The emergence of artificial intelligence has always been fraught with questions about humanity and its relationship to what it might have to create. Stephen Hawking believed that creating artificial intelligence is dangerous for all and that the emergence of A.I. is not far off. However, many other issues exist despite the many dangers associated with A.I. One of them concerns art. Recently A.I. has become so advanced that we can create beautiful pictures with the help of an algorithm, and all we need to do is search. Many people aren’t surprised how advanced A.I. algorithms in search engines like Google have become, but the creation of paintings by artificial intelligence, which can be compared to human creativity, is a huge step that raises many questions. Can an artwork created by artificial intelligence be considered art? Who becomes the author of such art, the artificial intelligence or the programmer who created the code? And this raises questions about the authorship of art and the limits of intellectual property because what an artificial intelligence makes is a synthesis of everything that exists in the art based on many other works. Moreover, using the various services that allow the synthesis of artistic masterpieces in the public domain, one can choose to create a painting in a particular style, whether a Da Vinci sketch or a Van Gogh style (PiXimperfect). The framework is so blurred that today it is unclear the authorship of many paintings of this kind. There is a probability that while a human is figuring out whether the image belongs to the user who set specific parameters for creating a painting or the painting still belongs to the one who made the algorithm, it is possible that with the new level of A.I. development, the artificial intelligence itself will claim its ownership, however, for this, we need to understand whether the artificial intelligence can create fake art.
The Art of Artificial Intelligence
Today, artificial intelligence art is virtually indistinguishable from the hand of a real artist, but this was not always the case. Like many algorithms, the algorithms created to recreate paintings often had errors, which had to be refined when artificial intelligence got it wrong. It would seem easy enough to ask an algorithm to create something like portraits, but it may not analyze the data correctly. For example, if we tell the algorithm to picture dumbbells, the algorithm might mistakenly draw arms instead of barbells. This happens because the algorithm most likely encountered a picture of dumbbells held by people and could not distinguish between the dumbbell and the hands. This is why one of the most famous paintings created by artificial intelligence is so unimpressive. Although the painting has a sense of style, “Edmond de Belamy” does not have a clear outline of the nose and eyes. This is most likely how the algorithm perceives art because it was asked to draw its picture the way it imagines and interprets it. This is why A.I. art can be either imitation and recreate masterpieces by mixing authors, or complete abstraction, as the algorithm understands it. It is not surprising that a painting by Edmond de Belamy seems incomprehensible to us, yet for the A.I., it is probably exactly how the A.I. imagines image and a particular style, for example, artists of the 17th century (Manovich).
Nevertheless, it is then that the question also arises of how humans can evaluate art that the A.I. has created. Human participation is inalienable as long as humans are actively involved in creating A.I. paintings, including writing code and setting parameters. Even an image by Edmond de Belamy cannot belong only to artificial intelligence. After all, it was the developers who got the royalties for that painting, not the A.I. So until A.I. becomes independent, as Stephen Hawking predicted, A.I. art will not be something that belongs only to A.I. For now, human beings take an active part in what exactly A.I. should mean by painting, in what style to create, and in what images to select as examples. However, even then, many services offer not just one option but several to choose from, which can be supplemented and detailed (PiXimperfect). This would not be possible if A.I. were to create paintings independently, and it is hard to imagine that the artist would have someone helping them while making the image.
Nevertheless, today many believe such algorithms play the role of a human apprentice, the one who helps and essentially learns from a human. Of course, an apprentice can also create, but only under the guidance of one’s mentor, and that is why the merit is often attributed to the teacher until the student starts to make oneself. A.I. art is still just learning to draw and understand what art is. So far, the A.I. cannot understand what it is and why skill is necessary. Of course, a human would like A.I. to be able to create, as it is essential for general development as far as a human understands it. However, in the A.I.’s understanding, art may have a completely different idea or be considered meaningless. Humans and A.I. may have completely different perceptions of art, which means that at some point, art will become human, and A.I. art, which means that only other algorithms will be able to criticize such art and use it for their purposes. We can only speculate whether A.I.s will find valuable skills because humans have found helpful art for themselves, while A.I.s are part of ourselves and should have the same desires as humans. On the other hand, A.I. art will cease to be only a kind of art understood by humans to be an offshoot made possible by advances in technology. We should also not ignore that the creation of A.I. will have a global impact on all of humanity, but it will also affect how we perceive the world. Perhaps the way the A.I. perceives beauty will become an absolute revelation, and art will no longer be the same as before.
