Incoherent, creepy and gorgeous: we asked six leading artists to make work using AI – and here are the results

For more than 30,000 years we have been the only art-making species on Earth, give or take the odd paint-throwing Neanderthal or chimpanzee. Art is the oldest and most spectacular triumph of human consciousness, from Lascaux to the Sistine Chapel. But a new generation of artificial intelligence (AI) art software may be about to end that. It will whip you up a Picasso or a Turner in an instant, or apply their styles to any theme you picture, from Liz Truss dancing in a supermarket to a brawl in a 1970s disco.

Stable Diffusion and competitors such as DALL-E 2 go far beyond previous claims for AI art. Easily accessible online, and in that sense open to full public scrutiny, they create precise, rich, convincing images in response to a typed-in text – for example “a sad cat in a mountainous landscape in the style of Turner”, or whatever combination of styles, keywords and subjects takes your fancy. Or you can ask more sidelong and existential questions, such as my request for “a photograph of a human”, which produced a bare-chested man who could be a museum exhibit of early homo sapiens – except for his mysterious earphone-like cables. For the expert there are others: “I’ve been experimenting in Wombo Dream, Midjourney and Google Colab/Disco Diffusion,” says the artist Mat Collishaw.

Until recently, I was deeply sceptical of the idea of AI art. I saw it as hype and casuistry, and with some cause: widely publicised efforts such as Ai-Da the robot artist obviously exaggerate the independence of the machine and play on our fascination with sentient artificial beings. But now the dream is coming true, at least in art. And art is surely one of the most inimitable expressions of the human mind.

Evangelists for so-called “strong AI” – full artificial intelligence that will replicate and exceed the human mind – are fond of making analogies with biological evolution. Over millions of years, mindless cells evolved the human brain; machines are now evolving much faster so why shouldn’t they become sentient soon? The evolution of AI art seems to vindicate that. In 2018, the sale of an AI generated-painting called Portrait of Edmond de Belamy at Christie’s for $432,500 (£360,000) was the latest thing in the field – yet this portrait was crude in the extreme, a pixellated blur easy to dismiss as a pathetic computerised pastiche of Frans Hals. Four years on, the detail and nuance of images produced by the latest AI art generators have grown more impressive exponentially.

What does this mean for art? Is it the end of our run as the only art-making species? Or can humans and machines work together to create something wondrous? To find out, I challenged six outstanding human artists, including three Turner prize-winners, to experiment with AI.

Gilbert and George

Gilbert and George have been flirting with post-humanity ever since they painted their faces and hands silver, like robots, for their 1969 performance-art masterpiece The Singing Sculpture. Their merging of creative identities – “two people, but one artist” – has a similar futuristic radicalism. That extends to the authorship of the Pictures they have made since the 1970s in which they appear while operating the camera and editing its images in a deliberately unexpressive way: always sharp and bright. The panel of four portraits they have elicited from AI is called Gilbert and George by AI, but firmly credited to them as artists.

These images of Gilbert and George are like them, but not. They are clearly not photographs: instead, the software has “painted” them from the information it has, in several efforts with various eccentricities. At the top left, Gilbert Proesch looks as if he’s in a 1960s film by Antonioni or Fellini: at bottom right, both men merge into the same grumpy caricature. In all the portraits, their eyes are cold and distant and odd. This is typical of what happens when you ask a machine to portray a human.

All existing AI art platforms, from the software that stunned Christie’s auction room in 2018 to the disconcertingly impressive Stable Diffusion, are types of “neural network” that excel at machine learning. Neural networks emulate the way neurons fire off each other in the human brain: they are capable of learning when fed with ever-vaster quantities of data. The power of machine learning is seen by some as an epochal breakthrough that makes AI “creative” and could soon lead to artificial consciousness – if it hasn’t already. This summer, Google sacked engineer Blake Lemoine after he claimed its LaMDA chatbot was starting to think for itself.

Others say these machines are only good at the job they have been taught: DeepMind’s AlphaGo can’t turn its skills to Scrabble or Cluedo and self-driving cars have big problems with unexpected situations outside their learning.

The trouble AI art has with depicting eyes, not to mention how many limbs or heads a human being possesses, may back up the sceptical view. The likes of DALL-E 2 and Stable Diffusion have been fed huge amounts of artistic and visual information yet don’t have any knowledge of, say, anatomy. They don’t realise this is a problem because they don’t “think”.

Or do they?

Elizabeth Price

Elizabeth Price takes on such issues in the sequence of 40 AI images she generated from text prompts. Instead of producing a finished artwork, she treated it as an experiment, sharing the results in an apparently casual way. Only when I flicked through the sequence like a slideshow did I see that it is as unsettling as her Turner prize-winning video The 1979 Woolworths Choir. Price engages with the AI as if it were indeed sentient, asking it questions rather than giving it commands, as she tries to get the software to reveal its true self. “I quickly became fascinated by how it was putting images together; how that process differed from the human mind; what it ‘knew’; what it ‘understood’; and how much we could think of its dataset and search modes as a kind of cultural memory.” Price says that if this is her artwork, it’s one that includes the questions she put: one of the limits of AI, she points out, is that it has an unsophisticated and conservative grasp of what art is.

She asks the AI: “What do you understand about love for a parent”; it produces a waxily real, warm yet ever so slightly creepy vision of an embracing family. “Do you understand politics in the UK?” elicits a shrill picture with devastating – deliberate or accidental? – satirical power in which hosts of shouting heads are juxtaposed with multiple union flags. “What do you understand about racial identity” leads to a photographic image of two Black children, one of whom holds up a photo of a Black girl as if she is a missing person. It’s striking, says Price, that the software should assume “racial identity” means Blackness, as if whiteness were the non-racial norm.

