Reading time: 17 minutes
Translation by AB – September 13, 2020
The crows of the Japanese city of Sendai have the “hacking spirit”1:
Walnut branches planted along the roads hung tasty nuts, but they were, in their green shells, inaccessible to our beaks. It was then that my congeners learned the traffic regulations. At the red light, the bird places its nut in front of the car, which crushes it at the green light, and whose fragments are picked at the next red light. Clever, isn’t it? And even downright clever.
If, for example, the “traffic regulations” represent our very human tendencies to laziness and addiction, if the “nuts” are our time and our data, then we recognize the “techno-crows” who share this hacking spirit and understand how to transform and use our “traffic” to their benefit. This way of doing things is neither new nor questionable, but AITs (Artificial Intelligence Techniques) now provide, in a fully digitized environment, unbelievable “hacking” possibilities.
These AITs have recently started to combine with art. A doubt then seizes us: is “AI-art”, which main idea is to produce “works of art” by “intelligent” artefacts, a more or less deliberate hacking enterprise aimed at directly produce attention and notoriety by dispensing with subversion, spirituality or aesthetics? In short, is it still art? Let us recognize that this question has shaped the history of (modern) art, at least since the gesture of Marcel Duchamp or that of Brâncusi, a history marked by ruptures and manifestos. The fold is taken: we will answer that AI-art is still art even if, for the moment, at its very beginnings, it seems to be born without artists.
To observe this curious infancy, and because it concerns above all techniques and objects that must succeed, we propose a detour through the origin of our techno-artefacts, the beating heart of Silicon Valley and digital marketing: Stanford University
Design Thinking is a collaborative design method invented in the 1980s at Stanford2. The central point of this method is the integration of the future user, the “stakeholder”, in the design process. The first of the five steps proposed by the Stanford d.school is called “Empathize“. This step involves interviewing stakeholders in order to empathize with them. Interviewers seek to establish what users DO, THINK, FEEL and SAY. The objective is to obtain a sentence such as: the stakeholder needs “something” because of “something else”.
Most of our digital tools and services are designed in the presence of the stakeholder, with whom the design team tries to “resonate”. The future user is, so to speak, “dismantled” (deconstructed) and “tampered with” as a computer was by the hackers of yesteryear. Design thinking is one of the fundamental tools of an economy of experience considering that the world functions globally as a machine that must be understood and then patched / hacked. In this “machine”, the user is subject to an authentic integration: the delivered experience conforms to his desires and vice versa. Thus, the digital world and AITs now expose us, thanks to Design Thinking and technical power, to countless integrations.
Thousands of startups are designing tools and services using artificial intelligence techniques: H20.ai predicts customer reaction, Metamind makes images speak, Snips hides technology behind a voice, Swiftkey writes for us, Dataminr detects potential “breaking news” one step ahead, Pony.ai offers autonomous driving algorithms, Eyeem determines the “value” of an image, Zendrive uses the sensors of our smartphone to improve our driving (and lower our premium insurance), etc.
Since artificial intelligence also offers itself to art, would it not be immediately contaminated by the virus of design, experience and hacking: doing “something” (MAKE) because of “something else” (HACK)?
In 1920, the French philosopher Alain wrote3:
It remains to be said in what way the artist differs from the craftsman. Whenever the idea precedes and regulates the execution, it is industry […] A beautiful verse is not initially planned, and then made; but he shows himself beautiful to the poet; and the beautiful statue shows itself beautiful to the sculptor as he makes it; and the portrait is born under the brush.
Alain observed this: the craft object exists before it has been manufactured, while the work of art is not defined until after its complete realization. But the techno-industrial invasion changed everything. The artist’s work has gradually moved from downstream to upstream, from the revelation by doing to the preliminary conception “because of something else”.
