x

Does Artificial Intelligence Really Have the Potential to Create Transformative Art?

Stephen Marche on the Possible Futures of Machine Creativity

I. The Situation

In 1896, the Lumiere brothers released a 50-second-long film, The Arrival of a Train at La Ciotat, and a myth was born. The audiences, it was reported, were so entranced by the new illusion that they jumped out of the way as the flickering image steamed towards them.

The urban legend of film-induced mass panic, established well before 1900, illustrated a valid contention if the story was, in fact, untrue: The technology had produced a new emotional reaction. That reaction was hugely powerful but inchoate and inarticulate. Nobody knew what it was doing or where it would go. Nobody had any idea that it would turn into what we call film. Today, the world is in a similar state of bountiful confusion over the creative use of artificial intelligence.

Already the power of the new technology is evident to everyone who has managed to use it. Artificial intelligence can recreate the speaking voice of dead persons. It can produce images from instructions. It can fill in the missing passages from damaged texts. It can imitate any and all literary styles. It can convert any given authorial corpus into logarithmic probability. It can create characters that speak in unpredictable but convincing ways. It can write basic newspaper articles. It can compose adequate melodies. But what any of this means, or to what uses these new abilities will ultimately be turned, are as yet unclear.

There is some fascinating creative work emerging from this primordial ooze of nascent natural language processing (NLP). Vauhini Vata’s GPT-based requiem for her sister and the poetry of Sasha Stiles are experiments in the avant garde tradition. (My own NLP-work falls into this category as well, including the short story this essay accompanies.)

Then there are attempts to use AI in more popular media. Dungeon AI, which is an infinitely generated text adventure driven by deep learning, explores the gaming possibilities. Perhaps the most exciting format for NLP is in bot-character generation. Project December allows its users to recreate dead people, to have conversations with them. But there’s no need for these generated voices to be based on actual human beings. Lucas Rizzotto concocted a childhood imaginary friend, Magnetron, which existed inside his family’s microwave, out of OpenAI and a hundred-page backstory.

These early attempts to find spheres of expression for the new technology are dynamic and exciting, but they remain marginal. This work has not yet resonated with the public, nor has it solidified into coherent practice.

The scattered few of us who use this technology feel its eerie power. The encounter with deep learning is simultaneously ultramodern and ancient, manufacturing an unsettling impression of being recognized by a machine, or of having access, through machines, to a vast human pattern, even a collective unconscious or noosphere. But that sensation has not yet been communicated to audiences. They don’t participate in it. They see only the results, the words on the page, which are little more than aftereffects.

Then there is the question of whether anyone wants computer-generated art.

The literary world tends to engage creative technology with either petulant resistance or slavish adulation. Neither are particularly useful. A novel about social media is still considered surprisingly innovative, and even the smartphone rarely makes an appearance in literary fiction.

Recent novels about artificial intelligence, such as Klara and the Sun by Kazuo Ishiguro or Machines Like Me by Ian McEwan, have absolutely nothing to do with actual artificial intelligence as it currently exists or will exist in the foreseeable future. They are, frankly, embarrassingly lazy on the subject.

Meanwhile, the hacker aesthetic has had its basic fraud exposed: it fantasized technologists as rebel outsiders, poised to make the world a better place, as a cover for monopolists who need excuses to justify their hunger for total impunity.

Both the resistance and the adulation are stupid, and so we find ourselves toxically ill-prepared for the moment we are facing: the intrusion of technology into the creative process. The machines are no longer lurking on the periphery; they are entering the temple, piercing the creative act itself.

The Lumiere brothers produced roughly 1,400 minute-length films, or “views” as they were called at the time, but nobody could see what these views would blossom into: A Trip to the Moon, and Birth of a Nation, and Citizen Kane, and Vertigo, and Apocalypse Now. Creative AI is not a new technique. It is an entirely new artistic medium. It needs to be developed as such. The question facing the small band of creators using artificial intelligence today is how we get from The Arrival of a Train at La Ciotat to Citizen Kane.

