AI Generated Art through a Surrealist Lens

The digital age has brought incredible advancements to the art world from developments in audio/visual technologies and virtually interactive technology to the boom in Non-Fungible Token (NFT) trading. To me the most interesting development we have seen is the introduction of art made by artificial intelligence (AI) programs, a hot topic of current internet discourse and debates. The main question being asked is simply whether these computer-generated images can really be considered art, this question is as much philosophical as it is to do with the law and neither argument is ‘right’. I will argue not just that these images are art but that they are fundamentally surrealist and I will highlight parallels between this new age, digital artform and the twentieth century Surrealist group practices.

My personal interest here lies with the surrealists, their influence on the art world and how their practices can live on in the modern world. I am not a computer person, but I was fascinated by AI art from the first time I came across it, specifically text-to-image pieces; “Text to image synthesis refers to the method of generating images from the input of text automatically” (Singh, Anekar, Ritika, & Patil, 2022, p. 194). Arthur I Miller’s book The Artist in the Machine- The World of AI Powered Creativity (2019) has been immensely valuable in helping me understand the more technical and scientific aspects of the world of AI and specifically it’s relation to art, it’s capabilities, limitations and history. ­

There is a large community of creators who are engaging with AI in their practices, some who create their own programs and others who use readily available ones such as the popular DALL●E●2 and Midjourney to create their work; many members of the community post their work to online photo-sharing sites like Instagram where they use hashtags such as #aiart, #digitalart and #generativeart to network and connect. “Generative” here refers to the type of AI that I will be focusing on, “AI can be broadly defined as computer applications that attempt to emulate human capabilities” (Troshani, Hill, Sherman, & Arthur, 2020, p. 181) but I am writing specifically about Generative Adversarial Networks (GANs). GANs are a type of artificial neural network or more simply a computer that is designed to mimic the human brain’s neuronic structure (Miller, 2019). Miller writes that “Artificial neural networks…need not be extensively preprogrammed…they are based on another sort of [unsupervised] machine learning in which data is fed into the machine with no specific instructions, in other words, the machine learns by itself.” (Miller, 2019, p. 43). GANs operate with the specific instructions to create images based on what the user inputs as well as its knowledge that it has acquired through this machine learning which involves two competing artificial neural networks: the Discriminator (D) and the Generator (G). GANs learn how to both recognise and produce images through a back and forth process between D and G, to simplify this process: if the user were to request the program to generate an image of a cat, the G will formulate a noisy, blobby image (known as latent space) based on D’s preprogrammed knowledge of what a cat looks like, G sends this image to D, D does not recognise this image as a cat and rejects the image, simultaneously D learns more about what a cat isn’t, with this new information G generates a new image and the cycle continues until the latent space resembles a cat enough for D verify that the image resembles cat and the program provides the user with the final product (Miller, 2019).  “The use of AI in the process of creating visual arts was significantly accelerated with the emergence of generative adversarial networks” (Cetinic & She, 2022) which came in 2014 when Ian Goodfellows thought of this breakthrough invention over a pub debate while he was studying at the University of Montreal (Miller, 2019).

 Dr Anahruda Reddy writes that “contemporary AI systems…appear to have imaginations of their own as they newly contribute to human creativity by adding layers of data-driven interpretation, prediction and generativity.” (Reddy, 2022), which brings to light the question of what is creative imagination? “Creativity is the production of new knowledge from already existing knowledge and is accomplished by problem solving” (Miller, 2019, p. 5). In his book, Miller summarises psychologist Graham Wallas’ four stage model of creativity as

(1) Conscious thought;

(2) Unconscious thought;

(3) Illumination;

 (4) Verification (Miller, 2019).

To simplify the AI process of image generation according to this model the user’s conscious thought (1) is processed by a computer program (2), the image is generated (3) and verified (4) by the computer’s D- if it follows the four stages of creativity then why should we not acknowledge the creative process and call this art? Miller highlights (2) unconscious thought as essential to the creative process- “During unconscious thinking, the mind mulls over many different approaches to a problem, using facts stored in our deepest memory” (Miller, 2019, p. 36) or known knowledge.

