Concerning the spiritual in AI art
From the age of Spiritualism to the Age of Spiritual Machines
Every work of art is the child of its age and, in many cases, the mother of our emotions. It follows that each period of culture produces an art of its own which can never be repeated.
— Wasily Kandinsky
In 1911, Kandinsky observed much of art becoming soulless, imitating older styles, increasingly mechanical, and specialized. But in a few pioneering artists (including himself), he saw “minds awakening after years of materialism,” as he writes in his essay Concerning the Spiritual in Art [1]. Like science, he writes, art is devoted to seeing the unseen, or the unnoticed. But the artist, he contends, sees the spiritual in the material, where the scientist sees only the observed.
Specifically, Henri Matisse, in his use of color, Kandinsky said, “endeavours to reproduce the divine,” his pictures being “full of great inward vitality” [2]. Paul Klee was similarly making the invisible visible, as he said himself around the same time [3]. Piet Mondrian was moving further into abstraction, distilling trees and seascapes into simple lines and colours; his later minimal, rectilinear paintings attempted to remove all traces of the real world, aiming to depict the universal through just a few colours and lines [4]. Pablo Picasso, too, was deconstructing the picture plane: as Kandinsky wrote, “In his latest works he has achieved the logical destruction of matter, not, however, by dissolution but rather by a kind of a parcelling out of its various divisions and a constructive scattering of these divisions about the canvas” [5]. Contemplating the abstract, Kandinsky said, allowed the artist to see what he called “the in-between” [6].
At the same time, the Swiss psychologist Carl Jung was becoming a pretty accomplished painter himself — he used painting to illustrate his theories about dreams, symbols and archetypes [7]. And in Sweden, Hilma af Klint was making amazing and explicitly spiritual paintings; she said she received messages from beyond, telling her what to paint. But Kandinsky was not aware of either of them, because their work was kept secret for decades to come.
We should therefore recognize, retrospectively, some problems with Kandinsky’s position. For him, the artist is an exclusively male prophet, and by contrast he refers to painting as ‘her’ [8]. He was firmly situated in the exclusively white, European art world, and primarily focused on painting as a medium. But I believe his essay serves as a useful starting point for investigating the spiritual in art a century later, specifically art that engages with new techniques and ideas of artificial intelligence (AI).
In this article, I will mostly use the term spiritualism and not spirituality to discuss AI art specifically in relation to the Spiritualist movement in late 19th and early 20th Century Europe and the Americas. Kandinsky can be included in that movement. I will not explore explicitly religious notions of spirituality. I will refer to machine learning, neural networks, and other aspects of AI, and in contrast to Kandinsky I try to incorporate as many female and non-western voices as possible — although access to the technologies and discourses of AI is still asymmetric. While this article brings together several strands of research I have followed, it is still a new area, and thus any conclusions I draw are tentative.
What is spiritualism in art?
Hilma af Klint, like Emma Kunz and others who came later, uses sacred geometry in her paintings. Sacred geometry, according to mathematician and artist Karen French, is what structures reality: it is the mathematics of the world, it is what we perceive. But she echoes Jung in stating that such geometry comes not from nature but from inside our heads — it reflects a deep symbolism. What differentiates sacred geometry from other geometry, she says, is intent:
What makes sacred geometry sacred is the purpose with which you use it. In mundane geometry, you build a chair so you can sit on it. In sacred geometry, the location, materials and process of construction is important. Every aspect of the way an object comes into existence and is used is sacred. [9]
We might say something similar about art in general: what makes something art is also about intent, placement, materials and process.
“Every era,” according to Susan Sontag, “has to reinvent spirituality for itself” [10]. Kandinsky said that it comes from within, and based on French’s description of sacred geometry above, we could say that this has remained consistent from one century to the next: art “validates that internal landscape” [11].
In machine learning, a similar kind of internal landscape is called latent space, in this case existing inside a machine — for our purposes, a machine involved in the creation of art.
“Every image embodies a way of seeing,” according to John Berger [12]. The image above is from a project by artist Edie Jo Murray, who took some of my AI research as a starting point to create artwork.
Can this image be considered spiritual? It’s not sacred geometry according to French’s definition, in terms of intentionality of the placement of elements. But we might think about the “materials” of the work — in this case including the training set provided to the machine, and data in the form of Murray’s text prompts and programming. And we might look at the process of construction — for example in nearest neighbour algorithms, decision boundaries, feature spaces and classifiers. Any intentionality in this case, I would argue, is inherent to the training set.
If we wanted to look for Jungian archetypes, we might consider the symbolism of a room, such as you might encounter in a dream: a room symbolises individuality and private thoughts, and a window symbolises “the possibility of understanding and of passing through to the external and the beyond”; it’s also a symbol for communication [13]. Windows are square, and squares, French tells us, do not occur much in nature; it thus symbolises rationality, and consciousness.