Can the creation of artificial intelligence be considered art?
The creation of A.I. raises many questions and, above all, apart from questions about authorship, the question arises whether A.I. art can be considered true art. After all, if we take the simplest definition of “art,” it implies that it is precisely what humans create. There are other more complex explanations, for example, that it is a figurative comprehension of reality and an understanding of the beauty of the world around and within oneself. However, so far, artificial intelligence can hardly define this, and most likely, if it can, that understanding will come from what has been programmed into the artificial intelligence by man. This close connection also prevents the A.I. from manifesting its individuality because, to create a picture, it is likely that material from other images must be downloaded into the A.I. The A.I. will read these paintings and use them to create their own. However, it can hardly be called art because everyone can copy; on the other hand, not everyone can create their style, like Van Gogh (PiXimperfect). An artist is not only capable of making one’s vision of the world but also of weaving one’s ideas and knowledge into the art and imitating an author.
Nevertheless, artists create their vision only if they do not want to reproduce someone else’s work. On the other hand, an A.I. can also design pieces that will be considered to be how the A.I. perceives the world and its views. However, if humans can sense everything around them, the A.I. is often programmed to perceive only what the programmer will allow. This is why most algorithms today only create paintings in a particular style. This can either be 17th-century art, which has many images that could be imitated, or modern art. We can put many paintings into the algorithm; the more varied they are, the more options will be offered. However, in general, the A.I. will create images in the same way, only if we do not specifically set it as a sketch of Leonardo da Vinci for example (PiXimperfect).
In addition, the A.I. will continue to be just a human tool, while it is not assumed that the device can create art. Considering what artworks have already been made by A.I., the portrait of Edmond de Belamy no longer seems so unusual. Perhaps the popularity of this portrait is just the result of the first such programming experience in art, and too much money was spent on this painting, although it later turned out that a completely different person initially prepared the algorithm itself. It cannot be said that this was not a genuinely momentous day because before, this man had not even imagined that such a thing was possible. Although the portrait was far from perfect, it was what society would have expected of an early A.I. Perhaps that is where the problem lies because it is now quite clear that A.I. is even capable in some ways of creating an image better and faster than the average artist. This can be seen in the example of the service midjorney.com (PiXimperfect). This service helps us develop pictures; all we have to do is enter the hashtags of what we want to see (PiXimperfect). In a few seconds, the A.I. generates four images from which we can choose the one we like or the closest to the idea. These images can be improved, or a new one can be generated, and the service is unlikely to offer the same thing every time a person develops an image. However, even if we specify a picture of an older woman walking up the stairs, the algorithm will likely suggest an abstract idea instead of a realistic one (PiXimperfect). It will be a pretty picture, which we wouldn’t expect from a machine algorithm. The painting will also depend primarily on how detailed a description of what a person would like to see from the algorithm, and here, a person’s imagination has almost no limits. The most striking thing is not so much that it is created literally in a few seconds, while a human would need much more time for this, but how diverse it is possible to make drawings. The service allows people to feel like an artist and do not need to put much effort into it. This is the essence of A.I., to make life easier for humans and to make many processes faster and more efficient. At the same time, a person does not need the A.I. to create itself, only for the sake of scientific interest maybe. However, first of all, it is a human who sets a pattern for A.I. to make, essentially directs the process, and rejects unsuccessful drafts, leaving only the best. A human being takes an active part in this, not only at the stage of developing the algorithm itself but also in the end result.
Edmond de Belamy and art from Midjourney
However, it is precisely because so many people are involved in the creation of A.I. painting that the problem of authorship arises. Not surprisingly, when the portrait of Edmond de Belamy was created, it turned out that a completely different person wrote the original code. Those who got the money at Christie’s auction only refined that algorithm, resulting in the famous portrait (Manovich). However, who owns the art itself in the end, the developer, the original developer, or the A.I.? The painting was sold as the first of its kind created by the A.I., and it was the developers who got the money. This suggests that all artwork created by algorithms belongs to the creators of those algorithms. Nevertheless, if the portrait has specific developers, then with the service midjourney.com, everything is entirely different. The developers allow users to create their masterpieces in exchange for a paid subscription, and the rights belong to the users.
