Artists against artificial intelligence: whose take?

Recently, a group of artists filed a copyright infringement lawsuit against the creators of neural networks. The essence of the lawsuit is that, according to the plaintiffs, the owners of neural networks violated the copyrights of “millions of artists” by using five billion images taken from the Internet “without the knowledge and consent of the artists” to “train” the algorithms of their services, and are now enriched by images created by neural networks based on the studied paintings, thereby damaging the art. Moskovskaya Pravda found out from experts how promising and justified such claims are from artists, musicians, writers and other representatives of creative professions.

Lawyer, Patent Attorney, Head of the Intellectual Property Practice at Afonin, Bozhor and Partners Alexander Afonin:

– It seems that the claims of the authors of works against neural networks on the fact of violation of rights most likely will not be recognized as justified in the current legal regulation and will be satisfied by the courts. According to paragraphs. 4 p. 2 art. 1255 of the Civil Code of the Russian Federation, the author has a personal non-property right to the inviolability of the work. Also, the author or other copyright holder has the exclusive right to allow or prohibit the reproduction, processing of the work in accordance with paragraphs. 9 p. 2 art. 1270 of the Civil Code of the Russian Federation. At the same time, the processing of a work means the creation of a derivative work (processing, screen adaptation, arrangement, staging, etc.), processing. Since the neural network, as the creators usually say, only “learns” using the works of the authors, but does not copy the elements of the original works, then in essence we are talking about a situation similar to that when one artist is inspired or learns from the works of another, which certainly cannot be recognized. violation.

According to international principles, as well as, for example, paragraph 3 of Art. 1259 of the Civil Code of the Russian Federation, copyright extends to works expressed in any objective form, including written, oral (in the form of public utterance, public performance and other similar form), in the form of an image, in the form of sound or video recording, in volume-spatial form.

Copyright does not apply to ideas, drawing style, drawing methods, compositions, which can be concluded from the content of paragraph 5 of Art. 1259 of the Civil Code of the Russian Federation. If the neural network processes (retains the recognition of the original work) the works of artists, then there will be a violation, just as if one artist copies the work of another.

Ilya Karpuk, director of the Printeria service center, 3rd class lawyer, developer of neural networks:

– These lawsuits in the “western” legal field are not uncommon. In the wake of the hype of neural networks, there will now be more and more such lawsuits in an attempt to somehow make money on the trend.

Most often, the authors of such lawsuits do not understand the principles and mechanisms of neural networks at all, seeing the result and, perhaps, not deliberately distorting the facts. But it is the principles of “case law” in Western courts that make it possible to bring such claims and even often win them.

Undoubtedly, there will be copyright disputes. Although the neural network as a mechanism generates the result of its “works” completely from scratch, and each time it will be different, there can be similarities with the work of real authors of works, since neural networks need to be trained, and for this it is necessary to use a lot of test material.

Using the example of a neural network that generates graphic content according to the description, you can see that the training set was 5-7 billion images. These images are obtained in an open way, classified and described for specific requests. After the training is completed, the neural network generates a result upon request, for example, “a girl in a polka-dot dress on a Van Gogh-style longboard, present the result as a picture.” The results will be very interesting, because many subsystems are involved (highly simplified) – background, longboard, girl, dress, style of Van Gogh paintings. And the smaller the training set, the stronger the similarity with the work of any designers, artists, photographers. If the training set is quite extensive, it included dresses, different polka dots, and colors, then the result will be interesting, and one might even say unique. Indeed, unlike a person, the neural network does not copy any specific style emotions, it considers training “pictures” as a kind of mathematical equation and then itself produces completely interesting mathematical results – in the form of images, sounds or text.

Of course, if you try to conduct an examination of such works, you can conditionally “get to the bottom” of similarities with people, strokes of famous artists, but such lawsuits will not work in the Russian legal field. Examination for plagiarism must show at least 30% similarity with the original. And this directly comes from the quality of the training set of the neural network.

Global companies involved in the development of neural network programs have achieved tremendous results, many works created by artificial intelligence are indistinguishable from the works of artists. And neural networks can already write essays, do school homework … But so far this is only in English, and is not available to ordinary schoolchildren.

I consider such claims to be unjustified. But I want to note that observing the progress in the development of neural networks, you understand and start to get scared how smart and natural the result of the machine’s operation can become for us. Here it is rather necessary not to file lawsuits, but to establish trade unions of creative professions and negotiate with programmers. In an age of gigantic progress and machine learning, I think we will soon appreciate the creativity of people even more.

Lawyer Kristina Yankovskaya:

– Neural networks are trained on the basis of existing images, and there is a risk that the neural network can create an image based on another image that is already copyrighted. The essence of this case will be built around substantive evidence. If each artist who joined the lawsuit can prove that artificial intelligence did not just use some of his specific images for training, but took them as a basis for creating new ones (recycled without permission), then there is a chance that the court will satisfy individual requirements. By the way, not all artists have a negative attitude towards neural networks, there are those who use them as a tool to create their own unique works.

Artist Egor Zverev:

– At the moment, as far as I know, this is the first lawsuit on this topic, so it will not work to operate with a case of precedent. The situation is such that the whole world is following this trial, and further outcomes of disputes (judicial and extrajudicial) will already be based on this decision. The field of legal legislation in the field of neural networks has just begun to develop, so it is likely that new rules will soon be adopted that will affect the decisions of the courts.

At the moment, the main claim of the plaintiffs is that neural networks learn from their work without their consent. It certainly is. However, other artists are clearly doing the same thing (consciously or not). Anyone who makes art and does not exist in an information vacuum is inspired and draws ideas, including from the work of other artists.

As for my opinion, if the neural network does not explicitly borrow entire fragments and copy the author’s style of any artist, then it is difficult to present something to it. In addition, you still need to prove that the neural network used some of your work, and this is difficult, despite the fact that the code of many neural networks is freely available.

In parallel, the history of the assignment of copyright to the works of neural networks is being actively discussed. One of the developers’ ideas is to tie works that enter the network to an NFT token that will show their creator, including the neural network, if one was used.

By the way, there are already both neural networks that determine whether the material was created using neural networks, and neural networks that help bypass this examination.

It is quite possible that the outcome of this court session will be the forced blocking of the ability of the neural network to learn from the works of a particular artist. This may satisfy the demand of artists for copyright protection, and will not stifle the development of artificial intelligence.

* * *

It is difficult to disagree with the statement that creative people, like neural networks, are inspired by paintings created by other masters of the brush. At the same time, no one accuses them of copyright infringement. Another thing is that painting a picture with a neural network takes much less time than creating a canvas by a real painter. And here we can already talk more about what works will be more in demand – works of artificial intelligence created according to mathematical formulas or paintings painted by an artist inspired by the muse.

Alena Bodrienko.

Photo by Alexander Avilov / Moskva Agency