When we respond to something in art, it’s because we recognise something in it, or we believe in the belief system of the artist
But this is only one interpretation; we might also look to the machine’s human collaborator, Murray. In this project, she had an explicitly “spiritual” intention, if that’s the right word, based on our discussions during the project, and she selected a subset of images with that in mind. For us as viewers, when we respond to something in art, it’s because we recognise something in it, or we believe in the belief system of the artist [14]. So in defining what is spiritual in art, in addition to the artist’s intent and and the content of an image, we can add the viewer’s reception and interpretation, which will also be linked to their context and their state of mind.
A nice example is UUmwelt, a 2018 exhibition by Pierre Huyghe at the Serpentine Gallery in London. He drew on research in which scientists recorded people’s brain patterns using functional Magnetic Resonance Imaging (fMRI) while showing them pictures of specific objects; the scientists then asked the participants to imagine the same objects, again recording their brain patterns, then using those to construct new images. The results are uncanny, flickering images that look somewhat like the real objects but have a distinctly dreamlike quality — which matches imagery created by Google’s DeepDream, and work generally created using a form of machine learning called Generative Adversarial Networks (GANs).
“GAN art” has indeed become a sub-genre of AI art, and since GANs were been made readily available online for anyone to use, this has resulted in a plethora of work with varying quality and intent (i.e. mostly bad). Huyghe’s work, however, displays something like the “internal truth” Kandinsky spoke of, I believe, making invisible (latent) mental processes visible, to reveal that “in-betweenness” that Kandinsky referred to.
Regarding the context of the viewer, Huyghe used the whole environment of the gallery to create an uncanny atmosphere around the flickering images, in which live flies were bred, and paint was scraped off of the walls to expose previous layers.
Spiritual=irrational
A scientific interpretation of Huyghe’s images would be concerned with fidelity to the original objects, and the feasibility of such a process of mental reconstruction. But Kandinsky was against rationality and scientific measurement, seeing these as opposed to the spiritual. Georges Braque, who developed cubism along with Picasso, said, “The only thing that matters in art is what cannot be explained” [15]. Another artist, Sol Lewitt, said something similar about conceptual art in the 1960s: “Conceptual artists are mystics rather than rationalists. They leap to conclusions that logic cannot reach,” and “The artist may not necessarily understand his own art” [16].
Murray named her project from the concept of the “third space” developed by ancient Persian philosopher Suhrawardi: it describes a space between the intellect (the rational) and the senses (the phenomenal or pre-conscious); he described “the land of non-where” that exists invisibly alongside the material world [17]. This concept provides a bridge between the internal, latent space of artists’/machines’ intentions, and visions that come from somewhere else, as Hilma af Klint claimed to experience. Such visions can be induced by scientists as well as other external forces [18], but perhaps the source does not matter: in terms of agency or intent, what extrinsically motivates us to do anything? Might we, for example, consider how computers might be using us instead of the opposite, particularly in light of the increasingly rapid development of AI [19]? One of the functions of artists, according to critic Jack Burnham, is indeed to specify how technology is using us [20]. Might another function be to act as a “medium” in terms of a spiritual communicator with other intelligences?
Experiences of alteration, loss of agency, and thoughts or forces believed to come from outside are common across cultures and time periods, according to psychiatrist Quinton Deeley; they are “foundational experiences which organised cultures in ways of life” [21]. In many contemporary societies, surrendering to such irrational experiences acts as a counterbalance to the techno-scientific rationality that orders our lives.
Symbolic and prophetic qualities in AI art
A fourth kind of spiritualism, in addition to that which arises inside an artist’s mind, that which might come from an unseen entity or place, and that of the viewer, is what Kandinsky called “the inner light of things,” which he saw artists like Matisse and Picasso draw out in their depictions of rather mundane objects.
Artist Julian Tapales trained a neural network to transform an object called a Bulul — a figure used in rituals in his native Philippines to dispel negative energies and protect the rice crop (Fig.3). Drawing from local materials as a training set, the network continuously manipulates the object’s form and texture as it turns on screen, accompanied by traditional music from the object’s region. In terms of intent, Tapales has taken an already symbolic object and given it a dynamic digital presence, in collaboration with AI. The original object was not static to those who believed in its symbolic or spiritual power; but dematerialised, decontextualised and set in constant motion and transformation, does it retain or gain any spiritual qualities as an artwork, when presented in a very different context (the Royal College of Art in London)? In Tapales’ words:
This work questions the ontology of analogue and digital materials, asking to what extent symbolic figures and objects are able to create a meaning beyond themselves and to have effects on their material context. Is the digital representation inseparable from its material counterpart, or has it ‘learned’ enough to acquire an agency of its own?
We might look at explicitly symbolic objects, which, like symbolic language (as in poetry) are indexical — they point to something else. This concept is encapsulated in the ancient tradition of the memory palace or memory theatre, explored in great depth by Francis Yates. In it, distinctive objects are created (often only mentally) as mnemonic devices to help remember particular concepts (for example the. names of gods); this was done in primarily oral cultures. Placing the objects in specific positions in a particular space can help someone remember a whole sequence of things, by recalling the objects’ positions when mentally moving through the space.