 

II. The Direction

One thing is certain: Nobody needs machines to make shitty poetry. Humans make quite enough of that already. The blossoming of AI art into its unique and particular reality will demand a unique and particular practice, one that sheds traditional categories of art as they currently exist and which engages audiences in ways they have never been engaged before.

One potential danger, at least in the short term, is that the technology is advancing so quickly it is unclear whether any artistic practice that emerges from it will have time to mature before it becomes obsolete.

Creative AI is not an expression of a self. Rather it is the permutation and recombination and reframing of other identities.

Every example of creative AI I have listed above uses GPT-3 (enerative Pre-trained Transformer 3). But Google just very recently released its own Transformer-based large language model, PaLM, which promises low-level reasoning functions. What does that mean? What can be built from that new function? Art requires technical mastery, and also conscious transcendence of technical mastery. Even keeping up with the latest AI developments, never mind getting access to the tech, is a full-time job. And art that does nothing more than show off the power of a machine isn’t doing its job.

Then there is the question of whether anyone wants computer-generated art. One of the somewhat confounding aspects of the internet generally is that it is hugely creative but fundamentally resistant to art, or at least to anything that identifies itself as art. TikTok has turned into a venue of explosive creativity but there is no Martin Scorcese of TikTok, nor could there ever be. Internet-specific genres, like Vine, are inherently ephemeral and impersonal. They aren’t art forms so much as widespread crafting activities, like Victorian-era collages, or Japanese Chigiri-e, or Ukrainian pysanky.

When people want to read consciously made, individually controlled language, they tend to pick up physically printed books, as ridiculous as that sounds. Creators follow the audiences. The top ten novels published this year are not fundamentally different, in their modes of composition, dissemination and consumption, from the novels of the 1950s.

But the resistance creative AI faces, both from artists and from audiences, is a sign of the power and potential of the new medium. The most exciting promise of creative AI is that it runs in complete opposition to the overarching value that defines contemporary art: Identity. The practice itself removes identity from the equation.

Since so few people have used this technology, I’m afraid I’ll have to use the short story that accompanies this essay as an example, although, to be clear, many people are using this tech in completely different ways and my own approach is representative of nothing but my own fascinations and capacities.

A few months ago, I received access to the product of a Canadian AI company called Cohere, which allows for sophisticated, nimble manipulations of Natural Language Processing. Through Cohere, I was able to create algorithms derived from various styles. These included Thomas Browne, Eileen Chang, Dickens, Shakespeare, Chekhov, Hemingway and others, including anthologies of love stories and Chinese nature poetry.

I then took those algorithms and had them write sentences and paragraphs for me on selected themes: a marketplace, love at first sight, a life played out after falling in love. The ones I liked I kept. The ones I didn’t I threw out. Then I took the passages those algorithms had provided and input them to Sudowrite, the stochastic writing tool. Sudowrite generated texts on the basis of the prompts the other algorithms had generated.

To generate Autotuned Love Story I had to develop a separate artistic practice around the technology. I’m not proposing my practice as a model; in fact, now that I’ve done it, I don’t see why anyone else would do what I’ve done. My point is that what I created here and how I created here is distinct from traditional artistic creation.

In its early years, hip hop had a defiance and a focus on effect which AI art should aspire to.

The love story below is my attempt to develop an idealized love story out of all the love stories that I have admired. It exists on the line between art and criticism. Autotuned Love Story certainly isn’t mine. I built it but it’s not my love story. It’s the love story of the machines interacting with all the love stories I have loved. I confess that I find it eerie; there is something true and moving in it that I recognize but which I also can’t place.

Creative AI is not an expression of a self. Rather it is the permutation and recombination and reframing of other identities. It is not, nor will it be, nor can it be, a representation of a generation or a race or a time. It is not “a voice.” Whatever voice is, it is the opposite. The process of using creative AI is literally derivative. The power of creative AI is its strange mixture of human and other. The revelation of the medium will be the exploitation of that fact.

