Robbie Barrat’s AI mechanisms:
CycleGAN
Pix2PixHD network
GitHub: https://github.com/robbiebarrat
On Barrat and Cosmotechnics.
Advances in technics and aesthetics in contemporary Digital Art are frequently discussed in relation to Cybernetics, Art History, and Politics. In this comprehension, I will be looking into Robbie Barrat’s work, a computational artist whose technicity consists mainly in the training of neural networks to create visual representations (image-painting), what is commonly regarded as Generative Art, or GAN Art (Generative Adversarial Network). The relevance of his work lies in the technical specificity he has developed for his computer to become a creative subject. In part, his works can be as a useful visual reference that points to where technology and applied data science in the Arts might be heading towards; but also, how the mentioned can serve as tools for research in aesthetics and language, by providing insights that otherwise would be difficult to achieve solely by using the logic specific to the human brain. I regard the computer as a ‘creative subject’ in Barrat’s process because the computer is not merely a tool, as in any GAN conclusion, the artist must allow the machine for a certain level of creative determination, it is within this unexpectancy that the magic of art occurs. The visual compositions then result from his collaboration with a machine, lighting up a way to look into the processes individuation of each, the artist, and its technical object as they involve in technical reciprocity in order to create a work of art. Furthermore, I attempt to make sense of this reciprocity in the domain of cosmotechnics, as developed by Yuk Hui, to run a semiotic analysis of digital contemporary art and its condition in the markets by making an example out of Barrat’s disruptions. I hereby break with the homogenous technologism that narrows our general understanding of the human-machine relation in respect to art.
I came across Barrat’s work just in 2018 on Twitter, he was training a neural network that created ‘its own’ classical paintings, namely Landscapes and Nude Portraits. I was moved by the way his work was contrasting–or bridging–which such clarity some of the notions concerning techno-aesthetics, which at that time, were heavily influenced by the training of neural networks–the AI aesthetics ongoing configuration–, against the aesthetic logic models established in classical painting and in West’s Fine Arts in general. What called my attention initially was encountering a GAN that was not built upon a sea of innumerable metadata, instead, it was trained on a very specific set of imagery (data) that belonged in very defined aesthetic categories, for an example a classical oil, or a piece of haute-couture. Artifacts that are designed with a high degree of technical craftship would then train the network so it can imagine possibilities on its own. The images that result from running the GAN are eerie but feel familiar nonetheless, because they were brought to existence using elements of
composition–such as certain colors and dimensions– that we’ve longtime understood, and usually conceive as natural forms in the realm of Fine Arts. The base of such weirdness characteristic to a GAN piece is the degree to which the produced image blurs, moves across, or neutralizes the lines of aesthetic delimitation between the abstract, the figurative, and the concrete, essential in the eyes of the human to understand an artistic composition. In other words, it can provide refreshing insights outside of the methodological frameworks by which we understand form and representation in the Fine Arts. A heavily human-influenced computer, an AI, as much as it is working in a closed system of references, has total disregard of broader categorical developments important for the history and ontology of the visual.
In his relatively short career, Barrat has gained considerable recognition, usually attributed to his young age, his artworks are both challenging but visually/conceptually relevant for the people. When he got out of Highschool he had an internship at NVIDIA, where he had access to great computational power, the things that he learned and tested there, I suppose, would be crucial for the further development of his own creative language. He got particular attention from the broader public with his project ‘Neural Network Balenciaga’, which consisted in feeding the algorithm with images of Balenciaga runways to train the AI to create its own interpretation of a Balenciaga runway. This quickly captured the attention of the luxury retail giant SSENSE’s board of content creators, who later promoted the work by publishing an interview with him titled ‘Do Androids Dream of Balenciaga SS29?’. He was 18 when his artwork was the first that was uploaded to the NFT platform SuperRare ‘Nude Portrait #1’ on April 2018. Later that year Christie’s held their first ‘Tech Summit’, its theme was the blockchain, and they distributed onee piece of Barrat’s collection of Nudes in 200 fragments as NFTs. The next edition of the Christie’s ‘Tech Summit’ in 2019 theme was Generative Art. There is a lot to unpack regarding the Market interest in Barrat’s work and in GAN in general. For instance, we could locate it within the artificial hype that Christie’s and SuperRare were pushing to promote the newly created NFT economy as a financially stable system. In this sense, Christie’s intervention in the blockchain economy was a statement in an effort to convince the public and collectors that NFT art was more than a pyramid scheme of speculative collectibles and that it was, indeed, an actual Art movement, and therefore a matter of technical and aesthetic seriousness.