Applying this concept to machine learning, such objects are like a training set, placed within a latent space. But in a machine learning model, the latent space is less a theatre and more of a labyrinth, with multiple layers and dimensions that are difficult for humans to comprehend. And in mental recall, things retrieved from memory are never exactly the same — the act of recall changes memories. This merging of stored memories with each other and with new data thus results in transformations, producing imagery that is often ambiguous and sometimes grotesque, as in Jessica In’s AI transformation of Piranesi’s drawings shown here. But AI artworks are mostly presented on-screen, and mostly in isolation; an area for further development with regard to spiritualism might be to place them in positions and contexts which might enhance their symbolic, indexical attributes.
The notion of art not as an object as such, but as a pointer or a position, is extended when we consider the prophetic. Kandinsky contended that new forms of abstract art, such as his own, have “a deep and powerful prophetic strength” in their ability to symbolise spiritual truths [22]. This was utterly tied to his contemporary era, but at the same time bore the marks of a previous period — for him this was “Primitive” art and its methods. He drew on the art of Africa and Oceania — colonialism brought about ethnographic museums to Europe in the late 19th and early 20th centuries, and there Kandinsky encountered other cultures’ art. For him and his circle, such art stood opposed to scientific rationalism, and adopting some of its features helped to achieve the kind of universal spiritualism that he saw as transcending any single culture. For him — as for scientists of the time — the primitive was seen as less developed, and hence older, than contemporary European culture. (This attitude broadly resulted in genocides, colonisation and extraction of natural and cultural resources.) For artists like Kandinsky, the primitive represented a more childlike state, free from rationality, but also a time when humans were closer to nature, a time referred to in the myths of many cultures as a “golden age” when God and the world were united, and to which the world might return one day [23].
Contemporary philosopher Federico Campagna differentiates the primitive from the archaic, the latter term referring to a period after a particular civilisation ends, and before a new one begins. He equates this period with adolescence — a time in between childhood and adulthood [24]. He specifically identifies the time before World War I (when Kandinsky wrote his essay) as such a time: subsequent developments in the 20th Century, he writes, solidified a reality system which values rationality, measurement and classification, dominated by science and embodied in computers. He further believes that we may now be seeing the beginning of the end of this era, moving into an archaic period of transition [25].
We might regard the present as a “primitive” phase of experimentation — hence all the bad art being produced
In this context, we might situate AI as an alternative means of computation to older, more rigid algorithmic means, in so far as it is able to learn from both training sets and new data input. As such, AI systems are increasingly regarded either as valuable collaborators, or as potentially dangerous overlords if they attain a self-sufficient level of “superintelligence”. In terms of their role in producing art, we might regard the present as a “primitive” phase of experimentation — hence all the bad art being produced. Regarding their role in relation to the spiritual, this is only the very beginning, but artists like Murray and Tapales are boldly exploring this new terrain.
In one sense, inventor Ray Kurzweil predicted something like this in his book The Age of Spiritual Machines [26]. He defines the spiritual as “a feeling of transcending one’s everyday physical and mortal bounds to sense a deeper reality” [27], equating it with consciousness more generally. In his belief that computers will attain something like human consciousness, he extrapolates that they will be able to experience spiritual states in a similar way to humans.
But it is far from clear at this point how much longer either humans or computers will survive. And arguably, Kurzweil’s vision is restricted to prediction — as an extrapolation of current trends — and not prophecy. To be prophetic, according to Campagna, means to step outside the current reality system, and to speak from another position. In this sense, both AI and artists seem more suited to take such a position. Artists are outsiders by choice and by definition, drawing attention to things we take for granted.
How exactly is such prophetic work done? Campagna shows one way, by creating different character types to construct a narrative, which will in turn serve as a future memory in a forthcoming civilisation. Such a narrative need not be truthful, he contends [28]. This approach prompts us to think, what stories do we wish to send to the next civilisation? Which memories to preserve, which to discard, which to change?
Campagna put his approach into practice in his PhD research, by creating a video game (with designer Jelena Viskovic) called Lamassu, named for an Assyrian god comprised of different animals. In the game, the player can be one of four characters, and each sees the same game-world from a different perspective, each also affecting each other’s experience. Only one character — a rock– can win the game.
Another example is Everest Pipkin’s 2020 project Lacework, in which he used a freely available dataset comprised of one million three-second videos showing different types of actions, from waterfalls to people exercising. Each video was classified with a one-word verb, with the help of people hired using Amazon Mechanical Turk, and the entire dataset is intended to train AI systems to classify and predict actions from input data. But Pipkin didn’t train any AI system; instead, he watched the entire dataset himself (all one million videos) during a months-long Covid-19 lockdown. He then selected 400 videos and stretched each one out to 15 seconds, also scaling it up (using AI interpolation software) and zooming in to a focal point of action.
After such a long process of watching so many videos of actions, he reports, he began to see patterns everywhere: “in the colour of paint and the types of trees, in movement, in texture that could unfold into ever more detail…. I wonder if this is how a sorting algorithm feels.” All the while, his AI software was also spotting patterns as it scaled up images, “imagining detail where there is none,” in a process that “has been described as hallucinatory, which is an accurate marker — it is a recurrent looking, a push in and in for ever more detail which then spirals into something else entirely.” In Kandinsky’s terms, seeing the in-between. “Repetition,” Pipkin concludes, “is devotional.”