Such apparent racism has become a problem with other AI applications such as facial recognition, and reflects the internet data the machines are fed. Price compares the intelligence she questioned to a “collective unconscious”. And it can throw up some seriously weird images as it trawls through digitally archived human memories. She types in the poet Emily Dickinson’s lines about her own future grave: “An Island in dishonored Grass – / Whom none but Daisies – know.” In one text prompt the daisies become beetles, and the computer depicts them as giant shiny creatures in a verdant landscape.

Some combinations of words unleash the truly unexpected from AI art generators, while others result in something flat, or incoherent. It feels as if you are searching for the right key to the vast banks of imagery that exist on the internet. AI in its current state is quite literally an unconscious mind, full of memory, but unable to make sense of it. In the early 20th century, artists and poets inspired by Sigmund Freud sought to release images directly from the human unconscious. That was the surrealist revolution. Could the AI age be fertile ground for a new surrealism in which human artists pry open the digital unconscious?

Mat Collishaw and Polly Morgan

Mat Collishaw and Polly Morgan work in a studio that seems to have its own subconscious. It’s a converted pub in south London, a bit of a fortress from the outside, where on the ground floor Collishaw works coolly with computers and hi-tech toys including a 3D video screen. Collishaw is fascinated by the technology of the image, from early cameras and zoetropes to AI – of which he is an early adopter. But downstairs in the pub’s depths, he tells me, Morgan is busy amid guts and gore, skinning snakes for her taxidermy sculptures.

Not that Collishaw’s art is lacking in surrealism. His current experiments with AI start with 17th-century still life paintings of flowers. He feeds them into the software, then creates text prompts to add in insects. At first, the picture looks charmingly beautiful, then you start to notice more and more insects – which then turn out to be flowers in disguise. Collishaw explains that it illustrates “Pouyannian mimicry, when a flower imitates an insect to attract and exploit other insects which inadvertently propagate the flower’s species”.

It is a metaphor for AI itself – for Collishaw is no techno-utopian. Like the flowers that imitate insects to trick them into propagating their pollen, the big tech corporations attract us with social media and internet searches so they can collect our data. This in turn fuels machine learning, which adds to the digital world’s glamour. Collishaw’s flowers are gorgeous yet deceptive.

When Morgan emerges from her bloody basement in her stained overalls, I suggest she, too, try out an AI artwork, and it arrives a few days later – a taxidermy sculpture strangely transfigured. The snake in it has been cross-bred with concrete. It glistens monumentally beside a female hand whose nails are not just long, but doubled, in one of those creative slips AI can make with human anatomy. Morgan compares it to the surrealist photography of Man Ray – it’s like a decadent remake of his portrait of Kiki of Montparnasse, her eyes closed in dreams, her hand on a carved mask. Except here, the dreamer is a disembodied hand and what it dreams of is an inorganic serpent.

Gillian Wearing

Gillian Wearing uses the fantastical possibilities of AI to create a truly disconcerting image full of unease called Imagined Mask of Joan Crawford as Bette Davis in Whatever Happened to Baby Jane. You start to laugh at the title, then stop when you realise that beneath this brutal and macabre parody of a human face, created by exploiting the inhumanity of AI portraits, is a suffering human being. This is very much a work of art by Wearing that happens to make use of AI: it’s the latest in a series of disconcerting and introspective works in which she has been exploring the nature of masks, and what they say about our public and private selves.

Using one of her own photographs, she has added the mask with DALL-E 2 AI so it seems to grow naturally out of her face, while being contoured to the shape of a skull all too visible under the skin. It uses the distortions that AI can create, yet within a portrait that is human and real. Wearing suggests mortality and madness in a vision of a face eaten away from inside. It’s strange, spooky, funny – yet like all her art it is also about the lumpen reality of being human. Even this mask morphing into two Hollywood legends is ultimately about plain sad facts. You get a sense of loneliness and anguish, crying from inside to outside, soul to soul.

So long as humans are involved, art will be all about us – whatever the technology.

Lindsey Mendick

I asked Lindsey Mendick to try AI because she works in one of the oldest of all artistic media. Pottery was made in ice-age Europe, prehistoric China and every civilisation since. Mendick makes hilarious, lubricious ceramics that swarm with monsters and dirty jokes. What might an artist who is used to delving into wet clay make of an art form that requires you only to type a text prompt on your phone screen?

Turns out she’s a natural. Typing her texts into Stable Diffusion, Mendick struck gold with the words “Las Vegas”, “Cher” and other celebrity names. She started by sending photo-style scenes of wild cavortings in “a Las Vegas buffet”. Then she added the term “impressionist painting”.

That’s how such contemporary masterpieces as Impressionist Painting of Cher and a Werewolf at a Buffet in Las Vegas and Impressionist Painting of Tom Cruise Feeling Sad Eating a Sandwich With a Werewolf in Las Vegas were born. In the funniest, most touching of her kitsch ultra-bright AI paintings of celebrities, a vulpine Leonardo DiCaprio shares a huge seafood platter with a werewolf. Man and monster break bread peacefully.

Like these collaborations between artist and machine, it could be the beginning of a beautiful friendship.

Elizabeth Price: Underfoot is at the Hunterian, Glasgow until 13 April. Lindsey Mendick in Strange Clay is at the Hayward until 8 January. Mat Collishaw’s AI flowers will be at Kew in 2023 and The Gilbert and George Centre opens 2023.