Because creative AI is not self-expression, its development will be different from other media. On that basis, two propositions:

Artists should not use artificial intelligence to make art that people could make otherwise.

The display of technology cannot be the purpose of the art.

Creative AI should, above all, be itself and not something else. And secondly it should allow users to forget that it’s artificial intelligence altogether. Otherwise it will be little more than advertising for the tech, or an alibi for the artist.

Fortunately, there is a predecessor that can serve as a model, and which follows the two directions above: Hip hop. Hip hop was an art form determined, from its inception, by technological innovation. Kool Herc invented the two-turntable setup that allowed the isolation of the break, and Grandmaster Flash developed backspin, punch phrasing, and scratching. These developments required enormous technical facility but also a concentration on effects. The artists shaped the tech in response to audience reactions.

Hip hop also demanded an entirely new musicality to maximize the effects of the innovation. Building beats and sampling required a comprehensive musical knowledge. The best DJs had the widest access to music of all kinds, and were each, in a sense, archivists. They engaged in “raids on the past,” using history for their own purposes.

Just as hip hop artists developed a consummate familiarity with earlier forms of popular music, the artists of artificial intelligence who use large language models will need to understand the history of the sentence and the development of literary style in all forms and across all genres. Linguistic AI will demand the skills of close reading and a historical breadth as the basic terms of creation.

And when we look at the bad AI art available now the failings of the art are almost never technical. It’s usually a failure to possess deep knowledge, or sometimes any knowledge, of narrative technique or poesis.

In its early years, hip hop had a defiance and a focus on effect which AI art should aspire to. They showed a willingness and capacity to create and abandon values. They did not worship their instruments. They concentrated on the results, and that spirit largely survives. A good question to ask as a rough guide to the creative direction of AI art: What would Ye do? WWYD?

 

III. The Stakes

Creative AI promises more powerful illusions and more all-consuming worlds. Eric Schmidt, at The Atlantic, recently offered an example of the future awaiting us:

If you imagine a child born today, you give the child a baby toy or a bear, and that bear is AI-enabled. And every year the child gets a better toy. Every year the bear gets smarter, and in a decade, the child and the bear who are best friends are watching television and the bear says, “I don’t really like this television show.” And the kid says, “Yeah, I agree with you.”

Despite this terrifying promise, AI art will probably remain small and marginal in the short term, just as film was for several decades after its birth.

The development of creative AI is much, much more important than how cool the new short stories or interactive games can be. For one thing, artistic practice may serve as a desperately needed bridge between artificial intelligence and the humanities. As it stands, those who understand literature and history don’t understand the technology that is about to transform the framework of language, and those who are building the technology that is revolutionizing language don’t understand literature or history.

The political uses of artificial intelligence will follow creative practices.

Also, the political uses of artificial intelligence will follow creative practices. That’s certainly what happened with film. A few decades after the The Arrival of a Train at La Ciotat, Lenin was using film as the primary propaganda method of the Soviet Union, and the proto-fascist Gabriele D’Annunzio filmed his triumphal entrance into the city of Fiume. Whatever forms creative AI takes will, almost immediately, be used to manipulate and control mass audiences.

Creative AI is a confrontation with the fact that an unknown number of aspects of art, so vital to our sense of human freedom, can be reduced to algorithms, to a series of external instructions. Marovec’s paradox—that the more complex and high-level a task, the easier it is to compute—is fully at play. Capacities requiring a lifetime of dedication to master, like a personal literary style, can simply be programmed. The basic things remain mysteries. What makes an image powerful? What makes a story compelling? The computers have no answers to these questions.

There is a line thrusting through the world and ourselves dividing what is computable from what is not. It is a line driving straight into the heart of the mystery of humanity. AI art will ride on this line.

_________________________________________________

AUTOTUNED LOVE STORY

[This story was generated by means of natural language processing, using Cohere AI and Sudowrite accessing GPT-3.]

