“Art Intelligence”, in: Intelligence it's automatic – From Surrealism to Artificial Intelligence 1925-2025, Stichting Liedts-Meesen, Ghent 2025.
In 2024, I published a book, Art Intelligence – How Generative AI relates to Human Art-Making(1), in which I speculate about the future for art-making and artists after the advent of generative AI. 18 years earlier, I had published an essay titled Controlled Production of Virtual Geo-political Reality through Failure(2). At the time, I had no idea of the parallels that would occur between the handmade work I referred to and the discussions that would be sparked by generative AI some fifteen years later. I argued at the time that it was impossible for an AI to have produced the work in question. I would like to think that this observation is still true. For how long?
The essay explains how I use an analog drawing method to generate (sic!) a series of twenty political maps of the European Union. The series begins with a map that was accurate at the time, which then gradually morphs into a vision of what the geopolitical reality of the Union might look like twenty images into the future. I have produced three such series so far: in 1998 (the subject of the 2006 essay), in 2009 (with a print version published in 2010), and in 2020. In each case, Europe and its Union are gradually but inexorably changing their shape.
Most of the work on the 2020 version, Psycho-Mapping Europe post Brexit, took place in Berlin during the pandemic lockdown. In those days of great uncertainty, it was eerie to see Europe shape-shift before my eyes. As for the work's potential as a prognosticator of changing geopolitical realities, what could be more appropriate than to see it in a Moscow museum on February 24, 2022? Next it will be in Ghent in February 2025, and who knows what will happen then.
Generative Artificial Intelligence does not work from scratch. It uses ingenious methods to combine fragments of existing material, material on which it has been 'trained', to create new output. In what follows, I will construct a new text myself by borrowing and juxtaposing paragraphs and fragments from these two earlier texts(3), both of which discuss generative art-making, although from different perspectives. One digital, one analog.
This is how I began in 2006: 'There is an important difference between an image that is a work of art and one that is an illustration. The content of the latter will have been defined before the image is made: its role is to serve, to clarify, to visualise. It should not add complexity or provoke new questions. A work of art, on the other hand, may serve a purpose or have an agenda, but it will contain a question that cannot be fully answered, a kind of intangible value that cannot be extracted without being redefined and changed in the process. This value makes the work unstable and encourages the viewer (a term that will include the artist) to shift their position.'
And in 2024 I claimed: 'After the introduction of generative artificial intelligence, anyone working in an artistic field will have to think carefully about what there is that cannot be digitised. And if and how they can use the new possibilities to further their artistic goals. It follows, then, that the introduction of generative AI prompts renewed reflection on what art is and what purpose it serves.'
This led to the following question: 'Artists have long used tools and various machines to make their work, but now there are machines that can also conceive the work. What impact will this have on how we continue to make art and how we understand it? How will it affect the role of the artist in society and the conditions under which artists will work in the future?"
From now on I will mix the two texts without indicating which is which. Triggered by the prompt, the AI combines bits and pieces into an image according to the logic of its algorithm. It is incapable of considering what meaning its image might have in the interpretation of a human viewer. An LLM-based AI determines which chain of words (for the chatbot) or visual configuration (for the image-making application) is an appropriate response to the question or request formulated in the prompt. The same process can also lead to invented facts, what we now call 'hallucinations'. Such errors, which can have potentially huge consequences depending on what the AI's answer is used for, will still be the result of a chain of isolated and distinct, mathematically based decisions. They will be unintentional, unguided by emotion.
The first image in Psycho-Mapping Europe is a political map of Europe and beyond. I collected many different maps (both political and geographical), studied and compared them (even on a limited landmass like Europe the effects of different projections are surprising) and chose one with an aesthetically pleasing projection. I then enlarged it and transferred it to my first sheet of paper using a light table. I drew the map image with a thin black ink line (using an accident-prone steel dip pen) showing coastlines and major lakes as well as political boundaries. No towns, mountains or rivers. The map image is framed by a thin black border approximately one centimetre inside the edge of the paper. This border was drawn first.
Having finished drawing (= tracing) the first map image in the series, I took a new sheet of paper, added the thin frame and began to copy the first map drawing line by line, dot by dot, using the same pen, trying to be as exact as was humanly possible without recourse (this time!) to the light table or any other tool (and limited by the extent of my patience). I imagined myself as a machine, a kind of human scanner, unaware of anything other than the precise task ('programmme') at hand: to copy, bit by bit, a system of lines from one sheet of paper to another (starting at the upper left corner and moving from left to right, from top to bottom) using only my eyes and my hand. As I was drawing with ink, erasing was not an option and I could not sketch. Every mark had to stay on the paper. Every line in the original had to be included, nothing could be deliberately left out. I was patient and careful, and my copy turned out well: it looked very much like its model. On closer inspection, however, there were small mistakes and deviations everywhere. I had to accept them, they could not be changed.
The human artistic decision-making process is full of influences caused by human friction. The person may be unaware of these influences, may embrace them or fight them, depending on the direction in which the person’s self-awareness leads them. While the archive used by the LLM is incomparably larger, the 'archive' used by the human creator will be in some ways broader and the variation factor will be different. It will include all sorts of vaguely defined feelings, unrelated ideas and fixations, as well as unspoken physical memories and recollections that refuse to come out of the fog. All of this will not be directly related to the self-defined task, but may still influence the choices made. Ideas about the self, animosities, desires, misunderstandings, competition – there is no end to the human friction that can be part of an artist's decision-making process. The individual is never an optimal specimen. He or she is necessarily flawed, a biological glitch. Their glitch talents are what they must exploit. Of course, the AI's programming can be tweaked to include all sorts of external and unrelated glitches, but until it has the self-awareness of a human, it will never be comparable. The procedure of careful copying was repeated nineteen times. Each time I made every effort to avoid 'correcting' (in relation to any lingering memory of the first map or any other extraneous knowledge) any 'mistakes' that had been incorporated into my model. My task – unusual in an artistic process – was to concentrate solely on the mechanical process of copying. Not to add, not to subtract: not to 'translate'. Stay calm, stay focused, just get it 'right' as it is. Avoid any input.
