On a YouTube advertising interruption during a swimming video – my passion for a few years – I did not manage to put it immediately on silent as usual, because I was operating the espresso machine just a few meters away from the computer. So I listened as Gemini, Google’s AI assistant, gave fashion advice to a young woman. Filming herself with a smartphone, she asked whether her dress was okay and whether she should wear a denim jacket. She also showed Gemini three shades of lipstick and asked whether her sneakers would fit the outfit better than her normal low shoes. The app replied that the dress would suit her very well (courtesy also costs nothing for the AI) and that jeans and sneakers as a leisure look are also better than formal clothes such as blazers or pumps.
I was fascinated. Not because of the program’s answers: That anyone would define sneakers and jeans as “casual wear” was not a particularly complex logarithm. What fascinated me, but rather negatively than positively, was the young woman’s confidence in her smartphone: that she gave more attention to the fashion judgment of an app than to her own inner voice for beauty.
To give simple everyday decisions into the invisible hands of so-called artificial intelligence corresponds on the practical level with a rather ideological opinion that crosses my path again every year. In small talk, at parties, in discussions, in chats: When it comes to AI, sooner or later people electronically socialized to the bone spout the statement that it is only a matter of time before computers, unidentifiable as such, will write better literature than human authors. So that my venerable profession is doomed, like the dinosaurs or ma-and-pa stores in the age of Amazon.
However, this gloomy prognosis for real poets is based on a fundamental error, both in terms of literature, but also about what constitutes authentic human intelligence in contrast to the artificial. The version of AI that this is about is the one that supposedly adopts meaningful features of human consciousness. In the scientific world, this potentially spiced-up, but probably only in the imagination, is called AGI (“Artificial General Intelligence”).
General artificial intelligence, for example, would mean that a computer could also correctly interpret body language in conversation with a human and would have a sense of constantly changing context. So it could deal with complex systems at least as confidently as a four-year-old child.
To anticipate it in simple words: Complex systems (for example, the behavior of a fox in a big city) are impossible to model mathematically, because they have too many mathematically indefinable elements and thus, because computers are mathematical beings, artificial general intelligence is impossible. Mathematically representable systems, such as facial recognition software at airports, which tries to filter out well-known terrorists from the millions of passengers, a kind of “concentration” game on a high level, are of course another, but in principle rather banal, Gemini-compatible matter.
But first a small, basic insert before we go on: In the current age of experts for virtually everything and everyone, people who make up their own minds without being proven professionals will automatically be very critically inspected. As independent spirits, they disturb the channeled flow of official discourse, but even more so remind it that thinking for oneself is always an option.
This blog also draws on Kant’s Enlightenment motto: “Sapere aude – have the courage to use your own mind.” Of course, not just off the top of one’s head, but inspired by certain information. The source I have drawn on here for information and that represents my inner compass is this book: “Why Machines Will Never Rule the World.” My text here is a particular literature-related application of the theories presented there.
So since it is clear that I am neither an expert, but also do not renounce them, back to the topic, namely why computers cannot produce literature that deserves the name. First of all, the data a computer spits out is always the result of logical operations. These were predefined and require a logically defined, i.e., limited space.
This clearly defined space (for example, a chessboard, a simulated solar system, the substantially reduced interaction possibilities with an ATM) of mathematically generated statements via computer is fundamentally faulty, or pointless the more foreign context openly determines the space is. For example, weather forecasts for a limited area and for the near future are often surprisingly correct, but the larger the area and the greater the time span of the forecast, the less realistic it becomes. Because then it exits the mathematically reasonably definable space and the dynamics of complex systems (such as the worldwide weather events) are no longer mathematically calculable: too many variables, too many dimensions.
If the notorious wings of a butterfly somewhere in the world can potentially trigger a tsunami, mathematics and thus the intelligence of computers would have to throw in the towel. An example of this kind of complex system escaping the understanding of AI – and calling all hysterical climate prophets to modesty – is this: Even the simple smoke plume of a cigarette (physical turbulence) is not predictable in its course, that is, it is mathematically ambiguous.
And here I come to the reason why I started this text with my handling of the espresso machine: Without this random handling with the espresso machine, this text would lack the concrete occasion (the Gemini advertisement that was not silent because I was busy with the espresso). It would never have been written in this form.
An artificial intelligence that would want to imitate the consciousness of an author would therefore not only have to be able to generate language (about which more later), but would also have to be able to mathematically depict handling an espresso machine at the time of a certain Internet advertisement as the initial ignition for a text. This is impossible, because there are too many variables (espresso machine, distance to the computer, time of advertising interruption, etc.), but also because there is no mathematically clearly defined space: espresso machine, Internet advertising, language, etc. – these are all their own, complex, dynamic systems that cannot be traced back to each other, but instead refer to each chaotically, i.e., mathematically incomprehensibly.
