After AlphaGo rolled the go world, AI was challenging the artist again-

Created by an artificial intelligence system called the CreativeAdversarial Networks (CAN). Picture: thanks to Rutgers University Art and Artificial Intelligence Laboratory
Imagine that a computer can create a piece of art, and the works of art looks and by artists, in the famous Art Fair Exhibition of the works of the same.
This is the art of Rutgers University (Rutgers University) and the Artificial Intelligence Laboratory (the Art and Artificial Intelligence Lab) new research team is trying to reach the goal. In June, they made a surprising discovery at the International Conference on computer creativity in Atlanta, USA.
The paper pointed out that: "since the dawn of the advent of artificial intelligence, scientists are exploring the use of machines to create such as poetry, stories, jokes, music and painting of human level creative products, are also looking for the ability of creative problem solving with the machine, (we research) concluded that humans can not distinguish between the machine the creation of works of art and appear in the top art expo, created by contemporary artists.. works of art. "
In 2015, two papers published in art and Artificial Intelligence Laboratory at Rutgers University and discussed one can identify the works of artists, genres and artistic style of the algorithm (basically can be used as a computer art historian), the new research is based on these two papers. In the past few decades, this algorithm has created new connections between different artists in different styles of art, and has led to a series of unexpected discoveries. In addition, it can be graded on art and found that Da Finch's best-known work, Monalisa, scored less on creativity than what he wrote. It was not a very famous work.
Picture: thanks to Rutgers University Art and Artificial Intelligence Laboratory
In this new research project, the laboratory has systematically improved the Generative Adversarial Networks (GAN). In the GAN model, the depth of the neural network can learn the existing style of painting, such as Baroque, Rococo, pointillism, color gamut and fauvism and expressionism painting style. One of these two adversarial networks creates pictures based on what they learn, and the other judges whether the results are up to standard.
The new version, known as the Creative Adversarial Networks (CAN), can create works that are different from any of the known art styles. According to the thesis, that is, maximizing deviations from the known art styles, and minimizing deviations from the art distribution patterns". To train the network, the team used 81449 pieces of art created by the 1119 artists open to the public in the Wikimedia gallery.
Without knowledge, the researchers asked participants to judge four sets of works, guessing that they were created by humans or created by computers:
Works created by CAN neural networks;
Works created by GAN neural networks;
Historical expressionism; the work of an expressionist artist;
And exhibited at the 2016 Basel Art Fair on non figurative works.
Picture: thanks to Rutgers University Art and Artificial Intelligence Laboratory
The results show that the subjects have the highest accuracy in judging Abstract Expressionist works, and 85% of them are judged by human artists. While 53% of CAN's works are regarded as works of human artists, GAN is only 35% mistaken for works of human artists. The most interesting is that only 41% participated in Basel art fair works are considered as human artists.
Ironically, when subjects were asked to give the conception of the work, the visual structure, convey inspiration and works of scoring, they think "the computer works were generally higher than human artist's work, whether it is the abstract expressionist works of art group and Basel art fair works group. "
Obviously, artificial intelligence can not completely replace the artist's work, but this new study shows that in the depth of the neural network in the world, they (Artificial Intelligence) may have the potential of artists.