Viewing his resulting work, I would not describe as devotional exactly, but is certainly mesmerising. Due to the interpolation between frames and pixels, faces and objects are smeared and flattened into two-dimensional blobs of colour, and even a plane crash takes on an abstract, poetic beauty. Repetition is indeed devotional in prompting spiritual states and out-of-body experiences, across cultures in ritual dances and chanted mantras. But here, mundane actions create a trance-like state — both for Pipkin as the artist and for us as viewers. This is art as investigation — both internal and external. Through such defamiliarization of the mundane, Pipkin enables us to view it anew, from outside — indeed from a machinic perspective, getting close to Campagna’s definition of the prophetic.
Breaking the system
Much of machine learning is concerned with classifying things, and one type of AI art is concerned with challenging or disrupting this process. For example, Georgia Ward Dyer pushed a neural network into the absurd by making drawings of nothing, in order to see what the network would classify them as — an example of what is called apophenia. Specifically, she used automatic drawing, which is not a machine learning algorithm but a method that dates back to Victorian Spiritualism, in which people possessed by external spirits or forces were prompted to transcribe drawings from the beyond. The method was later taken up by the Surrealists, who used it to create images from dreams or pre-conscious states. Ward Dyer exhibited her drawings alongside the classifications bestowed by the system, along with a second print of each drawing, highlighting portions of the image the system identified specifically with the object classifier — portions which signify the essence of the object, in machinic terms. Her abstractions then become symbols, pointers to real objects; and the process becomes one of unlearning — a process in fact used by Hilma af Klint, whose automatic drawing forced her to unlearn her formal artistic training [29].
Much AI art necessitates the use of large training sets to facilitate classification. Artist Anna Ridler usually creates her own datasets, through photography or drawing, before creating her own machine learning models. But like Pipkin, she created Mechanized Cacophonies (with Caroline Sinders) during a Covid-19 lockdown, during which she longed to travel to natural places but was forced to resort to videos found online — in this case of seascapes. The result is, however, even more moving than Pipkin’s, perhaps because it taps into a primal human mesmerisation with complex, natural motions such as the sea, fire, clouds, etc [30]. A similar process of interpolation results in a similarly dreamlike quality as in Pipkin’s work. Ridler and Sinders go further than Pipkin, however, by incorporating sound, also paying attention to the context of the viewer — suggesting the use of multiple devices to view the work, to create a more immersive, spatialized experience. Such an experience lies beyond linguistic description; in Kandinsky’s words, “No such theory of principle can be laid down for those things which lie beyond, in the realm of the immaterial” [31], and artists should “endeavour to awake subtler emotions, as yet unnamed” [32].
For Kandinsky, this meant going beyond merely representing or re-creating natural phenomena. This is why he moved increasingly into abstraction, and for him the ideal art form in this regard is music. He tried to bring the evocative, emotional power he heard from composers at the time such as Debussy, to his painting, for example Yellow-Red-Blue (1925). He began to call his paintings impressions, improvisations, and compositions, expressing a “desire for rhythm in painting, for mathematical, abstract construction, for repeated notes of colour, for setting colour in motion” [33]. But his goal was not to create beautiful, melodic works; in addition to Debussy, he mentions the rather more dissonant composer Schönberg: “His music leads us into a realm where musical experience is a matter not of the ear but of the soul alone” [34].
Kandinsky also mentions poetry and literature: “if the object is not itself seen, but only its name heard, the mind of the hearer receives an abstract impression only, that is to say as of the object dematerialized, and a corresponding vibration is immediately set up in the heart” [35].
In Parts of Speech, created in collaboration with artist Linnea Langfjord Kristensen, I tried to disrupt machine classification, like Ward Dyer, in my case alternating between text and image. Prompted by an algorithm that assembles stock video clips based on text input, I used as input a description of the system itself, provided by its author (the system was Text2Video by Zack Lee). I sent the video generated by the system to Langfjord Kristensen, who wrote a poem based on the visuals; I then fed that poem back into the system to generate new video. Finally, I passed the new video onto an AI system that generated bespoke music from it, and combined the final output into a two-minute video, processing it to be one-bit black-and-white, in reference to the way machine vision typically works [36].
I imagined this process as a multi-part conversation; conversations have been central to AI, and computing more generally, ever since Alan Turing’s famous test of whether an unseen conversant is human or not. Artist Behnaz Farahi provides another nice example in her project Can the Subaltern Speak?, in which two masks use mechanical blinking eyes producing morse code, and an AI system to attempt to interpret and communicate, developing their own language in the process.
In 2016, three MA students of mine at the Royal College of Art used an AI system as a “non-human mediator” for their collaboratively written dissertation about nonhuman intelligences. They used the word2vec algorithm that forms syntactic and semantic relations connected by vectors, treating it as a “weaver,” to synthesise new connections between their individually-written texts, which they might find themselves.