 The surrealists had  a deep fascination with the unconscious, their main aims can be summed up as a plight to rid their art and literature of conscious thought; the group practiced automatic writing and drawing, painted from dreamscapes, held ‘sleeping parties’ and used Freud’s practice of ‘free association’ to unlock their repressed fantasies, memories and fears in the name of art. In September of 1922, forefather of Surrealism Andre Breton and his wife Simone Collinette hosted a type of séance at their home. The séance was proposed by René Crevel and they would go on to host multiple through to the spring of 1923, eventually being known as “sleeping sessions” thanks to Breton (Bauduin, 2014) . It is important to note that although this precedes the common notion that the surrealist movement was born in 1924 when the first Surrealist Manifesto (1924) was published, Guillaume Apollinaire coined the term in the title of a play that he wrote in the turn of the twentieth century (Ellison, 2021), members of the group that we now know as the surrealists adopted the term later on; many were working and acquainted before the manifesto was published. Attendees include names such as Max Morise, Robert Desnos, Gala Diakonova, Paul Eluard, Max Ernst, Man Ray and others. Activities would include certain attendees slipping into a trance or ‘sleeping state’, they themselves and others made drawings and writings based on the experience. Some members of the group such as Breton and Ernst have been reported as unable to reach these trance states  while other members were adept (Bauduin, 2014); “In one of Man Ray’s photographs of a trance session conducted by the Bureau of Surrealist Research in 1923, the poet Robert Desnos is shown at the typewriter. According to Breton, he was particularly good at entering ecstatic states and bypassing conscious control”  (Turner, 2011).

 Another member, or more so outsider, of the group with a particular skill for entering sleep states was the “near phantom…surnameless Suzanne” (Rosemont, 1998, p. 67). Not much is known about Suzanne, just that she was a ‘sleeper’, she offered her ability to enter sleeping states, like an unconscious muse (Rosemont, 1998). Artists and writers would make work, sometimes automatic writings or drawings, based on Suzanne’s behaviour and accounts of her trances, through the process of drawing on known knowledge through unconscious thought. Suzanne’s participation in these sessions allowed the Surrealists to subvert the four step model of creativity, this process became:

(1)unconscious thought;

(2)unconscious thought;

(3) illumination.

Conscious thought became secondary, if not altogether useless compared to unconscious thought and verification was no longer necessary because what they created through this process were pure interpretations of unconscious knowledge- they were able to achieve their main goals. When comparing these practices, Suzanne and the sleepers become the text prompt, the writers and artists become the GAN and the end result is the same- art produced by interpreting knowledge through means of unconscious thought. Unfortunately, the sleeping sessions ended due to the sessions becoming “increasingly dark in time and even violent” to the point where a small group of members were planning a collective suicide attempt in a side room (Bauduin, 2014). Bauduin begs the question of whether the “emotional or psychological dependency upon Breton” (Bauduin, 2014) pushed the attendees, including Suzanne, to overexaggerate or perform rather than to organically experience these sleeping states; I cannot possibly testify as to whether they personally could have reached a trance like state or not but after witnessing my mother be hypnotized in Ibiza, 2013, I am inclined to believe that it is possible.

In the first Surrealist Manifesto (1924), although it’s a physiological term, Breton defined automatism as “the dictation of thought in the absence of all control exercised by reason and outside all moral or aesthetic concerns” (Breton, 1924) and became an essential part of the surrealist practice through automatic production to scratching techniques such as frottage and grattage where the marks made are largely down to chance (Tate, nd). Frottage, meaning rubbing, was developed by Ernst during the twentieth century; it is a technique that involves laying paper over a textured surface and rubbing a drawing material like pastel or graphite over the paper (Tate, nd). Also developed by Ernst, Grattage involves painting a layer of oil paint, laying the canvas on top of a textured object and then scraping the paint from the canvas to emphasise these textures (Tate, nd).

Chance became a key part of surrealism, most notably in games of Exquisite Corpse which was a favourite at dinner parties but also long distance via post. Although the name is not well known today, the premise of the game is that each participant draws a head on a piece of paper, folds it over so as not to be seen and passes it to the next player who draws the neck and the game repeats until the creatures, or corpses, are complete. When inputting a text prompt into a GAN program, the user will have a general idea of what the image will look like but it is ultimately out of the users hands and down to chance much like a game of exquisite corpse or a gratted painting.