With the portrait of Eduard de Bellami came an unpleasant situation that raised the question of authorship to a new level. While the A.I. still has no rights, anything the algorithms can create belongs to the developers. Nevertheless, it is known that the portrait was created based on other images, meaning that the algorithm copied the technique to make something of its own. Perhaps the algorithm could not have developed its picture at all if the developers had not put it in the A.I. Nevertheless, if humans are not allowed to use someone else’s property, then most algorithms need it to figure out what they need to do. The A.I. has no idea what a portrait is until several images are loaded into the algorithm. The A.I. cannot even figure out what a person is and why they need to be drawn. So the A.I. uses portraits that are already prepared. It is impossible to say precisely how much the A.I. borrows details or copies from other artists because even among artists, it is not always clear who is imitating whom and who is copying whom. This can only be understood if the artist blatantly tries to fake or reproduce someone else’s canvas, passing it off as one’s own.
Nevertheless, over time it becomes clear that the problem with the authorship of the painting by Édouard de Bellamy is because too much money was paid for the image. If the art from Midjourney had been worth that much, perhaps the developers would have wanted that kind of money, too (PiXimperfect). However, after a while, it became clear that the algorithm could produce too many paintings for them to be of any value. After all, the value of art is not only a measure of the artist’s skill and fame but also of how much work was put into it. A person must also create a concept, an idea, which would be the basis of this picture, so that if not the skill and labor, then at least the idea would be appreciated by others. This is especially obvious because in recent years, the scope of art has become increasingly blurred, and anything can be art, including simple strokes on canvas. However, if a human can come up with an idea behind these strokes, then an A.I. is incapable of doing so. The portrait of Edouard de Bellamy is probably one of the few that the A.I. has produced. That is why it is not easy to estimate such paintings’ value and authorship since, at different stages, the authorship can belong to other people.
Artist or programmer
Although most of the work in developing A.I. that can create art is done by programmers, it is not clear what contribution the A.I. itself makes. In addition, it is not enough that the programmer wants to create a robot artist because, to teach the A.I. how to draw and create art, it is necessary to have specific knowledge and skills. A programmer can hardly teach an A.I. to paint if the programmer oneself has never picked up a brush in one’s life. Even though most algorithms only need to read information from other paintings and take them as an example, the algorithm may not understand how to create its image. If the A.I. makes an image by copying all the existing styles, it is unlikely to produce something that can be sold at a Christie’s auction. It is considered much easier to copy than to create something new. That said, it is not surprising that one of the first A.I. paintings is a portrait, essentially a reproduction of a living person on canvas, a copy. However, the A.I. did not draw a picture from a live person. Edouard de Bellamy is not an existing persona that the developers themselves invented. Then, it is unclear exactly how the A.I. drew the portrait and what it was based on. To draw a picture, the A.I. not only had to look at all of the uploaded pictures but also understood that a picture is an image of a person. Paintings can be abstract, but the A.I. had to draw in a non-abstract style. That’s why the picture looks strange as if the programmers forced the A.I. to create a blurry portrait without a nose. After all, if the developers trained the A.I. how artists are usually trained, then the A.I. would know that certain principles draw a person’s face.
Nevertheless, it is possible that if the portrait were perfect, it would be impossible to accept it as a portrait painted by the A.I. In addition, the algorithms are far from perfect, and the developers of Edmond de Belamy are only slightly closer to being able to create something that can be attributed to A.I. Another thing is that services like Midjourney are already capable of avoiding such mistakes and the pictures that end up coming out are almost perfect. Even though it is unknown exactly how this program works, most likely, the programmers not only loaded pictures into the algorithm but also made sketches and drawings. If we look at a few pictures from this service, we can see that the images have some similarities. Even if we ask the A.I. to draw something that couldn’t exist, such as a sketch of a spaceship drawn by Leonardo da Vinci, the A.I. can reproduce it as if Da Vinci had drawn something like that (PiXimperfect). Either the programmer downloaded Leonard’s sketches for the A.I. to be able to produce similar issues or trained it to draw in that style. Either way, the developer’s contribution is quite significant. Many people think that creating software code like mid-journey is an art in itself because great creativity comes from the code. If we compare this to the way an artist creates one’s canvas, then, in essence, the program code, or algorithm, is nothing but the result of the work. However, an artist’s canvas would be considered art, while an algorithm that can create art would want to feel anything special. Of course, a programmer, like an artist, has received a specific reward for one’s efforts in recognition and fame, but an A.I., a robot, or an algorithm will not be considered art. Therefore, the artist has every right to what one creates, and there is someone to protect one’s rights. Whereas the programmer either has to admit that the A.I. is not yet independent and therefore has no right to what it creates, or acknowledge that the A.I. creates masterpieces itself and then anything ever produced by the A.I. the programmer will not have the right to take away from the A.I. While the programmer oneself needs to be an artist to develop such an algorithm, one will not be the artist who owns the authorship of what the A.I. draws. This is why the pictures made at the midjourney service do not belong to the A.I. or the developers but to the service users who pay to use the service. Nevertheless, we cannot say for sure that it is the A.I. that represents hell or heaven in a certain way and therefore draws such pictures (PiXimperfect). The A.I. only makes assumptions based on what has been downloaded. It is possible that this is how the developers represent heaven or hell, and it is not possible to understand what the A.I. itself thinks, nor is it possible that the authorship in such a case will not belong to the A.I.