Since then, generative AI systems have made significant advances in producing texts that are indistinguishable from human-written text, again based on huge training sets, and they are starting to produce literature, poetry, and even philosophy [38]. Allado-McDowell shows GPT-3 to produce convincingly prophetic, even spiritual, text [39]. But Murray and I have been using it instead to produce texts that are alternately absurd and profound, for the Third Space AI project.
I have proposed a conversational model for AI art, which contrasts language-based description with actions or images, linked by feedback loops [40]. While computers are faster in searching text than humans, humans are much faster at visual search. But the boundary between text and image is not clear: for example in terms of visual languages. In poetry, the visual presentation, multiple meanings of words, and pauses in between them, are just as important to its meaning.
But “meaning” is not a clear concept either: if computers use exclusively rational means to locate it as precisely as possible, humans easily find connections without any rationality whatsoever — including spiritual ones. Intuitions such as first impressions can be powerful; curator James Brett describes moving through an exhibition as I often do: “If you walk through the show fast… it makes sense — because visual language moves much faster than a verbal one” [41]. In Kandinsky’s words, “We must use our first impressions, lest imagination take over.”
Ridler undertakes a three-way conversation in her version of Edgar Allen Poe’s story ‘Fall of the House of Usher’. Taking a 1929 film version as a starting point, she hand-drew the first 200 frames of the film, then used them as a training set for an AI system to render the rest of the film. In the resulting film, the images become increasingly abstract as the system had progressively less data to work with. The content of the film, though, carries the work beyond a simple technical experiment: Poe’s was a story in the Spiritualist tradition of its time. The eponymous house is alive, and linked to its local landscape. The protagonist is a painter, and his paintings come to life. In abstracting the film through machine learning, Ridler not only evokes the ghostly content of the story, she evokes the “ghost in the machine” of machine learning.
More importantly, Ridler’s work illustrates what writer Joanna Zylinska proposes: that we depart from the idea (held by Kandinsky) of the artist as singular genius. All work, she says, is a product of a technical relation, and art has always been produced with the help of all sorts of technologies and machines. In addition, all tools (including language) can be regarded as “artificial intelligence” according to philosopher Vilem Flusser. He asks, Is the eye a tool or is the camera an organ? Both of them, he contends, have a kind of “program”. In this sense, humans too are machines to some extent [42]. We have always, says Zylinska, been artificially intelligent.
In Rehearsal by Oliver Smith, a computer looks at an interior scene and attempts to draw its contours. Its input comes from a camera, and its output is projected back onto the very scene it observes, causing a feedback loop which distorts the drawing process ever further. In a companion work, a computer reads out a text (a line from the story ‘Funes the Memorious’ by Jose Luis Borges) in an empty room, this time using a microphone and speaker for feedback, similarly misinterpreting the text in its iteration of re-creating it. Again, this is a form of disrupting machine learning (through a machine’s conversation with itself), through the use of positive feedback to create unexpected outcomes, but Smith goes a step further to impose the machine’s results and process onto the real world, in the process making us question our own process of memory formation.
Questioning artificiality, intelligence, art
Having formulated a definition of spiritualism and applied it to AI art in terms of irrationality/ineffability, symbolism, prophecy, and disruption, I now reflect on this to interrogate some of our assumptions about AI and art.
Tapales’s Natura Naturans, for example, brings to mind the question recently asked by Joanna Zylinkska: Who and what is art for? [43] Tapales gives one answer which directly echoes Kandinsky: “this work revisits the vexed question: whether the task of art is to represent nature as it is, or to replicate the generative activity that constitutes the essence of nature itself.” For Kandinsky, as for Mondrian, the “essence of nature” is indistinguishable from the spiritual. And here we arrive at another definition of spirituality — one which sees “intelligence” as something not only limited to humans, or machines: it might be found in a house, a landscape, or the “inner light of things” which Matisse evoked. If we take Tapales’s question at face value, then such intelligence/essence/spirituality is to be found not only in “spiritual” works, but also in the process of creating them, whether in Pipkin’s repetitive focus on mundane details, Ward Dyer’s disruption of systems created for another purpose, Ridler’s hand drawings (exhibited alongside or separately from the digital works), or Smith’s machinic feedback.
Smith’s work indeed addresses Zylinska’s other question — who is art for? His Rehearsal was never exhibited, only set up in an empty room and documented; so besides this documentation, its viewer was also its creator — the computer. The work is in the process, which we humans can only observe from outside.
But this is a paradox, for the spiritual is identified with the immutable; or perhaps more accurately, both the spiritual and the scientific seek out immutable “truths”, the difference being that spiritual (and artistic) practices aim at the immeasurable, where “science sees only the observed”, to again quote Kandinsky.