 Mike Tyka, a researcher working with Google started to experiment with artificial neural networks in 2015. Tyka has a PhD in biophysics and studied both biochemistry and biotechnology at the University of Bristol before he became an artist, since then his work has been shown in the New Museum in Karuizawa, Japan and at Ars Electronica in Austria (Interalia Magazine, 2018). Tyka used the GAN program DeepDream (2015) to produce his series Portraits of Imaginary People of which he published a book in 2019. Tyka fed the GAN’s D with thousands of images of faces  from the image sharing site Flickr (Interalia Magazine, 2018), this part of the deep learning process is known as supervised learning. The results of his request for DeepDream to generate images of faces resulted in eerily uncanny portraits,  “the brain instinctively registers that these are not real people” (Miller, 2019, p. 89) yet these faces are compellingly human. That is the epitome of the uncanny, a staple of surrealism. “The 'uncanny' is that form of terror that leads back to something long known to us, once very familiar" (Freud, 1919); the uncanny relates to the two German concepts of Heimlich and Unheimlich. There are no direct translations for these words but roughly, Heimlich refers to a feeling of homeliness and familiarity and unheimlich is the opposite- when combined they form the uncanny which is somewhere between an emotion and an experience, a comfort in discomfort or comparable to that nagging feeling that something is not quite right but not being able to put your finger on it. Obviously, Freud’s uncanny is much more complicated but these concepts form the basis of his theory. Like the story about my Mother in Ibiza Tyka’s portraits are merely anecdotal here, the truth is that all images generated by AI are ‘imaginary’, they do not exist in our physical realm, they are interpretations of the imagination of the person who inputs the text created by a machine. Professor Kriss Ravetto-Biagilio suggests that digital artwork has created a “new type of uncanny experience” (Ravetto-Biagioli, 2019, p. 5), he writes that “the uncanny no longer emerges from our inability to tell what is a machine from what is human…but from the fact that it is no longer possible to make such distinctions.” (Ravetto-Biagioli, 2019, p. 4). A number of AI programs are capable of style transfer, a skill that enables them to create pieces of art in the style of a particular human artist, the images that are created are familiar, the style is recognisable but the image itself, maybe also the subject matter is unfamiliar and there may be some tell-tale signs that it isn’t by the artist the viewer thinks it is which creates this sense of uncanny. The thought that a computer is capable of mimicking human creativity is uncanny, the knowledge that something so inhuman is capable of human tasks is uncomfortable. Juxtaposition is at the heart of the uncanny.

 Some AI artists who are sharing their work online are experimenting with mundane subjects, trying to create the most realistic face or a family of cats wearing socks, others are creating surreal worlds with their own narratives or imaginary creatures. The Surrealists are renowned for creating dreamscapes like Salvador Dali’s Persistence of Memory (1931), uncanny spaces such as Dorothea Tanning’s Eine Klein Nachtmusik (1943) and unusual creatures such as Max Ernst’s Fireside Angel (1937). The surrealists were not tied down stylistically, nor did their work share content or context unanimously in the movement so it is difficult to summarise their interests in terms of what they painted, drew, wrote about and imagined but generally the surrealists are associated with dream/nightmare-scapes, occultism, psychological study and distorted realities- all of which are similarly associated with AI art. The introduction of GANs has allowed users, who may otherwise not be able, to express the thoughts and concepts that they don’t fully understand whether it is visualising a dream or helping them to envision their idea of what a utopia would look like. In this way, AI acts as a surrealist tool, much like the group used psychological tools such as Freud’s theory and practice of free association to aid their creativity. Free association was developed by Freud as hypnotism fell out of his favour, rather than inducing the patient to talking about their trauma, they were encouraged to speak freely and without thought to what they were going to say next (Storr, 2001).

 In the first Surrealist Manifesto (1924) Breton defined surrealism as “psychic automatism in its pure state, by which I proposes to express- verbally or in any other manner- the actual functioning of thought. Dictated by thought, in the absence of any control exercised by reason, exempt from any aesthetic or moral concern” (Breton, 1924) I would argue that the use of GANs is absolutely  a form of psychic automatism, if not the most pure; computers are not able to think consciously and impurify the product of creative unconscious thought. And if Breton himself accepts these expressions in any manner, then AI generated images are fundamentally surrealist. Not only this, AI art has given rise to not just a “new type of uncanny” (Ravetto-Biagioli, 2019) but also a new form of surrealism, one that has the uncanny ingrained into everything that it generates.