Authorship in the Art of Artificial Intelligence
The authorship of the Art of Artificial Intelligence is still only disputed between people. The A.I. cannot assert rights because it does not yet have an opinion of its own. Perhaps very soon, the A.I. will pass the Turing test, and everything created by the A.I. will belong to them. For now, however, most of the rights to everything created by A.I. belong to the programmers. Nevertheless, many algorithms, like search algorithms, use the information available on the web for their own learning to create their picture in the end. Such algorithms are capable of analyzing much information in a short period. However, the A.I. copies other people’s pictures and developers cannot always track how much the A.I. copied and created its own. Ideally, developers should forbid the A.I. to use what many have rights to in reality.
In reality, however, the A.I. uses other people’s property to create its pictures based on it. Considering such paintings can cost a lot of money, many people might resent having their property used. This is likely why Edouard de Bellamy was painted when the algorithm was loaded with paintings from the last century whose authors are already dead (Todorovic). Perhaps images that did not belong to specific artists and had no ownership rights were chosen for this purpose. However, if we do not control the A.I., it can create a painting using different brands, violating property rights. To prevent this from happening, the A.I. must understand that this is not allowed, which means the programmer must include this in the algorithm. This could be prohibited using something like this or loading only verified, non-copyrighted information. In an era where almost everything has property rights, it is straightforward to fall into the trap. This is especially true for paintings. After all, it is much harder in the art to understand and appreciate an image’s value. History knows when many famous artists sold their paintings for nothing to survive, but their paintings became priceless after their death. However, if the A.I. cannot claim ownership of their pictures, the more so the A.I. will not understand what ownership is and why some images can be used while others cannot. The developer, as the one who holds the rights to everything the A.I. creates in the first place, should keep an eye on this. That said, an A.I. is perfectly capable of learning art on the Internet, just as ordinary artists do. No one can forbid an A.I. to take inspiration from other artists, copy their styles, or imitate someone else. However, as with different algorithms, people face the problem that when algorithms explore the Internet, they may see much personal information, and even if this is aimed at learning, does the A.I. have the right to follow people’s lives and use other people’s paintings? Also, the A.I. needs to understand what commercial brands are. However, as long as the developers can control it, they can also be sued.
Nevertheless, if the A.I. becomes independent, it may not obey the laws created for humans. After all, many regulations prescribe rules only for humans. When the A.I. becomes independent, new regulations will be needed and perhaps even different directions for the A.I. However, until then, the A.I. will be able to use the information without interference, which applies not only to pictures but also to many other things. The authorship of the A.I. will not be questioned, but only if the A.I. creates its masterpieces rather than repeating what humans have created. As long as the A.I. is at the mercy of humans, we are essentially allowing the A.I. to learn from us. Including the use of some paintings for learning that has no rights. However, perhaps in the future, the A.I. will be forbidden to use anything resembling human art.
The Limits of Intellectual Property
Nevertheless, the authorship of art was primarily created for humans. The fact that in the future, A.I. will be able to create their art was never envisioned; therefore, all art concepts mainly apply to humans, including authorship. Zelinger discusses in his book that treating A.I. as human and using all rights, including property rights, is wrong (Zelinger, 57). Even at the stage when A.I. is still under human control, most rights continue to belong to the developers. However, intellectual property is fundamental when it comes to the development of such algorithms. After all, in the case of Edouard de Bellamy, the developers only refined an algorithm that was already ready. This led to the fact that whoever initially worked on the algorithm that would have created the picture oneself got neither fame nor money. All because one did not protect one’s intellectual property in the form of A.I. So there may be other cases where the A.I. created will not belong to anyone because no one will claim ownership rights. Will the A.I. then become independent, or will the A.I. belong to anyone who finalizes the code? Or will the A.I. become the property of all humanity? This way, many problems with the right of ownership are solved. After all, the Mona Lisa does not belong specifically to the Louvre or France, even though it is located there. The painting belongs to all humanity. Despite the dangers, Stephen Hawking speaks of, the creation of an A.I. that passes the Turing test will mark a new era in human development. After all, A.I. will be capable of creativity and many other things to which humans have not yet been able to find solutions. Therefore, it is likely that A.I. will also become the public domain.