One artistic way of approaching this is through a method developed by Juliette Pénélope Pépin called poetic objectivism: an attempt to remove any subjective judgement of the artist by revealing links that are unnoticed or deliberately hidden, a process she describes not as unlearning (as with Hilma af Klint) but “[Re]-learning with the unchangeable” [44]. Here we can identify a role for AI, for example in the Latent Semantic Analysis used by Baum, Leahy and Walker, which
generates a reading that is alien to humans. These insights are impossible to find by hand and obscured by readers patterned by human languaging processes. This is a paradoxical perspective of being post-language and pre-human. In the reduction of human language to numbers-in-space the [vector space model] looks for meaning on the informational level. Yet as we have started with human language our numbers and spaces are only the vectors of thought bound by the syntactic structures of human language. [45]
For her part, Pépin applies machine learning to reveal similarly hidden, “in-between” spaces in natural imagery, specifically “organic movements that are still little known to the human eye”. She writes, “I want to give a glimpse of non-human visions between dream and nightmare,” echoing what many people, I believe, feel about DeepDream-type GAN imagery [46]. Her project, like many discussed so far, harnesses the database as a source of artistic and spiritual inspiration — a practice, she observes, that has been little-explored in contemporary art but is “rich in poetic potential” [47]. Like Pipkin, Ridler, and Baum et al, she æstheticises not only the database or archive, but the artistic research process also, and as a result, the spiritual might be found in either of those as well as in “final” works produced.
Pépin’s work, along with Baum et al, also explicitly points to the more expansive definition of “intelligence” to include the natural as well as the artificial — indeed questioning the historical opposing of these terms. According to philosopher Henri Bergson (writing at Kandinsky’s time), we have exaggerated confidence in the individual mind; instead, intelligence is “as vast as reality” [48]. Rational intelligence can only understand inert, not living, matter, he contends. This expanded definition locates intelligence in the creativity of an organism’s exchange with the environment, as Alfred North Whitehead conceived it; or in J.J. Gibson’s identification of possibilities in an organism’s perception, movement and action as environmental affordances.
As Zylinska notes, artistic works long before AI have been produced in conjunction with nonhuman agents such as viruses, organic and non-organic substances [49]. And as Michel Foucault proposed, “the author should rather be understood as an ‘author function’” [50].
Here I would also note the work of Jenna Sutela, who used AI to creates quasi-linguistic messages from a bacterium, an explicit attempt to “connect with a world beyond our consciousness” by viewing the machine as shaman or medium. She drew on the Martian language, as interpreted by a human medium in the 19th Century, combined with the movement of bacteria believed to have lived on Mars and travelled here on a meteorite — an “intelligence” that is literally alien and radically other. The results is a new language not comprehensible to humans.
If intelligence is “natural” and more-than-human, and if we have always been “artificially intelligent” through our relation with our tools, as per Flusser and Zylinska, then what is “artificial”? As Baum et al note, the “Nonhuman Turn” dissolves this old dualism. I would point to the natural foundations of computing: since the hardware of AI is formed from materials of the Earth, then in using it, we implicitly form a relationship to these natural materials, and the places and ways they are extracted. In many indigenous cultures in the Americas, stones are considered ancestors, and speak through people.
In terms of software, we could point to algorithmic processes that can be found in nature, even in the origin of life itself. What is artificial, I would argue, is when processes or practices become sufficiently detached from natural ones — the prime example being our current system of digital time. While based on natural processes (oscillations of atoms or quartz crystals), by reducing time to a linear number measured in milliseconds or nanoseconds, we lose our connection to natural rhythms [51].
Ridler takes up this topic in her work Circadian Bloom, which revives the flower clock, a living garden which tells time through plants’ circadian rhythms — specifically, flower species that bloom at different periods of the day. Ironically, the flower clock was developed by Carl Linnæus, whose taxonomic system for classifying natural organisms is still used today, and is, arguably, thoroughly artificial.
Ridler sites her work on screens in a real garden, the GAN-generated imagery speeding up the natural process undertaken by the surrounding real flowers. There’s the difference: real flowers in a real garden, as against their on-screen representation. Is everything we see on a screen artificial, or merely a “virtual” representation of the real? What about a printed photograph, or a flower viewed through some other device such as a telescope or microscope? I would reply again that it is when such representations become sufficiently detached from the natural that they might be regarded as artificial, but making such a distinction is ultimately subjective and semantic. We refer, for example, to “artificial neural networks” as those which are simply outside the ones in our bodies, but are no less real.
Sofia Crespo uses artificial neural networks to confront Linnæus’s legacy directly in Artificial Natural History, producing what she calls a “natural history book that never was”. Linnaean taxonomy is part of what she calls “the very renaissance project of humanism” which not only aims to scientifically classify all living things, but counterposes them against human exceptionalism. With the help of AI, she disrupts the taxonomic system and our desire to categorise and classify everything, by creating creatures that resist such classifications — again ironically, since such neural networks inherently rely on classifying incoming data as either one thing or another [52]. (Her creatures also evoke genetically modified organisms.) Further dissolving the natural/artificial distinction, she imagines new “natures” created with such technologies: “Starting from the level of our known reality, we could ultimately be digitizing cognitive processes and utilizing them to feed new inputs into the biological world, which feeds back into a cycle. Routines in artificial neural networks become a tool for creation, one that allows for new experiences of the familiar.”
Returning to Kandinsky and spiritualism, Crespo, like most of the other artists mentioned here, evokes the ineffable, spiritual nature of things by exploring “the in-between”, taking as subject matter organisms which are at once familiar and alien, combined with our inner, dreamlike interpretations of them, and counterposing the inherently rational “nature” of computers with the rather more irrational possibilities brought about by “artificial intelligence” which can distort, learn, and even teach us something new, by enabling us to approach that which we think we know, like natural organisms, as something new.