As early as 1924, André Breton called for a wakeup that is still valid today4:
We are still living under the reign of logic, but the logical processes of our time apply only to the solution of problems of secondary interest. The absolute rationalism which remains in fashion allows for the consideration of only those facts narrowly relevant to our experience. Logical conclusions, on the other hand, escape us. Needless to say, boundaries have been assigned even to experience. It revolves in a cage from which release is becoming increasingly difficult. It too depends upon immediate utility and is guarded by common sense. In the guise of civilization, under the pretext of progress, we have succeeded in dismissing from our minds anything that, rightly or wrongly, could be regarded as superstition or myth; and we have proscribed every way of seeking the truth which does not conform to convention.
What we can observe today of artists trying their hand at artificial intelligence falls under this trap: the use they make of AI is, so to speak, “compliant by design”. And today, unfortunately, we don’t have any surrealist proposition: there is no Freud today to show us new and mysterious territories escaping industry: the unconscious, the mechanics of dreams … Instead, we have neuroscientists! We may have to seek the dream outside of the dismantled, deconstructed human, the stakeholder. Why not in these machines that we design and that, perhaps also for this reason, we seek to provide an intelligence, a consciousness and therefore a mysterious “unconscious”?
But this industrial and systematic design activity, which invades our ways of thinking, can and must itself be questioned artistically. Can we make design itself an “objet d’art”? Between September 2017 and early 2018, the French Stereolux’s Arts & Technologies laboratory and the Design Friction studio conducted a project called “utop/dystop(IA)“5:
The approach used throughout this project is that of “design fiction”. It is an approach to design that imagines and designs fictional objects serving as a support for reflection and discussion.
It is therefore a matter of thinking and discussing, not yet of creating. But already starts a movement of liberation from the final cause (in the real world): MAKE. Thus, conceived by design fiction, the so-called “Spice up Life-Assistant!” [ in French “Pimentez Assistant-Life ! ” ] provokes our personal AI-based assistants (Siri, Alexa …) and tries to hack them to get us out of their bubble:
Faced with this technology [chatbots, personal assistants …] which smoothens our lives, a group of activist hackers, the REVEYEZVOUS [ WAKEUP ], mobilizes and invents a chip, the Spice, which interferes with the functioning of artificial intelligence. One watchword: “cultivate your difference”! The principle of the Spice is simple: temporarily reprogram the artificial intelligence of your Life-Assistant and let yourself be surprised.
Another example is “The Empathizer“, this fiction consisting in helping us to understand the emotions of others and thus to improve social relations:
A connected object captures the emotional state of your interlocutor and an artificial intelligence translates it to you to make it understandable and perceptible. The idea is to no longer prejudge the other person’s reactions, but to experience them through stimuli that help to feel the emotional state of others […] Very quickly, the product is adopted and promoted by public authorities who see it as a miracle solution to restore social cohesion.
These fictional objects and services are aimed at a use that questions our digital shortcomings: confinement with the intelligent personal assistant, which would therefore be a question of hacking (spicing up) in turn, social atrophy which could be resolved by a “prosthesis” (on prostheses: The mirrors of the “I”), fake news which would be automatically revealed and corrected in real time by the “Verity” algorithm, replacement at work by robots which would be slowed down by a coach, etc. Or even6:
“The Creative Mixer”: what if you described your personality to obtain a “mix” of your visual identity? “The Empathy Writer” [ empathy again! ]: what if you could find the exact words to express your feelings? “The Belief Checkout”: What if the supermarkets of the future took your values and ethics into account when they automatically order products for you? [ Homo Amazonus ]
Design fiction thus operates as a serious mode of questioning the future through design. It just about fails to go beyond Design Thinking, to “ironize” it, to propose fictional services that are absurd, funny, surreal, or simply beautiful. And why not, therefore, fictions in which the stakeholders are machines?
In any case, from design thinking to design fiction, design is the conceptual matrix of today’s “AI-art” and its first institutional avatar: “GAN-art”.
Unction: birth of GAN-art
Backward in 1926, abstract art was officially recognized with this work by Brâncusi7:
“Bird in Space” is a one meter and thirty-five centimeter polished bronze sculpture. Its tapered and clean shape evokes that of a feather. When the object arrives at the port of New York, customs officials deny it entry into the territory as a work of art. [the work was released under the mention “kitchen utensils and hospital equipment”].
