I recently had a heated conversation with one of my colleagues here at Balkan Hotspot about an art competition in his hometown. With a furious look on his face, he showed me the winning poster for the upcoming city festival, and on closer inspection I realised that his anger derived from the fact that it was created using AI. He angrily lamented about how much work him and other artists from his hometown put into their designs for the contest, only for an incohesive, soulless AI poster to triumph them. The problem is that, even though the internet is full of it, a lot of people do not know when they are looking at art created by AI. But should AI art be assessed differently than art made by humans? And beyond that, can it even be labeled as art?
Defining AI Art
The first step towards tackling a complex topic like this is defining what exactly is meant when talking about art. Of course, that is not an easy task, and a subjective topic that has been in dispute in various contexts, specifically regarding artisans, modern art and art made with digital technology. But thankfully, it is not the purpose of this article to define art in all its variations, but to address specifically the side of it that is most affected by AI: the visual arts. For this purpose, this definition as per Britannica seems to be sufficient: an object or experience consciously created through an expression of skill and imagination. The term art encompasses diverse media such as painting, sculpture, printmaking, drawing, decorative arts, photography, and installation.
The second unavoidable prerequisite for this discussion is a basic understanding of how generative AI works. To begin, a large dataset is gathered, that diversely represents the content intended to be generated. In this case, as we want to generate visual art, this data is mainly images of artwork spanning different categories, styles and subjects. Those images are transformed into raw data that is standardised and can be fed into the model. When talking about AI models for visuals, the most commonly used one for generating visuals are Generative Adversarial Networks (GANs). They consist of two parts of code: the discriminator and the generator. While the generator is tasked with creating content based on the underlying patterns and complexities of the dataset, the discriminator evaluates the generated visuals against real ones. Through a competitive learning process, both networks push each other to improve the output and make it as close to real artwork as possible. What we now have is a model that generates random images from noise that are close to real artwork, but ultimately just imitate the pictures it is being fed. To make it follow the prompts we give it, it is exposed to labeled training data, meaning real art and pictures that are marked with a description. The model learns patterns and relationships connecting words with visual content, and also typical styles, motifs and how certain scenes are often represented. Thereby it is close to impossible to control the generating process in a detailed manner, as the information given to the model contains human bias and ethics violations that are recreated in the training process. Nevertheless, it is continuously adjusted and fine-tuned to achieve the desired output and transform it from random noise into representative images.
Is AI creative?
Looking at this process, it becomes clear that the requirements for its results to be called art are not even met in the slightest. Artists hone their craft for years, they create countless unsatisfying works only to achieve one that they deem worthy. They know where to set shadows, how anatomy works, what colours and composition to utilise to awake certain emotions in their viewer. The insufficient rip-off simulated by AI, which does not possess the knowledge or experience to be able to explain why it is setting a certain shadow or creating a specific line, could never come close to humans’ hard-earned knowledge. When comparing a child’s imperfect drawing to an AI-generated artwork with flaws, the child’s work gains value through its unique creative expression and motor skill diversity, whereas the AI piece lacks that genuine artistic contribution. Because the point that is even more consequential in the comparison of AI and humans is one simple fact: AI is not creative. It generates, but it does not invent. It lacks the ability to understand context, it has no perspective, no values, it sends no message. Rest assured that AI has never created anything even remotely interesting, and that it will stay this way for the foreseeable future.
As of now, creativity and thinking outside the box are skills that only humans have managed to acquire. An AI can generate a picture, but can it tell you if it is a controversial one or not? Philosophical debates like the quandary of how to deal with morally complex art are constantly negotiated in the art world. Should controversial, possibly problematic works be exhibited, and how should they be reflected on? None of these subtleties are being reflected in AI-generated art.
Theft of intellectual property
Besides this fundamental dissonance between AI and anything that would be considered art, there are also more tangible, day-to-day problems arising in this debate. One of the biggest points of discussion concerns copyright and intellectual property. As we now know, a big dataset is necessary to train AI. What makes this difficult is that artists who post their art online have no control over the fact that their work is being used without their knowledge or consent. It is almost ironic that their work has been used to train their own nemesis, who is threatening to become their replacement. One might think that this is an unvariable status quo, and it may be true that protecting copyright and intellectual property is more difficult than ever nowadays. Nevertheless, this is not a reason to leave artists to comply with their unfortunate fate. The American Software Company Adobe sets a positive example of how this can be handled: Only licensed or public domain content is used to train their visual AI Firefly. This is a start for the limitation of AI’s scope of influence concerning art, but it is one of the few examples of artist’s rights being considered in creating AI. We have seen disappointingly little action from governments and international unions in the regulation of AI, especially in the field of the arts. While the first regulation of AI was passed in the fom of the EU AI act in march of 2024, none of the included provisions regulate the use or creation of art in any way.
The ethical dilemma of technology
Beyond its legal implications, AI creating “art” is simply a solution to a non-existing problem: artists enjoy making things, in fact, human creativity is arguably one of the strongest forces that shaped the modern world. There is no moral reason to replace this valuable motivation of society with a machine, and if you have ever talked to real artists, you will have realised that not a single one of them feels the urge to relay their craft to an AI, no matter how hard and frustrating the process gets. This goes without saying because AI is one of the main threatening forces for the livelihood of artists. Not because the technology is getting as creative or technically skilled as them, but because it is cheaper and quicker than them, and the ones removing artists from the creative process are those who do not care about quality, and benefit from fast, cheap production.
