Fully Automatic Dreams
essay by Dr Stephen Haley
“Do androids dream of electric sheep?” wondered Phillip K. Dick in his eponymous novel of 1968.[1] Artist Sam Leach has a likely answer - “Well, yes, probably - if you instruct them to.”
Leach’s exhibition Fully Automatic evokes many dreams. Digital dreams, utopic techno dreams, musings, wonderings, the dream of landscape painting and, perhaps, even the odd nightmare. These paintings are strange imaginings from an emergent digital age, an age truly unprecedented in human history, so entirely unique to our time, that these works would be impossible in any other era. Yet, despite being spun from the gossamer stuff of dreams, just like the digital itself, these otherworldly images are based in hard materiality and process.
To the nuts and bolts then. The artist worked with physicist Dr Matthew McAuley from Belfast to build an Artificial Intelligence capable of inventing images. Using algorithms from an Open Source code - DCGAN (Deep Convolutional Generative Adversarial Network) - the artist feeds the AI with a series of images. These are drawn from the history of art and architecture and often pair unlikely bedfellows, among them - Fragonard, Super Studio, Boucher, Archi-Zoom, Bosch and selections from the artists’ own past work.
The algorithm both generates and discriminates. One part ‘looks’ at the constitutive pixels and their surrounds, seeking spatial resemblances and differences to create an image. The other part compares the output with the original dataset to determine whether the image is successful - either ‘fake’ or ‘real’. This is particularly startling and begins to resemble something unsettlingly like consciousness. The generative AI reconsiders its efforts, attempting to predict the next logical painting from prior efforts, hoping to trick the discriminator into classifying it as ‘real’. Hence the ‘Adversarial’ in the DCGAN - both elements of the algorithm question and consider their success adapting their code to ‘improve’ their outputs.
This process takes many hundreds of thousands of repeated passes. Early iterations resemble a cloud of dots and even highly resolved images are smeared, blurred and uncertain. A kind of mechanized, computer generated collage then, but this also sounds disturbingly like an artist - re-examining their work, shifting to seek new and better results. The machine remains entrapped within the premises presented to it, the raw material chosen by Leach. No android can dream of electric sheep without instruction - at least - not yet.
The resulting images are an electro-pixel fog of strange, uncertain visions. These are not precise and mechanistic, but miasma swirls of colour and indistinct form. They are, in fact, evocations of form, not pictures at all. The artist then looks at these electric dreams and imagines back into them. Dreaming back into a dream, he sets about making a painting from them. The artist provides the solidity and certainty they lack, refining, discerning, inventing and amplifying the absences and uncertainties between the smears to produce a final image. Given all this work, the exhibition title is more than a little ironic.
This process recalls a previous attempt at full artistic automation, the automatic drawing of Surrealism. Some Surrealist artists attempted to unselfconsciously produce marks, creating images supposedly unfettered by the rational mind to reveal unconscious manifestations. Here, the machine becomes the Surrealist, charged with the first stage of unconscious automatic speculation. The second stage of wilful imagination and image creation is, however, much closer to an older method – one described by Leonardo Da Vinci when he advised artists to create images by looking at “a wall spotted with stains, or a mixture of stones, .. to devise some scene” or to “discover a resemblance to various landscapes. …strange faces and costumes, …like the sound of bells in whose jangle you may find any name or word you choose to imagine.” [2]
Leach is a highly skilled painter and an adept artist acutely aware of the history of art, science, and the interconnection between these still allied fields. His past works are highly rendered, imaginative visions that surprisingly conjoin a variety of images collaged into a seamless whole that evoke the conditions of our age - so why invent such a convolute and complex process? Several reasons I suspect.
Science and art are linked by experiment. To have a hunch, then set up and test the idea in the physical realm, with the hopes of unexpected discovery is the very stuff of experimentation; common to both science and art. In this sense, these works are experiments in the emerging realm of AI. For me, they also recall the ludic writings of the OuLiPo group, who predetermined set structures to create unimaginable possibilities. Their leader, the polymath Raymond Queneau reportedly remarked the OuLiPo were “rats who devise the labyrinth from which they propose to escape”. (Examples include George Perec’s A Void a 290 page novel without the letter ‘e’ or Queneau’s own Exercises in Style a simple 3 paragraph story told in 100 stylistic variations.)[3] Similarly, the restraints imposed here, actually free the artist to invent in paint. The algorithm presents the artist with new and exciting problems for which a solution must be found - how to materially realize these visual conundrums in paint?
These works regard the fundamental changes the digital age has bequeathed upon us. Within art history, this is not unusual - artists often directly address and engage with the possibilities presented by new technologies. As machines began their rise in the 19th Century, Modern artists - from Monet’s railway stations to Duchamp’s bicycle wheel - speculated on these new technologies. The machine age has passed, superseded by the computer age, but in the same manner, progressive artists address this latest iteration of the new.
What exactly AIs will be capable of, is still an emerging story. Will they ultimately become Fully Automatic? It has been 23 years since the IBM computer Big Blue defeated then reigning world chess champion and grand master Boris Kasparov. Since then champion humans have been soundly thumped at even more complex games. Watson wins Jeopardy (2011), AlphaGo wins Go (2016). AIs are currently better at diagnosing skin cancers than human doctors,[4] and make much faster (and infinitely more sober) article clerks when finding legal precedent.[5] It goes on. Now the machines march forth to usurp the intellectual professions and cognitive labour of the 21st century in the same way they overtook physical labour in the 19th and 20th centuries. Some revel in these utopic technological dreams, imagining AIs will ultimately supplant human capability. But, as these painting reveal, this is not a simple binary. While no human can now beat the world’s best chess computers, these same computers cannot beat a human when paired with another chess computer. In the same way that humans partnered with machines could produce the previously unimaginable automobile, humans, when paired with AIs, will produce the next intellectual possibilities.
Apart from being stunningly strange, beautiful, evocations of where we are now, these paintings explore the dream states of digital possibility, the dreams of the machines and, in themselves, become a type of dream machine for those that view them.
Stephen Haley
[1] Dick, Phillip. K. Do Androids Dream of Electric Sheep?, Doubleday, USA, 1968. Ridley Scott later adapted this novel as the film Bladerunner in 1982 .
[2] Leonard Di Vinci, The Notebooks of Leonardo di Vinci, Entry 508 “Developing and Arousing the Mind to Various Inventions”, Trans Jean Paul Richter, London 1888, reissued Project Guttenberg, 2004
[3] Perec, George. A Void, (La Disparition) Gaillimard, Paris, 1969
Queneau, Raymond. Exercises in Style, (Exercices de style), Gaillimard, Paris, 1947.
[4] https://www.esmo.org/newsroom/press-office/artificial-intelligence-skin-cancer-diagnosis from Annals of Oncology, May 2018 last accessed 19 July 2020
[5] If you are in the law, you might be concerned to read: Re, Richard.M. & Solow-Niderman, Alicia: “Developing Artificially Intelligent Justice, Stanford Technical Law Review, 22. 2019. https://law.stanford.edu/wp-content/uploads/2019/08/Re-Solow-Niederman_20190808.pdf