Stepping inside AI
Standing opposed to western “naturalism” which privileges scientific methods for understanding nature and humans’ place within or beside it, animist traditions see a continuity between humans and nonhumans. In particular, Japanese Shinto does not regard a strict separation between the scientific and the spiritual, nor between mind and body. This extends to both living and nonliving things, seeing both natural and technological phenomena as divine, embodied with human characteristics. Anthropologist Anne Allison likens contemporary Japanese techno-animism to an “animist unconscious” mixing advanced technologies and spiritual capacities [53].
Artist Yena Park was born in Korea and grew up in Japan, and draws specifically from Shinto in her film Why you should put yourself into a block of tofu. The title refers to the Hari Kuyo ceremony, explained in the film’s voiceover: “Every year, house-keepers take their sewing needles to shrines, and place them into blocks of tofu, acknowledging the strains and stresses of the tools of their craft.” While interacting with a robot in alternately friendly and antagonistic ways, in the film Park takes what she calls an anti-systematic approach, alternating also between Japanese and English in spoken narration and on-screen subtitles, and she additionally mixes Shinto with the western philosophy of Gilles Deleuze.
Objects, she says, have been “controlled and demoted by our system of language, subjected to a status operated on by linguistic code… The system of language subjugates reality to only what it can capture. Systems, through the production of interiors, always create exteriors, considered or not. Systems cannot capture the infinite.”
This echoes Campagna’s prophetic approach, which seeks to step outside a reality system. “If systems are fed our already coded reality,” Park says, “what is returned cannot step beyond our abstractions. Instead, we can only think through what we know to truly exist. To form new perceptions of the world, we should construct and experience new perceptions of reality.” But paradoxically, for her this means investigating “the subjective reality afforded to objects by their material constraints” — to become an object, and explore from within. And here we can see echoes of Kandinsky’s reference to “the inner light of things”, manifested in various of the works discussed previously. To step outside the world in Park’s approach, one needs to step inside and inhabit another’s body, as kami spirits are believed to inhabit all things in Shinto thought.
Is this similar to what Pipkin does when immersing himself in a dataset, in order to think like an AI system? Or Ridler or Pépin, in laboriously constructing their own datasets of particular objects? (Shinto is also part of Pépin’s research.) The kami spirits are believed to also possess human beings, and speak through them. And they are seen as inhabiting machines. In both embodiments, they are capable of both benevolent and destructive deeds, thus are given offerings and praised in ceremonies.
When audio recording technology was invented, it sparked “anxieties about the spirit world embodied in the very matter of electrically animated objects”
When audio recording technology was invented by Thomas Edison in 1877 (the height of Victorian Spiritualism), it sparked “anxieties about the spirit world embodied in the very matter of electrically animated objects” [54]. Animation and animism share the same root; as Park points out, language can either limit or liberate us from unfolding different meanings. Campagna views language similarly: the reality system he sees dominating our world since Kandinsky’s time relies on absolute language which measures, classifies and separates things into their “essence” — of which Linnaean taxonomy is a prime example. But like Kandinsky, he holds up poetry for its symbolic power to transcend such rigid categorisations, more suited to another, magical reality system which regards instead the essence of things: not what they are, but that they are [55].
Now that machines can write increasingly convincing poetry, and create poetic imagery, we see similar anxieties, and potential, as in the equally magical technology of Edison. And in AI animations, as created by GANs, we might glimpse the spiritual “in-between”, in between frames and objects, and in between AI and ourselves. With computers now being used in less rational ways, perhaps we see “minds awakening after years of materialism”, as Kandinsky described, with artists — both human and not — leading the way.
Acknowledgements
This article is adapted from a talk I gave at Korea Advanced Institute of Science and Technology in December 2021. Thank you to Jinjoon Lee and KAIST for giving me the opportunity to begin researching this fascinating area. I am grateful to all the artists who graciously shared their work and made suggestions, to Edie Jo Murray and Linnea Langfjord Kristensen for collaborating on projects, and to my students, who always inspire, and who carry theory and practice ever further.
Footnotes
- Kandinsky, W. (1977 [1914]), Concerning the Spiritual in Art, p.43. Tr. M.T.H. Sadler, Dover Publications.
- Ibid., 76. See e.g. Matisse’s painting Goldfish (1912).
- Klee, P. (2013), Creative Confession and other Writings. London: Tate Publishing. See e.g. his painting Foehn in The Garden of Franz Marc, (1915).
- See e.g. his painting Tree (1912). His ideas are detailed in Holtzman, H. and James. M.S., eds. (1986), The New Art — The New Life: The Collected Writings of Piet Mondrian, New York: Da Capo Press.
- Kandinsky, Concerning the Spiritual in Art, 77. See e.g. Picasso’s painting Man with a Clairinette (1912).
- Kandinsky, W. (1947 [1926]) Point and Line to Plane, 11, Detroit: Cranbrook Press.
