Forests of the Unreal

The unsettling implications of deepfakes remind us that deception in art is no more pervasive now than it has been throughout history.

Screenshot from showing a deepfake of former US President Richard Nixon.

The first moon landing, in 1969, was a failure. After staggering across the lunar dust to plant the American flag, Neil Armstrong and Edwin “Buzz” Aldrin became stranded, their spacecraft having been built to get them there, but not up to the task of getting them back. They were fated never to return to Earth. They died slowly and soporifically, counting breaths until their oxygen supplies ran out. Recently, ‘proof’ of this version of events has surfaced: in November 2019, MIT’s Centre for Advanced Virtuality published a website dedicated to the fictional “moon disaster,” the main feature of which is a video clip, made in 2019, of former US President Richard Nixon delivering his “moon disaster speech.” In this public address, made to seem as if it were broadcast shortly after the landing, Nixon tells America that Armstrong and Aldrin will not make it back to Earth. In July 2020, the video gained traction on the internet after it featured in a documentary by Scientific Americanthat anticipates the future impact of so-called synthetic media. Now it’s an iconic piece of history that never happened.

This slightly blurry, vaguely twitching Richard Nixon is a ‘deepfake’, a simulation generated by artificial intelligence. It was produced by breeding samples of actual video and audio footage of Nixon together with a present-day recording of a voice artist reading a real speech that had been drafted in case of an unsuccessful lunar mission. MIT made the video for educational purposes, to demonstrate the level of realism achievable with current deepfake-generating technology. The most common use of this technology, however, is more sinister; within the last year, non-consensual deepfake sex tapes, usually of female celebrities, have taken the pornography industry by storm.

The unsettling implication of deepfake videos, no matter their application, is that just about anyone, saying or doing anything, can be convincingly faked. Audiovisual ‘evidence’ is no longer just that, but rather part of the elaborate staging of ‘fact’, which, to recall another US president’s enthusiasm for so-called alternative facts, may or may not reflect reality.

This creeping presence of digital imposters in our midst forces us to confront a question that art has circled around, in one form or another, for centuries: what is authenticity? The question is evasive because it really plays out through two different, but related, concepts. On the one hand, authenticity is a matter of representation. The correspondence of the artistic image to a reality, its ability to double the world convincingly enough to transform it, is one of the most basic operations of historical Western art. On the other hand, authenticity is a question of the status of an object – in this case the artwork – as itself legitimate, or ‘real’.

Bearded Man, c.1630, possibly by Rembrandt van Rijn.

In the Dutch Golden Age, these two inflections of the concept of authenticity danced off each other in the studios of the celebrity artists of the day, the most lauded of whom was Rembrandt van Rijn. In Rembrandt’s studio, around fifty pupils and assistants toiled away in imitation of the master’s style. Their ability to move up the ranks, eventually gaining recognition as artists in their own right, was a function of how closely they could emulate, and often directly copy, Rembrandt’s work. Their paintings were not only representations of their world, expressed through the contents of each picture; they were also deliberate representations of Rembrandt paintings. This was a lucrative racket for Rembrandt, who sold many of the paintings produced anonymously by his pupils under his own name. There was nothing duplicitous about this in the early 1600s. It was an accepted practice whose ubiquity, much to the consternation of contemporary Rembrandt scholars, is the reason there are so many ‘fake’ Rembrandts in circulation. Indeed, many of the ‘real’ Rembrandts out there could very likely have been co-painted with pupils, casting doubt on the ascription of authenticity in the first place.

All of this came to a head in August this year, when a painting long presumed to be by one of Rembrandt’s pupils, and therefore ‘fake’, was redeemed from a basement of the Ashmolean Museum in Oxford. Titled Bearded Man (c.1630), the postcard-sized painting is the subject of an elaborate authenticating process, which has involved a dendrochronologist (someone who specialises in dating wood) matching the piece of wood that is the support for this painting to another work confirmed to be a Rembrandt, Andromeda Chained to the Rocks (1630). According to the Guardian story which broke the news that the painting is “possibly genuine,” both works were executed on wood from a single tree felled in the Baltic region between 1618 and 1628. One can imagine any number of situations in which this doesn’t prove anything – Rembrandt could have bought his painting supplies in bulk and shared them with his assistants. But the match forms the basis of the hypothesis that Bearded Man is the real thing.

“The Next Rembrandt,” a 3D printed imitation of a Rembrandt painting.