Attracting the attention of these industry giants has rendered much of the writing done about his work for journalistic and marketing purposes, telling us more about the Art industry’s relation and future plans with Digital Art than of the implications of his technics in Art-research creation. While Christie’s and other giants in the crypto marketing complex make a case example of Barrat’s work in their effort to accommodate Digital Art into the logistics of the Art Market by pushing the narrative that blockchain technology is the pathway into building a decentralized infrastructure of wealth distribution to benefit creators, other texts, such as SSENSE’s interview, try to explore the potential of GAN use in other sectors of creative industries like in Fashion’s processes of design and creative direction. At the same time, they used his work as a proxy to promote Balenciaga, which is one of SSENSE’s bestselling brands. I do not doubt SSENSE’s good intentions in promoting the work of a young talented artist exploring the limits of fashion design with technology, it should be addressed though, that the interest of such a big retailing corporation in computer-generated art can only underlie in the utility that such Art can have to maximize profits.
Furthermore, none of those conversations are able to detach from the idealization of AI technology as a tool for the production of symbolic and capital value. We can observe, in a broader discussion, that much of the questions asked in regard to the relation of technology-art, are placed upon the technical objects that such relation produces; the apparatus, if it is the piece or a mechanism, is judged by its capacity to produce, reproduce, reveal, reorganize a certain set of values. With machine learning this conversation gets interesting, as it supposes that such processes can be automated and optimized with enough computational power beyond the limitations of the human condition.
It is this possibility of an automated reproduction of value that is at stake in the imaginary of Christie’s, SuperRare, and other platforms trying to knot a system in which blockchain technology and GAN ‘advances’ the conditions for production and commerce of works of art. It is value the central axis of their approach. Moreover, what is an NFT if not a resignification of the symbolic and capital value of a digital (object) asset? On the other hand, in SSENSE’s interview, Arabelle Sicardi approaches Barrat’s work from the perspective of ‘Creative Direction’, discussing the possibilities of AI applied to Fashion design, they draw the idea of AI as a tool for creating new things. First, we need to understand Creative Direction as a practice component of the larger process of Merchandising. Creative direction is the axis that conducts the processes of thinking, designing, producing, marketing, and ultimately selling a product. Creative direction is defined by productivity.
I consider these approaches a misunderstanding of the actual reconfiguration of value implicit in Barrat’s work, which lies in the process of individuation of its algorithm, and not on the technical application of such as a ‘machine that produces art’. As Barrat makes a selection of artworks that already underwent a process of curation and archive, he becomes a mediator of aesthetic value that will be suitable for the machine he has designed. Barrat’s preference to explore these notions through the use of Fine Arts is not independent of the discipline of selecting, organizing, and maintaining artifacts in an effort to consolidate a collective memory, which was the prime interest of the museums curating and preserving these pieces in the first place, understanding the role that Museography had–or intended to have– of mapping the history of the human intellect and creativity from the West’s perspective. The computer (machine) when is fed with such curated imagery, has the task to make a reinterpretation of such aesthetic values that had been previously carefully selected but it does so by neutralizing the symbolic and working solely on the aesthetic composition. What we have is then an abstract compression of fragments of a mediated collective memory, a technical reimagination of the Nude and the Landscape as categorical visual form.
