Artificial Intelligence Art: A Quasi-Kuhnian Reframing of the Creative
In October of 2018, a portrait, titled “Edmond de Belamy,” signed minGmaxDEx[log(D(x))]+Ez[log(1-D(G(z)))], was sold at a New York auction for $432,500. The title, “bel ami” (a French translation of the phrase “good friend”), is a nod to Ian Goodfellow, the inventor of the now-prevalent generative adversarial networks (GANs) method in deep learning. The signature constitutes an excerpt from the GAN code, authored by Robbie Barrat, that was used by the art collective Obvious to generate the lucrative print. In preparation for the creation of the ultra-modern work, the Paris-based trio fed fifteen thousand man-made portraits, scraped from WikiArt’s storage of 14th to 19th-century paintings, into Barrat’s software as training data and then tasked the program with generating a painting-prototype to replicate what it had learned. Garnering both criticism and acclaim, the sale became a seminal milestone in the long-developed timeline of artificial intelligence art.
Among the phrases that lambasted the print’s success are “It is not the work of the artist’s imagination,” “But ‘art’ implies creativity, and creativity is just what artificial intelligence can’t do,” and “The machines are on their way, and they’re coming for your jobs.” Surfacing nearly unanimously among critics was the notion that the image is without a bonafide artist, as there was no creative who exhumed the image from her depths and viscerally transferred it onto canvas. Rather, the creation of the work was the occlusion of cleverness and opportunity, as Obvious found Barrat’s code online, made minor alterations, and employed a novel strategy for training its neurons to achieve their task. The trio did not only receive criticism from art traditionalists but from other AI artists as well, several of whom thought that Obvious’ borrowing of Barrat’s code (which had already been used to generate similar works, distributed freely online) made the creation of the print disingenuous and insincere.
Since Christie’s pioneering sale of the portrait, there have been many new and progressive auctions and installations, collectively demonstrating that the AI art movement is blazing a path in the industry, making salient efforts to secure a respected and influential place in the zeitgeist. Offering work in the media of fiction, music, and poetry, among others, these neural networks actively broaden their horizons with time, inhibited only by the willingness and ability of programmers to write concise, efficient code which aims at the pursuit of beauty and the experimental reflection of human experience.
But even in this exposition of the circuitous endeavor, a latent traditionalist intuition rears its head to pry: is it really experience that these programs are rendering, or are they merely summarizing the canon of man-made work that has preceded them? Probing deeper, questions with more relevance might be, does a work have to cognizantly render some aspect of a lived or creatively imagined experience to be artistically validated? Should art be defined by the statistical qualities of its output, or rather the nuances, the intention inherent in its creative pursuit? What is gleaned from an analysis of AI art is not an opinion about whether or not its product is authentic, but instead an observation: AI art extends an opportunity for the redefinition, the revamping of the artist archetype; a foundational rethinking of what it means to collect creative influence; and most resoundingly, a slick and meta-level rendering of the creative process itself. What is being toyed with here is the blueprint of art creation, and it follows accordingly that the hackles of some traditionalists are raised by these efforts. Consulting similarly profound rethinking of discipline, this transmutation of axioms is redolent of the twentieth century’s quantum mechanics conversion, or the sixteenth century’s Copernican cosmology proselytisation. It is a shift of paradigm, but not as bonafide as in Kuhn’s 1962 thesis, and may be more accurately called quasi-Kuhnian.
Quasi- because, although it is a reframing of art which is incommensurate with historical paradigms, there is no salient threat of AI art overthrowing the institution of traditional art, which stands monolithically adjacent to, and not in competition with, its neural network relative. But appealing to Kuhn’s Structure of Scientific Revolutions to impose theoretical infrastructure onto the AI art phenomenon proffers an interesting framework for analysis, and a toolkit to explain why this contemporary current experiences the trials and triumphs that it does.
By interpreting the AI art framework as a Kuhnian paradigm different than the traditional art paradigm, we can better understand it as a set of theories and ideas which define rational possibilities and relevant methods, which is in particular incommensurate with the coexisting paradigm of traditional art. Having converted a minority of art community members and still regularly facing criticism, the machine learning paradigm has found itself in a state of candidate acquisition, with a need to demonstrate its value to those gatekeepers who influence the conglomerate perspective of the coterie. The significance of Obvious’ work is that it provided evidence for the successful adoption of the new paradigm — a declaration of viability for the nascent movement. The stage that is thereby entered, according to Kuhn, is debate on whether the candidate paradigm is suited to address future problems, or whether it carries value that rivals that of the established paradigm significantly enough to warrant a laborious coup of perspective. But in this perceived rivalry, it must be remembered that rigorous competition of paradigms doesn’t sufficiently apply to the situation, as AI art is vying for a tenured position next to and incomparable with historical art’s seating. In explicating this phenomenon, Kuhn writes, “these examples point to the … most fundamental aspect of the incommensurability of competing paradigms. … the proponents of competing paradigms practice their trades in different worlds.” And this is fitting, as a survey of the AI art insurrection suggests that the work produced is inherently different in its proceeding, somehow disqualified from traditional criticism, and standing in duplicitous and unjustifiable competition with its human counterpart of creative rendering.