Since antiquity, art has imitated nature thanks to the precious know-how of artists. American legislation was based on this principle to put Brâncusi’s work on trial. […] On that day, the judge had the power to redefine a work of art. After many debates, the verdict came: from then on, the artist can transgress the rules if he expresses his personality through his creation.
This reminds us that art is not free to define itself but always comes under an external and official unction. Without this convention, the work of art would have no aesthetic, no spirituality nor, of course, market value. We recently recalled this episode (Adam Curtis and the Strange World): Christie’s sale of the first “picture painted by an artificial intelligence” (entitled “Edmond de Belamy“), the work of the cunning French collective “Obvious”.
This time the unction did not come from an American judge but from Christie’s: “Edmond de Belamy” was awarded for the modest sum of $432,500! To the question of the journalist Rémy Demichelis (before the sale) “It remains to be seen whether a sale at Christie’s is a criterion for recognition as a work of art?”8 we can answer unambiguously: yes! But we have to admit that art has been hacked by an excellent design fiction that was made concrete and that is ultimately the real “oeuvre d’art”: a painting machine (a kind of “nut-cracker”…).
Of course, this event sparked much debate: what exactly did the computer do? What did Obvious do? It is now necessary to speak a little technical and of this new artistic current which has become monetizable: the “GAN-art”.
“GAN” is the acronym for “Generative Adversarial Network”, which designates a particular class of machine learning algorithms invented in 2014 by Ian Goodfellow, a researcher from Stanford. These algorithms make it possible to produce fictitious images or videos as real as life (AI and its shadows). Ironically, the very principle of how they work is luring. A “generative” neural network produces images (or even videos, voices, etc.) by “drawing inspiration” from numerous examples (15,000 portraits from the 14th to 15th centuries for “Belamy”), trying to pass them off as “real” to a second “discriminating” neural network. These two networks compete with each other until, by successive adaptations (what is called “learning” or “machine learning”, i.e. mathematical convergence techniques), the first manages to pass its productions off as “real” to the second as often as possible.
In 2016, by injecting 300 real Rembrandt paintings into its GAN, Microsoft succeeded in producing a true-false painting by the Master, soberly entitled “The Next Rembrandt“9:
Other GANs still produce images of fictitious people (is this art? Why not ?)10:
The process is purely mathematical and obviously does not involve any kind of intelligence11. The painting “Edmond de Belamy” is moreover cleverly signed by the formula of Ian Goodfellow which underlies the process and which mathematically proves that the play between the two networks, generative and discriminant, results in a state of equilibrium: the final painting.
To claim, as Christie’s does, that Belamy’s portrait is the first “painting painted by an artificial intelligence” is obviously a joke. We might as well pretend that Nicéphore Niepce’s “Point de vue du Gras” is the first photograph taken by a camera! The GAN-artists have a kind of spirograph in their hands and look for the hypotrochoid that suits them by trying the different holes.
The author, or rather the designer, therefore continues to play a decisive role in these works. But is he still an artist?
One of the best-known representatives of GAN-art is Mario Klingemann, a German artist resident of “Google Arts and Culture“12. It explains at length in an interview with art historian and curator Emily L. Pratt13:
I try to figure out how things work together and hang together, and the rules that govern them to see how they can be hacked. In my ideal world, there is always a way to map everything out, or to take everything apart, and to look at all parts separately. I then can put them into a different order so that you can learn something about it.
As a GAN-artist Mario Klingemann thus creates works in the spirit of the “portrait of Belamy“. Take, for example, this recent work, entitled “Memories of Passerby I“, an installation which consists of a composition of multiple GANs. A chestnut console contains the brain-computer (“AI computer brain”) and two screens project calculated portraits in real time and without interruption:
This work was presented in February 2019 by Sotheby’s which priced it around £ 30,000 to £ 40,000. It ended up selling for £ 32,000, a real flop compared to Belamy’s portrayal … Something went wrong very quickly in GAN-art and this probably for two reasons. First, the use of AI here is literal, accomplished in the first degree: it is a matter of running Goodwin’s equation ad libitum to produce images. In a Mundus Numericus already saturated with images one cannot claim that this is a revolution. Secondly, in principle it is a question of (having) fabricated “realistic” variations around the same theme, even if it is a fictitious theme.
