Whenever artificial intelligence has been used to create works of art, a human artist has exercised a significant element of control over creative processes. But what if a machine was programmed to create art on its own, with little or no human participation? What if it were the main creative force of the process? And to whom would the work be attributed, if it were to create something novel, attractive and moving?
At Art Laboratory and IA from Rutgers we have created a program that could be considered an almost autonomous artist, who has learned the current styles and aesthetics and manages to generate innovative images on his own. People really like the work of AICAN, and they can not distinguish it from that of human artists. His works have been exhibited all over the world, and recently one of them was even sold at auction for 16,000 dollars.
The emphasis on novelty
When we design the algorithm, we adhere to the theory proposed by the psychologist Colin Martindale. This raised the hypothesis that many artists try to make their works attractive by rejecting the forms, themes and styles to which the public is accustomed. Artists seem to intuitively understand that they are more likely to arouse the curiosity of viewers and capture their attention by doing something new. In other words, the novelty is the queen.
In the algorithm have been introduced 80,000 images representing the Western artistic canon of the last five centuries
So, when programming AICAN, we use an algorithm called "contradictory creative network", Which forces the program to face two opposing forces. At one extreme, try to understand the aesthetics of existing works of art. In the other, you will be penalized if, by creating your own work, you closely emulate an established style. At the same time, AICAN assumes what Martindale calls the beginning "Of the minimum effort" in which he argues that too much Novelty will annoy the viewers.
This guarantees that the art generated will be novel, but it will not depart much from what is considered acceptable. Ideally, create something new but based on what already exists.
Regarding our function, we do not select specific images to "teach" AICAN a specific aesthetic or style, as do many artists who create art through AI. What we have done, however, has been to introduce into the algorithm 80,000 images that represent the Western artistic canon of the last five centuries. It is as if an artist took an introductory course in the history of art, without focusing specifically on any style or genre.
With the press of a key, the machine can create a printable image. The works often surprise us because of their breadth, complexity and variation. AICAN can judge how creative each of his works isusing our previous work on the quantification of creativity. Since he has also learned the titles used in the past by artists and art historians, the algorithm can even name the works he generates. He called her one Orgy, and to another The beach of Pourville.
The algorithm prefers to generate more abstract than figurative works. Our research on how the machine manages to understand the evolution of art history I could help explain it. Having been tasked with the task of creating something new, it is likely that AICAN will be based on more recent trends in the history of art, such as abstract art, which became fashionable in the 20th century.
Do humans distinguish the difference?
There was also the question of how people would respond to the work of AICAN. To prove this, we showed different people images of AICAN and works created by human artists exhibited in Art Basel, an annual fair where contemporary art of the avant-garde is exhibited. We asked the participants if each of them was made by a machine or by an artist.
We discovered that humans did not appreciate the difference: in 75% of the cases, they thought that the images generated by AICAN had been created by a human artist. It was not simply that it was hard for them to distinguish between the two types. They really enjoyed the art generated by computer, and used words like "inspirational", "communicative" or "has visual structure", to describe the work of AICAN.
In October of 2017 we began to exhibit the work of AICAN in halls of Frankfurt, Los Angeles, New York and San Francisco, with a different set of images for each exhibition. In the exhibitions we heard again and again the same question: who is the artist? I, as a scientist, created the algorithm, but I have no control over what the machine generates. She chooses the style, the theme, the composition, the colors and the texture. And yes, I established the framework, but the algorithm is the one that decides the elements and the principles of the art it generates.
For this reason, in all exhibitions in which art has been exhibited, I have attributed exclusively the merit of each work to the algorithm, AICAN. In the Art Basel that will be organized in Miami in December, eight works will be exhibited, also attributed to AICAN.
The first work of art in the AICAN collection offered for sale, which AICAN entitled Saint George killing the dragon, sold for $ 16,000 at an auction held in November 2017 in New York. (Most of the proceeds went to fund research at Rutgers and the Institute for Higher Scientific Studies in France).
What the computer can not do
Even so, in the artistic process of AICAN something is missing. The algorithm may create attractive images. But he lives in an isolated creative space, devoid of social context. Human artists, on the other hand, are inspired by people, places and politics. They make art to tell stories and make sense of the world. AICAN lacks all that. However, it can generate works that human commissioners can then take root in our society and connect with what happens around us. That is exactly what we did with the Alternative facts: the multiple faces of falsehood, the title that we gave to a series of portraits generated by AICAN that impressed us by its opportune coincidence.
Naturally, the mere fact that machines can produce art in an almost autonomous way does not mean that they are going to replace artists. It simply means that artists will have at their disposal a new creative tool, capable of even collaborating with them.
I often compare the art created through AI with photography. When this was invented, at the beginning of the 19th century, it was not considered art; After all, a machine did a lot of the work. The creators of tendencies resisted, but in the end they gave in: after a century, photography became an established genre of the fine arts. Nowadays, the photographs are exhibited in museums and auctioned at astronomical prices. I have no doubt that the art produced by artificial intelligence will travel the same path.
Ahmed Elgammal is a professor of Computer Vision, Rutgers University.
Disclosure clause. The income obtained from the sale of works generated by AICAN has helped to finance the work of the Art Laboratory and the Rutgers IA.