- Introduction
“Art is a human activity having for its purpose the transmission to others of the highest and best feelings to which men have risen.”[1] This is how Russian writer Leo Tolstoy, one of the greatest authors of all time, defines art in his essay What is Art?. The way Tolstoy feels about the connection between the creation of art and humanity is probably shared by many other people. It is a very common viewpoint to see art as something that is deeply human, serving to express feelings, opinions, experiences of one human, the artist, to others, the recipients. However, in recent years, artificial intelligence algorithms – having already successfully entered various other areas of our lives, ranging from education to healthcare – are gaining ground in this very area that is supposed to be at the core of what makes us human: the creation of art.
Naturally, this development gives rise to a remarkable amount of questions, for example: Can an algorithm truly be creative? Is it able to produce human-like art? And if so, what does this say about the human conception of art and creativity? Is art truly a „human activity“?
- Can machines be creative?
In order to approach an answer to these questions, one can start by identifying some of the different areas in which humans may produce something that is considered to be art by others. In a next step, one can examine exemplary works produced by algorithms in these artistic fields, which fit into human perceptions of art as well.
2.1 Visual arts
In 2018, the renowned auction house Christie’s in New York sold a painting made by an algorithm for the first time in their history, for $432.500. The painting, called Portrait of Edmond de Belamy, depicts a fictional man whose clothing gives the impression that he may have been an 18th-century clergyman. However, the painting differs somewhat from actual paintings from the 18th-century made by humans, as it features a more contemporary painting style. This „mixture“ of styles is due to the generating algorithm previously being trained on an astonishing number of 15.000 portraits from various periods of art history, ranging from the 14th to the 20th century. [2] The architecture behind this algorithm is a type of artificial neural network called Generative Adversarial Network (GAN), which consists of two networks trained with the same data set: the Generator, whose task it is to generate portraits and pass them on to the Discriminator, which is then supposed to decide whether said portrait is human-made or not. This way, those two networks help each other to get better and better, until the Generator is able to generate a portrait which is indistinguishable from one painted by a human to the Discriminator. [3] The painting that is eventually generated by the algorithm using this approach is not completely random, but merely a product of a new combination of features of paintings humans have used during the long centuries of human art history. One could argue that this is not so different from what a human artist does when they aim to create a new painting; they also draw from a long tradition of art. AI does a good job simulating this.
2.2 Sound art
AI algorithms have also already made their first attempts at musical composition, trying to walk in the footsteps of the greatest musical geniuses in history, like Bach or Beethoven. The people behind the AI Project Beethoven X made it their mission to finish Beethoven’s 10. symphony, which he himself had not been able to, due to a very human flaw: his mortality. However, the 40 drafts for his 10. symphony that he left offered an orientation and were used to finish what Beethoven could not do during his lifetime. The AI was trained in the context of the music of the 18. century, much like the young Beethoven himself. As one can imagine, the execution of this project proved to be extremely elaborate and took a computer more than 2 years to accomplish, with constant human intervention. Disappointingly, the reaction to the final result has been rather negative. What made this endeavor much more difficult than the generating of Portrait of Edmond de Belamy is that in this case the AI has to be able to maintain and develop a musical theme over several measures, in order to produce a wholesome musical composition. Apparently, it was unable to do so, as one example of criticism was: “Everything sounds conventional, small-scale – and only rarely really like Beethoven.“[4]. Similarly, the conductor Dirk Kaftan concluded after the world premier: „So, to make a long story short: This is no Beethoven.“[5]
2.3 Literary art
Another important artistic field in which humans have excelled for centuries is the field of literature. Exceedingly more powerful language models are being presented by researchers ever year, able to generate poetry or new chapters of Harry Potter. The most powerful language model at the moment is Open AI’s GPT-3, launched on June 11, 2020. Being the biggest language model of all time, it was trained on a massive amount of text, exhibiting over 175 billion machine learning parameters (before that, the largest language model in the world, introduced by Microsoft, contained „only“ 17 billion parameters). It is capable of performing a multitude of language-related jobs, including the generation of creative texts, like poems and novels. However, despite its impressive abilities, it still has a lot of shortcomings and can produce rather silly mistakes, because it does not have any understanding of the texts it is producing. It lacks an underlying semantic model of the world, and merely works on powerful statistical computations, based on the massive amount of data it has been trained on. Here, we are facing the same problem as with musical compositions; GPT-3 may be able to produce a paragraph that is logically sound, but when it is supposed to write a very long, coherent text, more and more mistakes arise. In order to create a lengthy, coherent written text, a human proof-reader is needed.[6]
2.4 Performance art
Also worth mentioning is the advance of AI in the context of performance art, using virtual avatars of artists in theaters or creating whole concert experiences within a virtual environment, with the help of virtual reality glasses. A recent example is the Swedish band ABBA, who have worked together with A/V wizards for several weeks in order to create virtual copies of themselves using motion capture techniques. An AI algorithm was then used to make the virtual performers look like their younger selves, who accurately dance, move and perform like the real band members. While this digital ABBA concert will take place in a physical location, other artists have already gone a step further in trying to make their performances a fully virtual experiences. [7]