Nevertheless, only a group of people, who do it with a particular purpose, participate in the creation of the A.I. Even creating a special algorithm that creates paintings only makes sense if it pays off. If the Mona Lisa was written a long time ago and its author is long dead, as well as there are no relatives to claim it, the developers of the A.I. are likely to be alive and can claim the rights. To train the A.I., developers use all the intellectual property on the Internet, and it is hardly possible to trace whether the A.I. uses something people have rights to. However, when the A.I. is ready, everything it will create belongs only to the developers, not the world. Of course, a significant contribution to the development of A.I. is indeed made by programmers, yet can we then compare such art to the art of people who spend more time on it than a machine? People will always wonder if it’s fair that A.I., with no effort, creates masterpieces that bring in big profits, while humans themselves have to try much longer to make something like that. So there is a rivalry between humans and A.I., which is why Stephen Hawking was so concerned. After all, A.I. is already capable of doing things that we would not be able to do. Creativity separates us from machines because we put the concept into creativity, work on it, and do not apply logic. Most of the time, a device is not capable of this because it thinks logically. Even if an A.I. becomes independent, a lot of its mind will be fixed by logic, making them different from humans. Of course, not everyone is capable of creating a masterpiece, but we can all appreciate something beautiful and admire those who can create it. We can let the A.I. be like us, but will the A.I. understand what is beautiful? Most likely, these concepts will be different for humans and A.I.s. So when an A.I. also tries to create art, humans feel something is being taken away from them. After all, why should artists create when an A.I. can do it better and faster? That’s why humans protect their intellectual property from A.I. so that A.I. doesn’t end up surpassing humans in everything.
The art of Artificial Intelligence is a whole new level of creativity, which is quite unlike what humans consider a masterpiece. Even at the stage of creating an algorithm for the A.I., programmers become co-owners of what the A.I. can create, be it a portrait or a drawing. The programmers teach the A.I. how to develop and what it is. At the stage where the A.I. is submissive to the programmers, they can quickly indoctrinate the A.I. that painting is indeed something without which it is not easy to imagine life. However, this will be the human idea of art, not the A.I.’s opinion. The same thing happens with A.I. authorship. While the A.I. is not independent and cannot think in such a way as to form its ideas, it can only gather information and learn from what already exists. Thus, the A.I. can only learn from humans and all that humans have created. When a person wants to know how an artificial intelligence imagines this or that concept in the various services that claim that only A.I. creates a drawing on demand, this question is most likely asked of programmers, not of the A.I. itself. After all, programmers teach A.I. to perceive the world as a human does. However, A.I. does not have its perception of the world, nor does it have the ability to form ideas and concepts that could form the basis of art. That is why A.I. art cannot be considered to be that art in the human perception, especially it cannot be said that the art created by A.I. belongs to it because today, A.I. is intellectual property, which belongs to those who made it and everything that makes intellectual property automatically belongs to the developers. However, this does not allow the A.I. to use someone else’s intellectual property, especially when creating paintings, which can eventually be auctioned off for big money. Today, the problems are not related to the fact that what the A.I. creates does not belong to it, but at what stage the developed algorithm passes from one owner to another, and in the end, it is the final owner who decides how to dispose of this algorithm. Now, A.I. is a human apprentice, a tool with which a person can carry out their fantasies. It allows humans to realize that perhaps what will be created with the help of humans can become better than humans in everything, which will inevitably lead to conflict. However, the intellectual property problem and A.I.’s authorship remain relevant now, even though while A.I. has not yet become independent, laws about it have yet to be created.
Manovich, Lev. Computer vision, human senses, and language of art. AI & SOCIETY, vol. 11(2020), https://doi: 10.1007/s00146-020-01094-9.
PiXimperfect. “End of An Artist’s Career? A.I. Creates Art for You! – Complete Guide”. YouTube, https://www.youtube.com/watch?v=704brywiyfw. Accessed 9 August 2022.
Todorovic, Vladimir. Reimagining life (forms) with generative and bio art. AI & SOCIETY, vol. 3(2020), https://doi: 10.1007/s00146-020-00937-9.
Zeilinger, M. Tactical Entanglements: AI Art, Creative Agency, and the Limits of Intellectual Property; Meson Press, 2021.