The rain in the market smelled like rusting metal and wet stones. The stallholders had no real need to sell nor did they care much for their customers. There was a cookery demonstration. There was a magician. There was a video games stall. There was a beauty parlour. The rain was like a mist at first, fine and barely noticeable, but not long after the streets were flowing with a torrent of mud and water.

Among huddles of people, they met in a stall that sold umbrellas. The eyes of one were large and green, soft and milky. The other’s eyes were like iced coffee.

Shyness came upon them at once. Shyness and fear. A butcher’s boy, with a beautiful nose, stood beside a post, making grimaces at a plan that was chalked out on the top of it. A ragged little boy, barefooted, and with his face smeared with blood, from having just grazed his nose against the corner of a post, began playing at marbles with other boys of his own size. Their smiles were interminable, wavering and forgetful, and it seemed as though they could not control their lips, that they smiled against their will while they thought of something else.

–Alone?

–Yes.

The rain became like a dirty great mop being wrung out above their heads. The market became more uneasy, and gave place to a sea of noises that on both sides added to the general clamour. The crowd began to press in on them, to snatch at their coats, to groan, to criticize and to complain of cold and hunger, of want of clean clothes, of lack of decent shelter. The rain was unremitting—just like the flow of people, the flow of traffic, the flow of tired animals. The crowd erupted and all at once it seemed that there were too many people.

When the crowd closed up again, the two were separated from one another. The rain died down and the market was now very different. They looked for each other like lost children in a train station. It was a different kind of a market, darker, older, dingier, more chaotic. The pavement was covered with mud and mire and straw and dung.

They met by accident, which is only a way of saying that we have not looked for something before it comes forward, that they were both in the world and the world is small.

*

They never met again, or maybe they did.

Maybe, at first, they had the same delight in touching, in meeting, in forming, in blurring, in drawing out. They had secrets, and they shared those secrets. As one’s hands rolled over the other, they lay as still as fish. It seemed to both of them that they could not live in the old way; they could not go on living as though there were nothing new in their lives. They had to settle down together somewhere, to live for themselves, alone, to have their own home, where they would be their own masters. They went abroad, changed their lives. One was a manager of a railway branch line. The other became a teacher in a school. And the large study in which they spent their evenings was so full of pictures and flowers that it was difficult to move about without upsetting something. Pictures of all sorts, landscapes in water-colour, engravings after the old masters, and the albums filled with the photographs of relatives, friends, and children, were scattered everywhere about the bookcases, on the tables, on the chairs. Love is like money: the kind you have and do not want to lose, the kind you lose and treasure. The thought of death, which had moved them so profoundly, no longer caused in either the former fear and remorse, a sound that lost its echo in the endless, sad retreat, a phantom of caresses down hallways empty and forsaken.

Maybe they lived that life. Maybe they didn’t. But in the market, among the detritus, the splintered edges, they had once found each other, and found each other and lost each other again. They had said only that, yes, they were alone.

The rain had smelled like sodden horses and rusting metal and wet stones.

Stephen Marche
Stephen Marche
Stephen Marche is a novelist, essayist and cultural commentator. He is the author of half a dozen books, including The Unmade Bed: The Messy Truth About Men and Women in the Twenty-First Century (2016) and The Hunger of the Wolf (2015). He has written opinion pieces and essays for The New Yorker, The New York Times, The Atlantic, Esquire, The Walrus and many others. He is the host of the hit audio series How Not to F*ck Up Your Kids Too Bad, and its sequel How Not to F*ck Up Your Marriage Too Bad on Audible, and is currently at work on a book about the possibility of a civil war in the United States for Simon and Schuster.





More Story
The Unnoticed Generation: How Russian Writers in Paris Grappled With the Complexities of Life Between the Wars It’s 1933 in Paris, and the années folles are dead. Dada and deco are out, and the Great Depression is in. The luminaries...