We like stories of pains taken to the extreme, of superhuman obsession. Even if Bearded Man ultimately turns out to be disregarded as an original Rembrandt, the level of exertion taken to get to that point is its own justification. In 2016, the Dutch retail bank ING, embarked on a similarly fanatical undertaking, only in this instance the objective was to create the ultimate fake Rembrandt. An elaborate marketing exercise, “The Next Rembrandt,” as the project is called, used a slew of algorithms trained on existing Rembrandt paintings to generate a composition that imitates Rembrandt’s style and technique in all their minutiae. The resulting ultra-high resolution image was 3D printed in eleven layers to emulate Rembrandt’s layering of paint.

Although “The Next Rembrandt” fell somewhat flat outside of the Netherlands, lacking the intrigue of the Bearded Man saga, it can be seen as a proto-deepfake. Fake people, like the AI-generated Richard Nixon lamenting the moon disaster, are the not-too-distant relatives of fake art. Both induce a standoff between an idea of the real and its opposite, the unreal, and both demand that we choose between two ethical regimes: one that insists on the real, on a single unimpeachable truth; and one that is hospitable not only to non-human intelligence, but to its possible inorganic, or artificial, instances as well.

This tension between the organic and the inorganic, and the real and the unreal, is neatly dramatised in a 2019 body of work by the Norwegian artist Toril Johannessen (made in collaboration with software developer Sindre Sørensen) titled Skogsaken (The Forest Case). Originally exhibited in parallel at Entrée gallery and Trykkeriet centre for printmaking in Bergen from April to June 2019, Skogsaken revolves around a collection of AI-generated tree images based on the trees in the coniferous forests covering the mountains surrounding Bergen. Displayed in a series of silkscreen prints and an animation shown on a mobile phone, the trees of Skogsaken are both spindly and dense, their branches snaking across the page or the screen in sometimes floating fragments. They seem alien, strangely embossed, catching the light in unnatural ways. And yet, though I can’t quite put my finger on why, they look unmistakably like they belong in Bergen

Toril Johannessen, Vinterskog ved Skomakerdiket, Svartediket eller Melkeplassen, 2019, silk print. Photo: Thor Brødreskift.

Like many other AI-generated images, these tree images were bred using a Generative Adversarial Network, or a GAN, a type of algorithm that pits two artificial neural networks against each other in order to arrive at an output that is as organic-seeming as possible. GANs work by training these neural networks using a large sample of learning images, reference material the GAN will use to generate its own images. For Skogsaken,a type of open source GAN called StyleGAN was trained using 3,400 of Johannessen’s personal photographs of the forests surrounding Bergen. One neural network is in charge of generating new images based on the reference material, and the other is responsible for discriminating amongst these and the sample images, scoring images based on the probability that they are ‘real’ sample images or ‘fake’ images made by the generator network. The output images evolve as the system learns, generating images that are increasingly difficult for the discriminatory network to identify as ‘fake’.

In Skogsaken, what the discriminator cannot possibly know, as it toils away in search of real trees, is that the forests of Bergen are themselves, strictly speaking, fake. They were planted as part of an extensive afforestation scheme in the mid-19th century whose objective, as Johannessen explains in the press release for the project, was partly to improve the appearance of what were thitherto Bergen’s “grey and bald mountains.”

The further back one goes, in art and in life, the less real things seem to be. This flies in the face of our intuition that to find the original we have to go backwards in time, that we have to look for reality in the past. The present may be rife with opportunities to displace or, rather, reinvent reality, but this is merely an echo of an ancient, even primary, displacement that rends the thing itself from its representation. In deconstructionist philosophy, this hiatus is the very operation of language. We can’t get to the thing itself because it is cloned and clothed by signification. Likewise, in a feminist post-humanist ontology, the likes of which is constructed by N. Katherine Hayles in How We Became Posthuman: Virtual Bodies in Cybernetics, Literature, and Informatics (The University of Chicago Press, 1999), the thing and its representation inhere in one another; they are fundamentally entangled.

Orson Welles ends his enigmatic 1974 film F for Fake, which relates the life story of the famous art forger Elmyr de Hory, by paraphrasing an oft-cited Picasso quote. Welles, playing himself, addresses the audience just before bidding us farewell: “Art, [Picasso] said, is a lie – a lie that makes us realise the truth.” He stops short of Picasso’s crescendo, which is “the artist must know the manner by which to convince others of the truthfulness of his lies.” For Johannessen as for Rembrandt, to try to discriminate between real and fake, original and copy, is to miss the point: Art circulates in an economy of unreality which, though unprecedentedly sophisticated today, is no more pervasive now than it has been throughout history.

Toril Johannessen, Normalbaum, 2019, detail. Photo: Thor Brødreskift.