From Dalí to DALL·E:
Is AI Art the New Surrealism?
Ralph St Clair Wade (University of Cambridge)
Harold Cohen, Stephanie & Friend, 1991 [Whitney Museum]
Vera Molnar, Interruptions à recouvrements, 1969. [Creative Commons]
Computer Art Has A History
The suspicion that computers could make art or music has been present since their invention. Ada Lovelace, the mathematical prodigy who wrote the world’s first computer programme, believed that her computer might ‘compose elaborate and scientific pieces of music of any degree of complexity or extent'. In reality, it took until the mid 1960s for that potential to emerge, in the computer-generated linework of Frieder Nake and Vera Molnar.
The British artist, Harold Cohen, was hot on their heels. Having represented Great Britain at the Venice Biennale in 1966, he’d gone on to develop a group of computer programs which he nicknamed ‘AARON’. The success of the AARON project levelled some difficult questions at the art world – not least, what the minimum conditions should be for an image to be labelled ‘art’. Cohen himself was conservative in his claims for AARON, which he continued to develop until his death in 2016. In the closing lines of his 1994 article ‘The Further Exploits of Aaron, Painter’, Cohen wrote:
I do not believe that AARON constitutes an existence proof of the power of machines to think, or to be creative, or to be self-aware … [but] if what AARON is making is not art, what is it exactly, and in what ways, other than its origin, does it differ from the "real thing?"
Harold Cohen, Untitled, 1981. [Outland]
New Media Practitioners Are Historically Conscious
For a young discipline, today’s new media practitioners seem to have an evolved sense of their own history. The work of Ben Millar Cole self-consciously creates a conversation between two computing tools of subtly contrasting ages: recurrent neural networks (or RNNs, from the 2010s) and today’s large vision models (LVMs).
Millar Cole begins with a database of BBC sound effects, or more specifically, of their descriptive captions – and sets his RNN to predict further titles. Without a prior knowledge of the capabilities of these different nouns, the results are poetically nonsensical: ‘…Hoods Pausing, Tender Roles’, ‘River Pig loading’, and ‘Comedy Gold, Before Scream’. These surreal prompts can then – as in the case of Hoods Pausing – be fed into a vision model. The combination of these two technologies feels almost like digital art’s answer to mixed media.
The difference between even these two recent tools throws the rate of change into sharp relief. Whereas both RNNs and LVMs have learning capabilities, older systems like AARON had to have every style manually programmed into it. This meant that Harold Cohen, like any magician pulling a rabbit from a hat, could hardly be surprised when AARON revealed something he himself had hidden within it.
Ben Millar Cole, HOODS PAUSING, TENDER ROLES II, 2024. [Palmer]
Post Photography is Surreal
We know that today’s AIs learn, and that whilst they are ‘trained’ and ‘aligned’, they are not manually taught. This means that – like any child – they often come out with things that surprise or shock us. In the case of recent technologies some of these unintended outputs have been extreme – and we, as human creators, have sought to distance ourselves from our digital offspring. In Theme 2: Value-Alignment, we look at the darker side of these visual capabilities. The trouble is – like any parent who demands to know where their child has learnt a swear word – the answer is usually close to home.
Artistic movements have, before now, dragged hidden social prejudices and neuroses into the public sphere. This year marks the centenary of one of the most famous – Surrealism, which began with the publication of André Breton’s Manifeste du surréalisme in Paris, 1924. To the public, Surrealism is synonymous with the theatrically bizarre, and the needlessly whacky. It’s a happy coincidence that the melting clocks and waxed moustaches of Salvador Dali (1904–1989) helped (alongside WALL–E, the loveable Pixar robot) to name our own present-day gamechanger: DALL-E.
The similarities don’t end there. In fact, there was a method to the Surrealists’ madness. They weren’t just making chaotic art to shock the public. Instead, drawing upon contemporary psychoanalytic thought, they made art which drew on the deepest, most instinctual recesses of human identity. Then, and now, this process revealed parts of our individual and social character that we hadn’t realised were there – and more often than not, wished we hadn’t seen. The Surrealists brought Freudian accounts of fetish, masturbation and confused parental relationships to a shocked public. Today’s vision models might be much the same – if it weren’t for armies of moderators in the global south, some of whom report long-term psychological consequences from extended periods viewing graphic or violent content. Even now, some new media practitioners have returned to Freudian themes. The Brooklyn-based Charlie Engman has been making art about his mom since 2009.
The axis of modernity: DALI, WALL-E and DALL-E.
