In October,
Christie's auction house is going to sell for the first time a work of art produced by artificial intelligence (AI) - shortly after the first exhibition of works created by AI, held in the Nature Morte Gallery in New Delhi. And although the market vividly awaits the purchase, the question arises about ownership, obsolescence and work in the world of art, which the algorithm cannot cope with.
What is art?
Mario Klingemann, "Chicken or Meat?"
Many creators of AI use
generative-competitive networks (GSS), a technology that allows a computer to study a library of images or sounds, draw independent conclusions about what was studied, check them on original materials, and then try again, gradually improving the result by trial and error.
The resulting work of art between two artificial neural networks - which can be a drawing, video, multimedia installation - often turns out to be disturbingly vital, some kind of flora and fauna from the supernatural world.
')
For example, Mario Klingemann, who works in Munich, trained the algorithm on the portraits of the “
old masters ”, and then set him on the video from the webcam, where he himself filmed. The result was a video from a flowing multi-eyed grotesque, which is often compared to the works of
Francis Bacon .
Selling art from AI
Memo Actin, "Deep Meditations"
London artist of Turkish origin Memo Akten was one of the first artists who sold the picture, painted by AI, having received $ 8,000 for it at a charity auction held by Google in San Francisco in 2016. Two years later, Christie's is preparing to sell at auction his first work AI: a picture of the Parisian team Obvious entitled “Portrait of Edmond Belaimi”, for which they plan to earn from $ 8,000 to $ 11,500.
In a sense, the art of AI (IRS) is similar to any other emerging form of art, trying to occupy its niche in the market. Aparajita Jane, director of Nature Morte, says she put down price tags for works from the recent Gradual Descent exhibition “pretty aggressively,” from $ 500 to $ 40,000 to help III get a new genre. This is noticeably less than the usual gallery prices, from $ 10,000 to $ 100,000.
One piece, sold in Nature Morte, was created by artist Tom White of Wellington, creating abstract paintings in
Kandinsky style, using AI for this, representing everyday things like binoculars or fans. Jane says that the exhibition has attracted a new audience, which suggests that III can help the market grow beyond the dominant audience consisting of financiers and real estate dealers.
“I saw a lot of atypical art collectors buying my work — including scientists, video game creators, computer vision researchers, and AI,” says White.
What belongs to whom?
Tom White, "The Electric Fan"
In the press about the exhibition "Gradual Descent", a representative of Nature Morte Gallery wrote that these works were "completely created by AI together with artists." Obvious even sign a job with a mathematical equation for the algorithm used instead of its name. But no matter how much the artists and gallery owners like to attribute the authorship of AI paintings and emphasize that it is impossible to predict what the algorithm will produce, from a legal point of view there is no doubt who the owner of the final work is - the AI ​​or the human artist.
AI is simply a tool used by artists, just as a photographer uses a camera or Adobe Photoshop to create images, says Jessica Fjeld, assistant director of the Cyber ​​Security Clinic at Harvard Law School. “People are very deeply embedded in all aspects of creating and training modern AI technologies, and this will be the case tomorrow and in the foreseeable future,” says Fjeld. “As for me, it is much more interesting which of these people will receive the rights to the results of work, and not whether the software has the right to own it,” she adds.
Fjeld and her research partner, Mason Kortz, identify four key elements that make up the IRS, each of which involves copyrights in different ways. These are 1) the input data, 2) the learning algorithm, 3) the trained algorithm, and 4) the results of the work.
All the works of art mentioned in the article are sold as finished works - printouts, videos and installations. A person who has attempted to copy these works and resell them will violate the copyrights of the human artist, just as if they tried to reproduce an oil painting or a photo. But AI creates several new difficulties.
Code ownership
Harshit Agraval, “Anatomy Lesson from Dr. Algorithm”
Although most of the sources are created using open resources, such as Google TensorFlow and Facebook Torch, Fjeld says that artists who create their own algorithms, like White, have rights to them.
“An artist could sell the code as his work, although I did not hear that this happened,” she says. This is an interesting idea that collectors may like - they can then use the AI ​​artist to create their own, never seen before works.
However, retaining the ability to run code in the form in which it was supposed to be done - especially when it interacts with proprietary software or hardware - can be difficult.
“One of the problems with maintaining code is that software platforms update very quickly, and trained neural network models become redundant over time,” says Harshit Agraval, an artist participating in the Gradual Descent project, living and working in Bangalore.
Actin is particularly worried about work integrating web technologies - "things like Google Translate, either sending a request to Microsoft's cloud-based API that recognizes faces, or using Amazon Cloud services, or even works living in an already-discontinued Vine."
“I already know a fairly large number of works that have“ died ”due to changes in the cloud API or its disappearance,” he says. The solution may be an attempt to treat the work of AI as a performance. “They work until the technology allows it, and then they end. And we still have documentation and memories. ”
Possession of a training set
Anna Ridler, Untitled
Many artists involved in IIIR, train their algorithms on images or audio recordings from the public domain. Popular examples of such libraries are ImageNet, SoundNet and Google Art. One of the reasons is that using copyrighted images as a training set can produce results that look too much like a certain image. “I am not aware of any lawsuits in this area, but I think that sooner or later we will face them,” says Fjeld.
In theory, as the curator of the “Gradual Descent” exhibition, Karthik Kalyanaraman, says that III is not copying images or audio recordings on their own, means that they can learn with impunity on copyrighted images - just like art schools study textbooks and trips to the
New York Museum of Modern Art . “
Fair use ” is one of the protection strategies that artists can use in court if their training set contains copyrighted materials. But at the same time “if we approach the matter carefully, then I, from a pragmatic point of view, insisted that the images from the training set (for the works presented at the exhibition) were not protected by copyright,” he says.
Anna Ridler, another artist from the “Gradual Descent” exhibition, is even more careful about copyrights, using her own sketches and photographs for her training sets. “It’s the database set itself - what to include there, what not to include, becomes an act of creativity and part of the final work,” she says. “Since these databases are, in a sense, in themselves, works of art (I created them), so it will be almost impossible for another person to repeat my work,” says Ridler.
If the artist wants to use a proprietary algorithm or training kit, and the results of his work will obviously be derived from them, then he will most likely have to agree on the use with the rights holders.
III will change the entire art market
"Portrait of Edmond de Belami", a group of Obvious
III does not threaten the well-being of artists-people. The works of people who used AI belong to them - if for their creation people used open or self-created algorithms and training kits. But the emergence of III will lead to longer-term consequences for the art market.
Kalyanaraman believes that he has the potential to change art, not connected with AI, much like photography has changed the pictures, giving rise to impressionism, expressionism and other schools that are more interested in the expression of a person’s unique perception and emotions. He assumes that artists using AI can easily create new art forms, unexpected and provocative conceptual art, up to and including direct visualization of the description. Artists whose works look new, or can be described, and not felt, may face a decline in interest in their works, and collectors of similar works - with a fall in their value, just as realistic images do not intrigue people in the era of photography, photoshop and digital illustrations.
Kalyanaraman cites the example of Mark Rothko and Paul Klee, artists who, each in their own way, emphasize work experience as a relationship between two thinking beings - the first one does this by plunging the viewer into a tsunami of fire, and the second tickles his heels as artists whose works will live on.
“All our perception is tied to our emotions,” he says. And such a feature of the algorithm will be much more difficult to approximate.