In cosmotechnics, Yuk Hui would regard Barrat’s AI within the realm of techno-organicism. The process of individuation of such machine is its becoming organic, yet it can never become an organism of its own. Hui holds that the technical objects first became organic with Cybernetics, since it is the relation of information feedback that allows the machine out of the mechanical condition and sets it to a recursive logic. The recursive expansion of the digital (through operations of feedback), overcomes the dualist logic of antithesis between the organism and machine that prevailed in general ontology, he states ‘the becoming organic of machines is in the process of producing a new totality through exponentially stronger connectivity and algorithms.’ Having algorithm technology in mind, he opens the question of what is the role of art in this epoch of organic machines? This question is of paramount importance for the art market circuit of production-commerce, as it is a central inquiry for philosophers and critics working across semiotics, technics, and aesthetics, an ultimately, it poses a challange for the artists that see no intersection of art and technology, but regard it as the same technē (the Greek work which mean both art and technic). I consider Barrat on the latter, as he is clear that his GAN is not a creative individual of its own, but a medium whose mode of existence is transformative interoperation between computational power (technic) and aesthetic values (wich are indeed technic as well as they are materialized through pixel organization, the prime matter of such art).
With the invention of the camera, pictorial representation no longer depended on the motor skills of the painter, ‘liberated’ from the medium specificity of the brush and the paint’s plasticity, picture was now created through the impression of light by a technical process that was not plastic but chemical. As the camera captures everything within the realm of the visible, the artist’s task ceases to be the impression of what is visible in the light but to explore the sensible beyond the visual realm of light. Franco Berardi defines sensibility as
the faculty that enables human beings to understand what cannot be expressed in forms that have a finite syntax (such as the verbal form). He conceives sensibility and aesthesia as the two main abilities that constitute aesthetic perception, for him, it holds a direct symbiotic connection with technological transformation. The camera elevated the plastic artist’s
challenge to sophisticate the level of aesthetic perception, because the challenge was then to materialize what is invisible to light, opening the path for abstraction and conceptualization, which intended to represent/materialize ideas and feelings rather than moments or stories. Just as the camera opened the path for different possible sensitivities in both, film and plastic arts, the introduction of algorithmic thinking also lights up new ways to materialize (visually) what cannot be thought conceived solely through the mechanism of light or with syntax.
Berardi considers rhythm as the common substance of signs, within his theorical frame work, semio-capitalism, he conceives the creation and circulation of value to be in a constant process of acceleration, at the rhythm of information machine interoperating making an abstraction of space and time. The rhythm of info-acceleration was first depicted in the Arts with Futurism, which, he notes, defined the relation between art and the social mind through the introduction of the cult of energy. The industrial mechanical machines allowed the abstraction of use value and productive work (as Marx described) in the industrial economies, but as we go through the first half of the XX century, with the introduction of Cybernetic science, mechanics were transforming into feedback machines. It was around the end of the 60’s that the official shift from the mechanic age into the computational age was announced (for the Arts) at MoMa’s exhibition titled ‘The Machine as Seen at the End of the Mechanical Age’ (1968), which happened at the same year of the emblematic exhibition ‘Cybernetic Serendipity’ in ICA, London. Both exhibitions were somehow intended to promote a new vision of world, the milieux of human connaturality with the new computer machines. This was backed by technology corporations that were becoming increasingly powerful and global through the establishments of the first global networks. Some years later, already at a stage on late-Modern capitalism, the techno-linguistic devices that derived from this shift in power, had created a new hegemonic dimension of abstraction, digital abstraction, which adds up as a layer on top of the capital abstraction (Marx). In this level of abstraction, information takes the place of things, and transformation and production no longer happen in the material condition of bodies, but in the network of accelerated in-flow of information allowed by computational infrastructures. As we can observe in what happened with the camera, and with the industrial mechanics of reproduction, then with cybernetics, then with digitalization, sign and meaning are constantly achieving new levels of abstraction as technological transformation occurs.