What needs to be mentioned here is that, as AI does not possess forces of creativity and identity, it is predicted to damage itself and get into an output loop in which it starts only consuming its own products and producing further on the basis of them. The conspirators of the so-called “Dead Internet Theory” claim that this has already happened; they theorise that activity and content on the internet, including social media accounts, are predominantly being created and automated by artificial intelligence agents. Even though a conspiracy as such might be a tad far-fetched and out of touch, it is no question that the internet nowadays is full of bots and fake-accounts. Should the amount of AI-controlled activity increase, not dissimilar to how Twitter is developing, an ocean of spam and fake pictures will eventually turn the replies and content in those spaces into mindless mush.
Ecological Implications
In addition, the processing of AI models takes place in data centres, which require a lot of computing power and are very energy-intensive. “The entire data centre infrastructure and data transmission network are responsible for two to four per cent of global CO2 emissions,” says Anne Mollen, researcher at the Berlin-based non-governmental organisation Algorithmwatch. This is roughly equivalent to the emissions of the aviation industry. “It’s not just down to AI, but AI is a big part of it,” says Mollen.
In a 2019 study, researchers at the University of Massachusetts, Amherst, found that training a common full-scale AI model can produce up to 284,000 kilograms of CO2 – almost five times the emissions of a car over its entire lifetime, including manufacturing. And even if we just measure the electricity, AI models use far more power than traditional internet uses, like search queries or cloud storage. According to a report by Goldman Sachs, a ChatGPT query needs nearly 10 times as much electricity to process as a Google search. For years, data centers displayed a remarkably stable power consumption, even as their workloads mounted. Now, as the pace of efficiency gains in electricity use slows and the AI revolution gathers steam, Goldman Sachs Research estimates that data center power demand will grow 160% by 2030.
But energy is not the only aspect when it comes to the environmental impact of AI. With the rise of it being used commercially, companies have significantly raised their water usage, sparking concerns about the sustainability of such practices amid global freshwater scarcity and climate change challenges.
The enormous amounts of water that data centres need to prevent overheating cause major problems in water-scarce regions such as Santiago in Chile. Google’s data centre there is exacerbating the drought in the region, and local communities are protesting against the centre and the construction of new data centres. Antonio Guterres, UN Secretary General said at the UN Water Conference that “Water is a human right and the common development denominator to shape a better future. But water is in deep trouble.” Being receptive to the changes in the way technology affects sustainability is the key for changing our behaviour in a way that can prevent this giant crisis of natural resources. But as long as we treat the usage of AI as a higher priority than the literal essence of life that makes our planet thrive, we are staying blind to the problems that are created by big companies, lying to their customers without batting an eye.
Data Bias
Looking at the ethical consequences of AI usage, they extend even further than the ecological aspect. Because AI generates based on various types of images and cannot decipher the ethics of what it creates, it will not hesitate to create content that is inappropriate, homophobic, misogynistic, or just blatantly wrong. It validates the data bias that exists in the art world, generating based on human stereotypes and furthering inequality through unbalanced representation. As AI ultimately only works on what humans put into it, it takes every bad aspect of humanity and counts it into its algorithms often unbeknown to humans. It is not a new fact to most people that the art world is not fair, that there are lots of perspectives that are largely missing or misrepresented. This gets even more obvious when taken into the AI world. For instance, art created by women makes up only a fraction of the artistic canon, their perspective largely missing in the art world. A 2019 study of artworks in the prestigious MoMA in New York deducted by Masterworks found that 80% of artworks in their collection are from male artists, 85% of their artwork comes from North America and Europe, and 77 countries in the world do not even have one artwork in their collection. While it may seem like the social background of the artist is irrelevant to the technique and skill displayed in an artwork, the message, motivation and intention behind the creation, as well as the creative process itself and the consequences it has for culture and the art world differ widely depending on the diversity of perspectives.
Considering depictions of sexual violence as an exemplary case, many artworks frame the male subject as central to the piece, with examples ranging from Greek sculpture to Renaissance paintings. The male is often presented as powerless to his lust, and we are invited to empathize with his obsession and inability to resist. The depicted female is often passive, deprived of voice and agency. On the counter side, if we look at scenes displayed by female artists like Susanna and the Elders (1610, Artemisia Gentileschi), we see a different perspective. The expression of the woman is disgusted, she is trying her best to say “no”, to repell them, and the fear in her body language is clearly visible. Despite the fame and recognition she received during her lifetime, Gentileschi is relatively unknown in online and offline galleries. These sobering, disappointing circumstances force us to reconsider the objectivity of the data that AI models are trained on, and whether they will further perpetuate and amplify the inequalities still existing in society.
AI is an Alien
When immersing ourselves in real art, we experience a new view, a thought that otherwise never would have occured to us, the fingerprint of a human life, channeling its energy into creation. AI is just a bland amalgamation of what has come before. If you have ever tried to use a generative writing AI for an essay or a story, you will hopefully have noticed that it only spits out forgettable, repetitive mush dressed in big words. It is like an alien, looking at humanity and the art we create from the outside, trying to imitate it by finding all the common factors and recreating them in an insufficient way. We appreciate art for the emotions it evokes in us, and artist for the way they manage to convey so much intention with often just a few brushstrokes. Our individual understanding of the context into which a work of art grows also has a significant impact on how we appreciate the work and its value. We find it difficult to attach much value to artificially produced images because they lack the authenticity and depth of meaning that we normally look for in works of art. It is this authenticity, and the lived experience that inhabits a work of art, which remains of value to us. Perhaps we will learn to appreciate this even more in the age of generated artworks.
P.S.: If you want to know how to spot AI art, stay tuned for our next releases and our guide to spot fake art.