In the attempt at probing this technological candidacy, again, the eminent question is suggested: if it is not human experience in the traditional sense, what is it that AI art is rendering? Consider addressing this question by giving a synopsis of the discipline — in the practice of AI art, a digital mind, equipped with the simulated behavior of a human’s, is created and ordered to pursue an objective. That objective is to examine the artistic work of the past generations, to draw pragmatic (and necessarily numerical) conclusions about that work, and to utilize these conclusions in order to produce an entirely original piece, an amalgamation and statement of influence. It is seen here that AI art is not rendering reality, but rather it is rendering the artistic process itself, promoting the paradigm up one level of abstraction, and touting a model of the artists rather than her world. This model is then translated into substance by its product, whose hues, melodies, or stanzas symbolize the creative process which birthed them. Standing analogously beside those of man-made art, these final products are not to be appreciated solely for their perceivable qualities, but rather as encapsulative symbols of the process which produced them. In AI art, this process is foundationally different than that of man-made art — it is a process predicated on the reimagining of the artist herself. This essential discrepancy of objective is the reason why AI art seems so incommensurate with the prevailing notions of art and artist, and it is this discrepancy that is misinterpreted as an invalidation of authenticity.
In referencing Kuhn’s model, we can account for the blossoming proliferation of AI art and AI artists that has been observed in the past two years. As the well-known and time-honored artist archetype has proven unable to exploit the 20-teen’s artificial intelligence obsession, we’ve been pushed to exit the stage of operation which Kuhn refers to as “Normal Science,” and have entered a phase where members of the community grasp for experimental methods, revolutionary perspectives, and nouvelle strategies. As we are immersed in this stage, which Kuhn christens “Extraordinary Research,” we should expect AI to be applied ingeniously to every subdiscipline of art (and this project is already well-undertaken, with a large portion of media being touched by AI methods) in an exploratory escapade whose purpose is to put previously concreted axioms on the altar. This effort reeks of artistic hauntology, harkening back to the exploration of revolutionary advances in technology like oil-based paint, the digital camera, and en plein air painting, all of which have graduated from novelty to become staples in the modern industry. Analogously to these historical leaps in practice, as this machine learning escapade experiences successes it will proselytize artists and critics who are ready to embrace a new paradigm, granting credence and credible weight to the benign insurgence. With Kuhn’s diction, the recent years have hosted “the proliferation of competing articulations, the willingness to try anything, the expression of explicit discontent, the recourse to philosophy and to debate over fundamentals.”
Certain aspects of the output published by AI artists have shown that the brush-and-canvas artist, for now, is in no threat of obsolescing; as there are certain tasks of which AI has so far proved itself an incompetent pursuer. A prevalent, protruding one can be found in the observation that AIs tend to mangle and distort human faces when producing original work. While the visages produced do have abstract charm, consumers of realism are unlikely to be converted by the bone-chilling crooked smirks of these portraits, and may be imagined turning their backs until advances in accuracy are made. In the digital music scene, Amazon’s GAN-powered AWS DeepComposer has earned a reputation as a “quick gimmick” instrument that “sounds terrible,” and is consequently unlikely to woo composers and critics who value euphonics over technological novelty. In literature, Kyoto University posted 2018 research on arXiv recapitulating the successes of their AI poetry program whose output, in spite of being received as “hilariously terrible” by some critics, was indistinguishable by the researchers from human poetry. These aesthetic crimes should be taken with a grain of salt, likely the result of an inexperienced, unpolished, and recently-birthed movement, and also likely to be ironed out as time introduces more accurate GAN methods (or introduces a methodical insurrection to oust GAN). The more valuable didacticism spoken by these challenges is about the difficulty of distilling art into objective, numerical statistics, as these GAN softwares attempt to do — these missteps only underline the contention that a successful approach must be sophisticated, nuanced, and informed by epochs of trial-and-error.
As the passing years will see our technological cutlery more accurately emulate, or even surpass, the wetware of human cognition, it’s conceivable that AI art will come to produce timeless classics in several creative fields. Lingering in the shadows is even the provocative possibility that AI art will eventually see such profound success that it will overshadow our now-central artist archetype, causing an endangerment or, at worst, an extinction of traditional profession. While this is prima facie chilling, different social perspectives recommend different methods of reception for this possibility, with some as radical as Nick Land’s breed of posthumanism urging us to sideline human wellbeing and welcome, even accelerate that revolution, in hopes that it is the best possible trajectory for evolution of industry.
But given the current state of the industry, AI art is liable to see constipated success frequently and promising breakthroughs, like Obvious’ print, seldom. What we can expect, again following Kuhn’s model, is that the coming generations, having developed an artistic perspective with the institution of artificial intelligence in mind, will push more adamantly for advancement and devote more collective energy to the development of strategy. When theorizing the gradual acceptance of a paradigm, Kuhn cites Max Planck to claim that “a new … truth does not triumph by convincing its opponents and making them see the light, but rather because its opponents eventually die, and a new generation grows up that is familiar with it.” It may be decades before the staunch modern criticism is mollified, but as AI art fights for a place, with hope, we can look forward to a new discipline equipped with its own standards of criticism and estimation and its own institutions of education and apprenticeship. In 2020 and in the following years, it’s all in the hands of the programmers.
 Kuhn, The Structure of Scientific Revolutions, ed. 1970, p. 194
 ibid., p. 91
 Planck, Scientific Autobiography, ed. 1950, p. 33, 97
Published exclusively in Brown Tech Review. Image source.