No matter how hard I tried, new imperfections and mistakes could not be avoided, and so a process of accumulating failure had begun. This, of course, was my hidden goal for the whole exercise. I can say that now. But it was of the utmost importance that no deviation from the model was deliberately added; all failures must be real. Any inclusion of deliberate failure would lead to something entirely else: design. My interest was to find out something I did not know, not to project ideas already conceived. I suspected that the work would somehow carry a visual trace characteristic of me, its maker, but I did not know at the time how that trace would be constituted, and I was determined to do everything I could to avoid it in the first place. The more I could remove my subjectivity from the process (i.e. slavishly following the technical rules I had set up to regulate the copying process), the more value I would add to conceptual aims of the work.
When all twenty ink drawings were finished, they were scanned and partly coloured in the computer: the member states of the European Union were each given one colour. Finally, they were printed in an edition of three.
Interactions with AI can lead us to attribute human characteristics or behavior to non-human entities. It's important to remember that while AI may be very good at mimicking human-like responses, it does so not from a point of shared understanding and emotion, but from an outside perspective, creating an impression of what human interaction can be like. Its responses are generated based on its programming and training data.
Every work of art is the result of a game, a play of possibilities combined to create something meaningful. André Breton spoke of the sudden appearance of 'le merveilleux,' the marvellous(4) . Once you have seen a glimpse of the marvellous (a phenomenon which is, of course, in the eye of the beholder), you want to see it again and be able to conjure it at will. And so chains of works are created: repetitions, variations, discoveries and rediscoveries. Until the spark fades. If, on the other hand, the production is guaranteed by a machine (possibly guided by parameters set by the artist), the spark will only die once someone clicks 'STOP'.
Eventually, we will find ourselves in a world where 'creative' content is infinite. No one will have to struggle to find the right illustration or decoration, the right pop song or mood music, the right horror story or political tract; it will be provided instantly, to taste and on demand. There will be no friction. Or so the idea goes. Modern advertising has been based on a similar idea for decades: 'Buy our product to assert your style and values'. Buy, to be. What about an image made by a machine? Can it be 'good'? Can it be 'bad'? If an almost formless drawing by a human artist can evoke a passionate response...what about a formless expression produced by an LLM? (Which, it must be said, will necessarily be based on archived examples of 'formless' expressions by humans.) Can we even speak about the 'expression' of a machine? The machine has no need to 'express itself' in order to prove anything. It can't think about what it's going to do before it does it. It cannot think about it in advance. It does what it is told to, without delay. The human client may think it does it very well, but the machine itself will have no opinion. How are we to understand extremely reduced, minimal, and monochrome art expressions after the arrival of art-making machines? And what about conceptual art in general?
The 1998 version of my Psycho-Mapping series has been exhibited a several times. Each time there have been viewers eager to prove their knowledge who have come up to me to discuss computer animation. Something about the way these maps look make them think – take it for granted! – that these images have been generated by a computer programme. It really is a fascinating misconception!
An LLM has no consciousness, self-awareness, or personality. Everything it produces is based on what has already been produced by humans. Over time, as AI-generated text and images proliferate on the internet and scraping continues, feedback loops will occur. Eventually, the text-scraping(5) robots will be scraping text written by robots. And the feeder robots for the training of image-generating AIs will use not only human-generated source material, but also images produced by DALL-E and the like. Over time, such feedback loops will be influenced by the filtering decisions made by the companies providing the services. These decisions in turn depend on both contextual circumstances and commercial strategies.
The new tools are here and we will use them. The question is how they will affect the art we make. And also how we will define 'quality' in the future. Are we going to make art with artificial intelligence? Will we make art about artificial intelligence? Will we make art against AI? What should we really care about? When AI produces strange combinations, or when it produces what are now called hallucinations, it is usually not due to a malfunctioning algorithm. The resulting answer may be logically 'correct' within the AI's framework, but incorrect, hilarious, or even dangerous when interpreted by a human living in the real world.
The creation of such a work requires a highly controlled process in order to achieve an unpredictable result that contradicts its assumptions. I am not a programmer, but I think I can say with impunity that there are no AI algorithms that come close to the dynamics of my Psycho-Mapping. But there is another aspect here that is just as important: I am responsible. There is someone behind this. Every quirk of line here is the result of a specific human error, and every millimetre of ink line and dot here can be infinitely magnified and related to the myriad aspects of reality – life – on the ground. Paradoxically, it is this aspect that allows the work to be seen as an 'illustration'. An illustration of how new value – content – information – culture – is born out of constant human failure to do the right thing. While never giving up trying.
What if our use of AI leads to a renewed focus on the core function of art? In recent years, the art debate and institutional politics have focused on art's need to fulfill agendas. Soon that role may be taken care of by machines. Art has never been a reliable agent of propaganda, both because of the way it works technically and by reason of the conflicting motivations of artists. AI will be able to produce unlimited amounts of decorative and instrumentalized art without effort. What will remain for artists is to create meaning with human effort, with human friction. Meaning for human lives.
Jan Svenungsson