Carl Van Vechten, Salvador Dalí, 29 November 1939. [Creative Commons]; WALL-E (Waste Allocation Load Lifter: Earth-class), 2008. [Pixar]; Homepage of DALL-E 3, 2024. [Microsoft]
Post Photography is Uncanny
It's not just a question of revealing dark secrets about the human condition. Gen-AI images also have a tendency to show us familiar things in a new way. We could call this tendency ‘the uncanny’. That’s another term coined by the father of psychoanalysis, Sigmund Freud (1856–1939), to explain the strangeness of familiar things, seen differently. Freud’s example was the awkward relationship that we have with dolls or waxworks, but it’s easy to see the contemporary relevance of the concept.
It fell to robotics professor, Masahiro Mori to coin the term ‘Uncanny Valley’. This describes the U-shaped curve that emerges when you measure affinity (‘likeableness’) on one axis, and human lifelikeness on the other. He noticed that humans show an increasing affinity to an entity (e.g. stuffed toy) until the entity becomes nearly human (e.g. corpse or zombie) – at which point that affinity turns to revulsion. Clearly, there’s nothing worse than being nearly human, because when total similarity is reached, affinity climbs back up again.
Today’s post-photographers have been quick to capitalise on Mori’s idea. Think about those raincoat-clad forms in Millar Cole’s Hoods Pausing, Tender Roles II – how they oscillate between being reassuringly normal, and freakishly contorted. Or take the disconcerting fusion of human and non-human animals in Charlie Engman’s Horizon Horse. It’s no surprise that Luba Elliot, one of the leading curators in this fast-expanding field, titled her most recent show ‘Post Photography: The Uncanny Valley.’
Left: Charlie Engman, Horizon Horse, 2023. [Charlie Engman]; Right: Karl MacDorman after Mashiro Mori, The Uncanny Valley, 2003. [Creative Commons]
Post Photography depends on Human Creatives
Exactly a year to the day before Luba Eliot’s show opened, the photographer Boris Eldagsen was busy turning down the Sony World Photography Award – arguably the sector’s highest honour. The reason? He refused to describe his winning image as a photograph (made with light) preferring a coinage of his own: a promptograph.
(We interviewed Boris. You can watch a rough-cut here.)
It’s not that Eldagsen thinks that promptography is a lesser category – but, he argues, it’s certainly a separate one. Eldagsen takes a high view of artistic practice, as the defining expression of human creativity. If the very function of art is the expression of human feeling, it is by definition safe from supersession by artificial intelligence.
In common with other leading creatives, Eldagsen sees the capabilities of Gen-AI as being akin to a choir, or orchestra. The role of the creative is as the conductor. Most of us could, by analogy, take up the baton in front of the London Symphony Orchestra and coax out a simple tune. That would be a testament to the expertise of the ‘tool’ before us, and not really a reflection of our own musicality. To get anything worthy of performance out of that vast and variously trained ensemble, however, you’d need to be a professional conductor.
A famous orchestral conductor will gain some of their authority from having listened to hundreds of recordings of the same piece over the preceding decades. They’ll be able to draw on these authorities, referencing them in precise and knowing ways during their own performance. And this – Eldagsen told us – is precisely his own approach. Using an intricate 13-point prompting structure, he coaxes highly nuanced, referential images from readily available models. Like a skilled conductor, he knows his orchestra well, and can extract very different colours from the slightest variation in prompting. It is, he admits, a mode of practice that favours an experienced creative, with a rich understanding of the canon.
Left: Boris Eldagsen, The Electrician, 2022. [Palmer]; Right: Michael Baxandall, Words for Pictures, 2003. [Yale]
Post Photography Depends On Text
Above all, however, promptography represents the triumph of word over image. There’s an academic debate about the relationship between words and images (it starts, in modern scholarship, with renaissance art scholar Michael Baxendall’s Words for Pictures), but in promptography the word has clearly won the argument. It is the governing, authoritative part of the image. The prompt has become, to use the language of the philosopher J. L. Austin, ‘a speech act’. The idea of speaking worlds into being is one with a rich – and sacred – heritage in the Abrahamic religions. The creation narrative in the Book of Genesis sees the world spoken into being – the words that the Judeo-Christian God speaks are at once acts of description and creation. ‘Let there be light’ is the moment when light is both first described, and created.
What do these ancient references mean for today’s post-photography? The challenge for promptographers is that they do not use any word for the first time – by definition, they can’t. They’re creating from existing materials, and not out of nothing. Each word is loaded with prior use and meaning. Is our own language – loaded down by the weight of past use and abuse – up to the challenge? And more starkly, can we as human actors be trusted to prompt images with good intent?