But here’s a little better, a more facetious designer.
Authors of authors
Artist Ross Goodwin (also aided by Google) fitted a car with various cameras, sensors, and a GAN that had swallowed up 200 classics of Anglo-Saxon literature. Goodwin drove this car-writer, this “design fiction”, for 2000 kilometers, between New York and New Orleans. She “observed” with her sensors during her journey and wrote a sort of “road novel” à la Jack Kerouac. It symbolically emerged a roll 127 feet long (the roll on which Kerouac had written “On the road” was 120 feet long). Revised and edited by Ross Goodwin, this work was published under the title “1 the road“14:
The car itself is the pen. But, then, who is the author? Goodwin calls himself a “writer of writers” and “not a poet.” Others might call him a programmer or “creative technologist.” Though Goodwin has surrendered creative license to the writing machine, he nevertheless created the machine and the rules by which it operates. Not insignificantly, he selected the sources of input, including “nearly 200 hand-picked books” that, in composite, form the linguistic matrix that informs the A.I.’s literary sensibilities, from word choice to sentence structure.
The “novel” begins like this:
Three seconds after midnight. Coca-Cola factory, Montgomery.
The GAN-artist and more generally the AI-artist is always present in his design-fiction. We would even say more: it is the machine, and not its digital products, which is the result of a genuine artistic work (so it is the machine that we should buy rather than the result…). Klingemann sensed this with the installation “Memory of Passerby I”. He just, in our opinion, failed to ironize on the emptiness of his images and thus to free himself.
The Oulipians define themselves, according to a phrase attributed to Raymond Queneau, as “rats which themselves build the labyrinth from which they intend to leave“15. They believe that the freedom to create paradoxically arises from formal constraint (think of Georges Perec’s famous “La disparition“, written without the letter “e”) because formalism determines all potential forms and thus frees the artist from having to accomplish this determination. All he has to do now is to realize, among these forms, an authentically free selection, a curation, which will mean the exit of the labyrinth16:
This is because we must not be mistaken: the potential is uncertain, but not hazardous. We know perfectly well what can happen, but we do not know if it will happen.
Whether it is “1 the road”, “Portrait de Belamy”, “Memory of Passerby I” or even the end of Schubert’s symphony n ° 8 (“unfinished”) composed by an IA of Huawei, these forms are well potentiated by Goodwin’s formula (and more broadly by the design fiction of the author-hacker) and “we know perfectly well what can happen“. Oulipian situation? Let us quote again Mario Klingemann17:
An early epiphany I had around 1985 was when I realized that a digital bitmap is theoretically capable of showing every possible image that there is as long as you are able to find its number or “address”, since that’s what a digital image ultimately is: a huge number.
In the same way that the library of Babel imagined by Jorge Luis Borges contained all the possible books and thus potentiated the discovery of any possible text, a machine can produce any possible image, any possible digital work. But surveying the library of Babel and choosing a book does not make the explorer a writer any more than the path of AI results makes the GAN-artist … an artist. Klingemann’s hesitation is palpable:
Emily S. Spratt: Curatorship is thus a huge part of being an AI artist.
Mario Klingemann: […] Ninety percent of what the models produce is not interesting, and that’s why it’s important to pick the good examples. But in the future, I might pass my curation to another model that knows what I like and then curates those models, so I would just meta-curate. […] This could get stale, however, if I have to train another model to train the initial model further, and so on and so forth.
The GAN-artist thus reveals a limit: whatever he tries, his machine produces works that are relevant to his taste. He shall never come out of the labyrinth because he does not create by himself and remains attached to his machine.