One can now admit that, in fact, a AI does not have to imitate the conditions of creation of a text in order to produce one, but simply comes up with a masterpiece indirectly through linguistic permutations and style analyses of world literature. It tries to do this for as long as it has, using increasingly refined, increasingly detailed analyses of the language of classic and modern texts, in order to create in this way, refreshed with contemporary content, for example novels, which are stylistically similar to those of world-famous authors.
Possibly even in a striking way. Artificial intelligence could mix the long sentences of Proust with the biting wit of Heine and the panoramic view of Tolstoy, loosened up with Japanese haiku minimalism, structured in terms of plot perhaps as a night sea journey, i.e., the path of the hero from darkness to light – and already the next Nobel Prize for literature is given to a programmer who may never have finished reading a literary book in his life.
AI groupies may be thinking roughly of this kind of computer-controlled creativity when they argue that Artificial General Intelligence (AGI) will make human literature and any human art redundant sooner rather than later. Not to mention all the other fantasies of consequences… But they are subject to the same structural error as those who believe that a complex system like the world’s climate can be mathematically modeled.
Because literary language is not a closed system related only to itself with rules and definitions like chess or go. On the contrary: Potentially every word of a demanding text refers to a reality that is also outside language. That is, AI can analyze and synthesize as much as it wants on the purely linguistic level; as long as it does not consider the immanent relation to reality that resonates in virtually every sentence in literature, it remains a caricature.
And the problem is that, at the end of the day, it can consider this all-important relation to reality at most rudimentarily and schematically, because the otherwise endless flood of changing but real meanings, as already seen in the world climate, cannot be represented mathematically.
An example. Even the use of the not particularly original, but literarily indestructible sentence: “I love you” should make artificial intelligence fail literarily. Both Proust and Heine and Tolstoy use this sentence each in his own way, thus revealing a specific world outside of language. But because their language is linked to personal experience of reality, the sentence generated by AI is simply a sentence without special, experience-rooted meaning. Logically so, because as a machine, a computer has of course never had real human experiences. That is, AI, since it cannot draw on its own experience, must generate the reality reference of the sentence: “I love you” by algorithm, i.e., by picking out some likely variant from the millions of possibilities of real meaning that make any sense in the respective context.
As a reader, however, one notices when a text has been assembled from such “probable” rather than organic building blocks. For example, “I love you!” the professor said to his anorexic student, while at the same time the mushroom of an atomic bomb lit up on the horizon and he wondered if he had already paid the ticket for his car that was towed away from the disabled parking lot.
Although such sentences are superficially correct in terms of content and form, and they even transport content, they have no real coherence for human consciousness. They break down into isolated fragments, similar to the sentences of schizophrenics; neither internal nor external reality is mirrored or generated in a meaningful way. But this is logical because artificial intelligence, like schizophrenic consciousness, has no independent, stable center from which it can meaningfully order the flood of perceptions or, in the case of the computer, the flood of data. At least not when the context of the data changes, which is the case with literature (as a complex system) for virtually every word.
Everyone knows the harmless human quirks when some people always briefly laugh loudly after a sentence or constantly weave „so to speak“ into their sentences. Exactly this kind of schematism, regardless of the situation, is inevitable for artificial intelligence. It must act “neurotically” as it were, i.e., repeatedly use certain calculation operations even if they make no sense, because this, namely computational operations, are exactly its essence.
The essence of literature, to take the art to which I have dedicated myself, however, is not computational operations, algorithms, probabilities, but rather a holistic, often completely irrational and not externally standard process. One could even go so far as to say that at the moment when literature becomes fully explainable, i.e., when a computer can generate it, it immediately ceases to be literature and decays into kitsch.
For artificial intelligence, the incomprehensible and intuitive – the invisible aura between the lines that is inherent in every real literature – is a book not with only seven, but with infinitely many seals.
What on the unconscious level possibly explains the sympathy of many people for artificial intelligence: They feel instinctively without being able to put it into words that the “all-knowing” machine has no real insight. And nothing is more beneficial for a good friendship than the bond of a common weakness.
Joke aside, maybe. Rather, it seems that the unreflected human-electronic submission to everyday issues in contact with Gemini and other variants of AI represents a kind of current shift of religious and fallow energy. The law of the conservation of energy applies also to people’s transcendent energy (C. G. Jung even spoke of the transcendent function of the psyche): Spirituality or religiosity or faith are never entirely dead; they only constantly change their face. At the moment, it is often the caricature of the smartphone or the face of fanaticism that has replaced real spirituality. Two years ago in Israel, it was even both: Filming and posting a massacre, the smartphone and the machine gun united psychotically.
World literature, on the other hand – to return to the subject once again – has nothing to do with the caricature or the face of man, but shows, often without knowing or wanting to, that human consciousness is an unresolved mystery that evades every calculation, but is alive: equally distant from banality and cruelty.