- The question of aesthetic autonomy
After having examined the artistic work of algorithms in various fields, what becomes apparent is that as of yet, AI algorithms still need a lot of guidance and a lot of effort on the human side to be able to produce something we might call art. This leads to what many people would refer to as one of the defining features of humans and human artist: (aesthetic) autonomy. Human beings create art because they want to create art. They have intentions and inner emotional states which they can transfer onto others through their work [8]. In this sense, it is obvious that a machine lacks this feature, it can not decide to generate a painting, or finish Beethoven’s 10. symphony. Rather, it is always humans being responsible for and involved in every single step of development, whether it is coming up with the original algorithm, choosing and curating the learning data, or training the model. Thus, at least at the moment, human creativity is always the basis for anything we might call machine creativity. It would be presumptuous to claim that any AI is creating art on a whim which was not given to it by humans. This is also why the sensationalist tone of the media, announcing for example that a painting made by an AI has been sold, is misleading, because one could easily question the ownership of that painting; is it really a painting made by the algorithm or is it made by the people who created that algorithm in the first place? I would argue that it is in fact both.
- Human-machine interaction from an aesthetic point of view
As hinted at above, it would be most accurate to call the creations of art by machines we have seen so have a collaboration between humans and the machine. The machine itself would simply not create anything if it were not for the humans programming it to do so. However, it is important to note that this does not mean that the algorithm’s output is deterministic. With the rise of machine learning algorithms and deep artificial neural networks it has become possible for algorithms to „surprise“ their human creators with outputs which have not been intended or expected by them in advance. It is this unpredictability that can be quite useful when it comes to the human-machine-collaboration: Because humans tend to get stuck in their ways of approaching creative possibilities, which constraints them and blocks their view for things that could be interesting to explore in their art, the „creative“ machine can then serve as an auxiliary tool for the human artist, used to broaden their horizon and fulfill their own creative goals. It can offer incentives, provide ideas, show humans what is possible and thereby promote human creativity. This means that art creators and AI do not behave like subject and object, or subject and instrument, but rather they form some sort of „interobjective“ creative context.
- The aesthetic value of machine art and the anthropocentric concept of creativity
Some people may dismiss the artistic value of generated paintings or musical compositions. The machine does not possess any emotions of its own, so they argue that it simply can not transmit anything through their art, which makes the art „soulless“, „cold“ or „anemic“. However, just because a generated painting can not transmit feelings that belong to the algorithm, this does not necessarily mean that its art does not transmit anything. One could argue that all the intentions and emotions of the original human artists that went into the paintings are being transferred to the newly generated one because the algorithm has been trained on them. Moreover, it has been shown that often humans cannot actually tell that a work of art has not been produced by another human if they have not been told so in advance. That means they as the recipients project their own feelings onto the art and interpret it in a way that suits them anyway, so it does not matter that much if the „artist“ did have emotions or not. Because of those arguments, I do not think it is really justified to outright deny machine art its artistic value.
Underlying everything is, of course, the question of what is meant by art in the first place. So possibly the debate about AI and art could serve as an opportunity to question our human concepts of creativity and art production. Authors like Andreas Sudmann have pointed out that these concepts have a profoundly anthropocentric or anthropological logic, which we simply cannot seem to get rid of. [9] This means that we currently value such accomplishments of machines that we also value and admire in humans, whereas maybe what we should do is let AI create or do something, which we do not evaluate according to a current human concept of art. If art is a medium that conveys feelings and moods, experiences of life as a human being, then how could a machine ever achieve that? A machine has no body and can therefore simply not experience the world the way a human being can. And then again, even when it comes to artifacts made by humans, we are not necessarily in agreement on what to call art either. Sudmann also mentions that there exists this paradox concerning the usage of the terms art and creativity, which is reflected in the current debates around machine art as well: „On the one hand, they are used in such an inflationary manner that basically everything under the sun can be considered creative or art; on the other hand, there is a tendency to mystify and esoterically transfigure corresponding artifacts and activities.“[10]
Dieter Mersch discusses the problem of the concept of creativity being „chronically fuzzy“[11] as well. He furthermore criticizes the equalization of thinking and calculating in general, and the paradigm of algorithmic rationality in the context of art creation in particular. Machines beating humans in games like chess address areas that can very well be mathematized, whereas the concept of creativity, according to Mersch, can not. He claims that „ […] creativity does not form an erratic faculty nor a generative machine, but a principle of freedom.“[12] Here he refers again to the already mentioned aesthetic autonomy, which to program into a computer would be inherently contradictory, and is therefore impossible.