AI-art is still in its infancy but it has already been welcomed into art by the unction of Christie’s and Sotheby’s … It is supported by the major digital players (Google, Microsoft, Huawei, etc.) which probably satisfy two major interests. First, the artistic guarantee reinforces the idea that AI has a creative capacity, in the human sense of the term, and that, therefore, their technologies are “human-friendly”. Second, these artists they support are genuine good hackers who, through their artistic research, allow them to improve technologies that are then integrated into our smartphones, bots, cars, etc. In short, AI-art exists, but today it consists in trying to make us take, through sometimes edifying linguistic and scenographic staging, the productions of “intelligent” machines for authentic creations18:
A multimedia exhibition starts this Wednesday, June 12 at Oxford University. Paintings, sculptures, videos, performances… All these works have one thing in common: they have been created by a robot artist, the first humanoid robot capable of representing what he sees – in his own way.
This is what we have called “hacking of art by AI“. Let’s recall that by systematically integrating and dismantling the “stakeholders” (i.e. us) in our ways of producing objects and services, “AI-art” seems to have been born this way: as excellent design leading to good staging, good sales and good buzz.
When it comes to AI-art, therefore, it is not the artificial productions that should be looked at, which nonetheless remain authentic technical and mathematical exploits, but the artefacts that carry them out.
Gilbert Simondon, philosopher of techniques, knew how to put his gaze and his words on (and in) artifacts, when we consider only their use. This gaze was, moreover, almost tender: it gave objects their own existence, an independent internal logic, and above all a prerequisite for their use. But it also gave man a necessary place, which we must imperatively remember at a time when techno-powers are trying to pass off – this time through art – AI as genuine intelligence (we underline)19:
The machine which is endowed with a high technicality is an open machine and the set of open machines presupposes man as a permanent organizer, as a living interpreter of the machines in relation to one another.
On these bases (forgetting about usage, thinking of man as the organizing center), technique can be invited into the cultural field. First of all, it is necessary to rid the technique of its dominating, hacking essence in order to re-establish this creative opening that Simondon tells us exists within our artifacts. AI-art can then truly be born. Patience…
1. ↑ Aline Richard / Slate – May 30, 2019 – La signification des meuglements des vaches, et autres histoires animales surprenantes
2. ↑ Wikipédia – Design Thinking
3. ↑ Alain – 1920 – Système des Beaux-Arts
4. ↑ André Breton – 1924 – Surrealist Manifesto
5. ↑ Stereolux – January 25, 2018 – UTOP/DYSTOP(IA) : quand le Design Fiction interroge l’Intelligence Artificielle (broken link)
6. ↑ Dean Malmgren & Jure Martinec / Ideo – May 8, 2018 – What the AI Products of Tomorrow Might Look Like
7. ↑ Clémentine Picoulet / KAZoART – July 27, 2017 – Marcel Duchamp et le Ready-Made
8. ↑ Rémy Demichelis / Les Echos – August 23, 2018 – L’œuvre d’une intelligence artificielle française bientôt en vente chez Christie’s
9. ↑ The Next Rembrandt News Room – June 17, 2016 – The Next Rembrandt – Data’s new leading edge role in creativity (lien cassé)
10. ↑ Browse hundreds of fictional portraits on thispersondoesnotexist
11. ↑ A rather pedagogical mathematical reading of Sameera Ramasinghe on Medium.com – 14 février 2018 – Generative Adversarial Networks — A Theoretical Walk-Through
12. ↑ Many examples on Experiments with Google
13. ↑ Emily L. Spratt / WRDS – SPRING 2018 • VOL.24 • NO.3 – Creation, Curation and Classification: Mario Klingemann and Emily L. Spratt in conversation
14. ↑ Bomb Magazine – December 14, 2018 – A.I. Storytelling: On Ross Goodwin’s 1 the Road by Connor Goodwin
15. ↑ Wikipédia – Oulipo
16. ↑ Oulipo (Jean Lescure), Atlas de littérature potentielle, p. 25
17. ↑ Marion L’Hour / France Culture – June 12, 2019 – Le premier artiste robot expose à Oxford
18. ↑ Martin Dean / Sotheby’s – February 25, 2019 – Artist Mario Klingemann on Artificial Intelligence, Technology and our Future
19. ↑ Gilbert Simondon – 1958 – Du mode d’existence des objets techniques