The following image shows the results of searching for Barrat’s image in Google’s ‘Lens’ for image recognition. It discloses a set of images with similar features in composition, none of them done through GAN. This gives us an extraordinary insight, because it reveals that the abstract-figurative neutralization done by a machine, was done in a similar way by a human mind before, yet none of these images were fed into the GAN that Barrat trained.
As we move back and forward between value (re)configuration and algorithmic creation, we find ourselves in the field of the technical specificity of digital matter, it is important here to lay out the constitutes of matter for as GAN art goes and how it gets its form. In his book ‘On Mode of Existence of Technical Object’, a fundamental piece of work for Hui’s thought and mine, Gilbert Simondon writes:
‘Matter is matter because it contains a positive property that allows it to be modeled. To be modeled is not to undergo arbitrary displacements but to organize matter’s plasticity according to definite forces that stabilize the deformation. The technical operation is a mediation between an inter-elementary ensemble and an intra-elementary ensemble. The pure form already contains actions, and the raw material is the capacity of becoming; the actions contained in the form encounter the becoming of the matter and modulate it. In order for the matter to be able to be modulated in its becoming, it must have a deformable reality.’
Now, I want to apply this to the condition of AI and GAN generated art. It is important because such art’s plasticity consists of one of the prime matter/substances of our time: digital data. In the realm of the visual, it is almost always materialized by pixels, just as everything we encounter at the light of the screens that surrounds us. Digital matter is a compound of several other technical processes occurring simultaneously. In the case of algorithmic configuration and creation, the ‘deformable reality’ (of a matter) is given when such matter is digitalized, a work of art can be modified and turned into a new composition when its image is digitalized. But it is not only a matter of bringing things into the digital realm, it is mostly about the (coding) mechanisms that are collectively designed and optimized that allow the evolution of technology in respect of contemporary digital art.
On a final note, I want to focus more specifically on the last crucial component of the process of individuation of a GAN algorithm, which is its Open-source nature: the track of possibility for such technology to evolve. What cosmotechnics provides is a theoretical framework by which we can conceive this art as an effect of individuation that is not individual (as authorship), but is a result of psychic and collective effects of the technological, the natural, and the social. Both Hui, and Simondon, regard the struggle for reappropriation of technology. And this is particularly important in the nature in which GAN art is created. GAN art is able to exist and expand through a collective network of sharing technicity and advancing on it. Barrat shares his codes as a fundamental component of his work, as his work is based on prior codes that were shared online and allowed him to advance on them. The piece becomes meaningful as it exposes its processes of individuation. It sums to the struggle of accessibility, contributing to the annihilation of the paradigm of the computational technical object as a black box, the technical and theoretical mechanisms are revealed and put up for use and optimization. Therefore, the transparency of processes is not only a gesture, it is an actual component of the work.
Yuk Hui does not address Rabbat’s work directly, but in his chapter ‘The Status of Machine Intelligence Today’, while talking about Art and Automation, he brings up the piece ‘Edmond de Belamy’ (2018), an AI-generated portrait that ‘broke’ the internet when it sold as the most expensive AI-generated piece to date *at Christie’s*, it was authored by a marketing collective. The piece was created using a code written and uploaded (as Open Source) by Robbie Barrat, naturally causing controversy in the media regarding authorship and authenticity. In this sense, Robbie’s work embodies the succession of Open Source in a fundamental contradiction, as Open Source breaks with conditions of impossibility by destroying the functionality of ownership and allowing accessibility and potency, it easily becomes an artifact for speculative capitalism, reinforcing the abstraction of value and the skepticism of machine-human relations. Nevertheless, what concerns us is something far beyond the matter of ‘authorship’, and we disregard the conditions of the Art Market system, instead, we are following a line of work that concerns the actual relation of technical processes applied to art. If Walter Benjamin opened the question of how mechanical technology transforms art, Hui reverses it by asking, how art transforms technology? My own is: how can we use technology applied to Art creation as a method to study the history of human aesthetic sensitivity?