something really creative and impressive, but at the same time that it is not extreme and not pleasant aesthetically,” said researcher Ahmed Elegamal of Rutgers.
After the artificial intelligence system created a series of paintings, the researchers asked volunteers to evaluate them together with paintings of real painters – without knowing which works are human and which not.
Surprisingly, researchers, not only was the audience unable to discern which painting came from a human being and which did not, but many artificial intelligence artifacts gained slightly higher scores than human, complexity, innovation, inspiration, mental uplift, and other criteria.
Already there are artificial intelligence systems that can mimic the style of famous painters, and even relevant mobile applications like DeepArt are available. But the new system does not modify a pre-existing work of art, but it creates a completely new one from the beginning. Somehow, the galleries of the future will probably be unrecognizable.