- Jung, C. (2009), The Red Book: Liber Novus, W.W. Norton.
- The volume’s translator may share some blame for this.
- In The Medium’s Medium: Spiritualist art practices from the turn of the century and beyond (2019), 13, London: Gallery of Everything. This accompanied an exhibition of spiritualist art at Gallery.
- Ibid., 48.
- This according to psychic Avril Price, in Ibid., 49–50.
- Berger, J. (1972), Ways of Seeing, 23, New York: Penguin.
- Cirlot, J.E. (1962), A Dictionary of Symbols, 262, Tr. J. London: Sage, Routledge & Kegan Paul.
- According to Museum of Everything founder James Brett in The Medium’s Medium, 17. Museum of Everything (and its offshoot Gallery of Everything) is devoted to ‘outsider art’.
- Braques, G., Cahier (notebook), 1917.
- Lewitt, S., ‘Sentences on Conceptual Art’, 0–9 no. 5 (January 1969)
- In Campagna, F. (2018), Technic and Magic, 9, London: Bloomsbury.
- So claims psychiatrist Quinton Deeley in The Medium’s Medium, 16.
- There is a great deal of writing on how technology companies use humans, for example Lanier, J. (2011), You are Not a Gadget, New York: Penguin.; and some research on the addictive qualities of social media, e.g. Greenfield, A.(2017), Radical Technologies, London: Verso. An increasing number of voices raise concerns (and hopes) about the emergence of ‘superintelligent’ AI — see e.g. Vickers, B. And Allado-McDowell, K., eds., (2020), Atlas of Anomalous AI (Ignota Books.
- Burnham, J. (1968), ‘Systems Esthetics,’ Artforum, Sept. 1968.
- In The Medium’s Medium, 15.
- Kandinsky, Concerning the Spiritual in Art, 49.
- Campagna, F. (2021), Prophetic Culture: Recreation for Adolescents, 119, London: Bloomsbury.
- Ibid.
- Ibid.
- Kurzweil, R. (1999), The Age of Spiritual Machines, New York: Viking.
- Ibid., 151.
- Campagna, Prophetic Culture.
- The Medium’s Medium, 19.
- The term ‘mesmerism’ as a synonym for hypnosis, derives from Franz Anton Mesmer, an 18th Century pioneer of the Spiritualist movement in Europe.
- Kandinsky, Concerning the Spiritual in Art, 63.
- Ibid., 44.
- Ibid., 79. I note here banal works such as Play a Kandinsky by Google Arts & Culture are probably not what the painter had in mind for evoking the spiritual; for that, his original painting suffices.
- Ibid., 74.
- Ibid., 68.
- Computer vision typically uses a simple threshold filter on input imagery, turning pixels black on one side of the threshold, white on the other, in order to more easily identify edges and outlines for classification.
- Leahy, R., Olesen, A. and Walker, R. (2017), Weaving Worlds: A collective venture into human-nonhuman entanglements, MA dissertation, Royal College of Art.
- See e.g. Vickers and Allado-McDowell, Atlas of Anomalous AI; Allado-McDowell, K. (2020), Pharmako-AI, London: Ignota Books; Steyerl, H., Department of Decentralization, and GPT-3 (2021), ‘Twenty-One Art Worlds: A Game Map’, e-Flux Journal 121 (Oct 2021).
- Allado-McDowell, Pharmako-AI
- Walker, K. (2019) A conversational framework for machine learning. Human-Centered Machine Learning Perspectives Workshop, In CHI ’10 Extended Abstracts on Human Factors in Computing Systems (CHI 2019); Walker, K., ‘Learning from Machines: Conversations with artists about machine learning. Medium, 27 Jan 2020.
- The Medium’s Medium, 19.
- In Agüera y Arcas, B. (2020), ‘Art in the age of machine intelligence’, 117, in Vickers and Allado-McDowell, Atlas of Anomalous AI.
- Celis, C. and Ortuzar Kunstmann, P. (2021), ‘We Have Always Been Artificially Intelligent: An Interview with Joanna Zylinska’, Culture Machine 20.
- Pépin, J.P. (2021) Biocœnose project proposal.
- Baum, Leahy and Walker, Weaving Worlds, 10.
- Pépin, Biocœnose, 4.
- Ibid., 7.
- Bergson, H. (1911), Creative Evolution, 200, tr. A. Mitchell, New York: Henry Holt.
- Celis and Ortuzar Kunstmann, ‘We Have Always Been Artificially Intelligent’.
- Ibid.
- Walker, K. (forthcoming), ‘The time machine stops’, in Majaca, A. and Pfeiffer, L. (eds.), Incomputable Earth: Digital Technologies and the Anthropocene, London: Bloomsbury.
- This is the case in supervised machine learning using training sets.
- Allison, A. (2006), Millennial Monsters: Japanese Toys and the Global Imagination, Berkeley: University of California Press.
- Marenko, B. (2014), ‘Neo-Animism and Design: A New Paradigm in Object Theory’, 219, Design and Culture 6(2): 219–242.
- Campagna, F. (2018) Technic and Magic, London: Bloomsbury.