- Conclusion
In conclusion, due to the lack of artistic autonomy, AI cannot be described to be creative on its own. Humans are still involved in almost every step of the production of machine art. Some of the artworks that AI algorithms are currently able to produce can fool humans into thinking they are human-made and can give them a pleasurable artistic experience, which is why they definitely exhibit a certain artistic value and an aesthetic relevance. It is just that as of yet, art producing machines do not actually add any value themselves, but merely recombine what has already been done by humans. However, as AI is able to offer surprising results and new perspectives which can be very valuable to human artists and serve as an inspiration for their work, it can be integrated in human creative processes as an important component. As long as our conception of what creativity and art means stays deeply connected to the experience of being human – being a physical body in the world, capable of inner emotional states – art will stay the „human activity“ that Tolstoy claimed it is. But if we are willing to broaden our understanding of art, then there’s no reason why in the future this could not include „art“ made by machines as well, in whatever form this may be. Thinking a lot farther into the future, when it has been achieved that machines have developed a kind of consciousness of their own and are capable of being autonomous, then all these questions will have to be asked anew. Obviously it is currently questionable when and if this will be possible at all and if it is, how it will look like. But for me it is a very interesting and exciting thought to imagine that the art of these machines will then maybe serve a similar purpose as it does for us humans: giving others insights into how it feels to be a human or to be a machine, respectively.[13]
Autorin: Janina Bodendörfer
[1] Tolstoy, L.N. What is Art? In: The Novels and Other Works of Lyof N. Tolstoy. Translated by Aline Delano, New York: Charles Scribner’s Sons, 1902, pp. 328-527. Available at: http://www.gutenberg.org/ebooks/43409 (last access: 01.02.2022).
[2] cf. Sudmann, Andreas. Computerkreativität. Maschinelles Lernen und die Künste Künstlicher Intelligenzen. In: Bernhard J. Dotzler/Berkan Karpat (Eds.), Götzendämmerung – Kunst und Künstliche Intelligenz, Bielefeld: transcript Verlag, 2021, pp. 85-98.
[3] cf. Sudmann, 2021, p. 89.
[4] Hübert, Henning. Künstlich ist nicht künstlerisch. In: BR-Klassik, 2021, https://www.br-klassik.de/aktuell/news-kritik/kritik-urauffuehrung-beethoven-10-symphonie-kuenstliche-intelligenz-computer-bonn-100.html (last access: 06.02.22).
[5] Hübert, Henning, 2021.
[6] cf. Reddy, Varshini. The Limitations of GPT-3 and its Impact on Society. In: Univ.ai, 2020, https://www.univ.ai/post/the-limitations-of-gpt-3-and-its-impact-on-society (last access: 30.01.2022).
[7] cf. Marr, Bernard. ABBA’s Virtual Concert, The Metaverse And The Future Of Entertainment. In: Forbes, 2021, https://www.forbes.com/sites/bernardmarr/2021/09/06/abbas-virtual-concert-the-metaverse-and-the-future-of-entertainment/?sh=9b2fae56844d (last access: 30.01.2022).
[8] cf. Du Sautoy, Marcus. The Creativity Code: How AI is learning to write, paint and think, London: 4th Estate, 2019, p. 301.
[9] cf. Sudmann, 2021, p. 90.
[10] Sudmann, 2021, p. 96.
[11] Mersch, Dieter. Kreativität und Künstliche Intelligenz. Einige Bemerkungen zu einer Kritik algorithmischer Rationalität. In: Zeitschrift für Medienwissenschaft. Heft 21: Künstliche Intelligenzen, Jg. 11, 2019, Nr. 2, p. 65– 74.
[12] Mersch, 2019, p. 74.
[13] cf. Du Sautoy, 2019, p. 306.