Faceless Portraits Transcending Time, a new exhibition displaying AI-generated art currently at HG Contemporary, brings the art world’s current obsession – and debate – with AI center stage, posing essential questions concerning the validity of visual arts, and the role of aesthetics on social consciousness.

Text: Banyi Huang
Images: Courtesy of HG Contemporary

In Death’s End, the last book in Liu Cixin’s trilogy Remembrance of Earth’s Past, a character wakes up from deep hibernation in the year 2269 to find extraordinary pieces of postmodern contemporary art at the Shanghai Biennial[1]. She was stunned to discover that they were made by ‘Trisolarans’, an alien species who had been communicating with humans on Earth. By quickly processing massive amounts of data about Earth’s culture, they began to transmit recreations of artworks done in imitation of human models. The artworks became so popular that they soon superseded human production.

While Faceless Portraits Transcending Time, a new body of AI-generated art currently on show at HG Contemporary, is not created by a technologically-superior alien species ready to reduce the Earth to ashes, it nevertheless poses unavoidable questions concerning the state of the visual arts, aesthetics, and social consciousness. The exhibition features two series of works on canvas—the collaborative product between Dr. Ahmed Elgammal, a computer science professor at Rutgers University, and AICAN, an algorithm that the former developed based on 100,000 images culled from the history of Western art.

At first glance, these digital prints are remarkable, appearing to barely depart from conventional portraiture. Faceless Portrait #3, for example, can be described as a Caucasian, featureless man seen from the front, his windswept head attached to a body punctuated with blocky, feather-like textures, lit in the same dramatic, chiaroscuro way as paintings by Rembrandt and Caravaggio. Other works are replete with large swatches of symbolic colors such as red, gold, and black, invoking the human anatomy, Christian iconography, and visual demonstrations of political power. Ironically, these “faceless portraits” do feature faces, however abstract or contorted they may seem. The sublime expressiveness is undercut by a sense of the uncanny, a nightmarish hybrid between Goya, Francis Bacon, and modernist grids, seen from the pixelated lens of machines.

It comes as little surprise to learn that these images are generated from algorithms written with specific intent. In fact, much ink has been spilt over how GANs—Generative Adversarial Networks—function. Trained on the same data-sets, two neural networks are pitted against each other: while the generator is tasked with artificial outputs, the discriminator attempts to distinguish between the ‘fake’ outputs and the original data. Through this dynamic, iterative process, unpredictable elements start to emerge. However, Elgammal and his team, not content with the basis of imitation on which GANs operate, have adapted the algorithm and given it a new name AICAN, which stands for “Artificial Intelligence Creative Adversarial Network”. In his own words, the software is charged with generating images that are “stylistically different”, but not too drastic as to be unrecognizable. To most outsiders however, these parameters remain relatively obscure.

Circling back to the formal characteristics of AICAN’s two bodies of work, the disposition of the palette, composition, and motifs are undeniably shaped by a Western lens. The act of defining something as “stylistically separate” is to remain attached to the features it wants to depart from in the first place. To cheekily reference Derrida: “there is no outside-text.”[2] In other words, the outcome heavily depends on the range and selection of the dataset that is fed to the competing networks. Mario Klingemann, another AI-aided artist, incorporates pornographic images into his dataset, citing the maximum exposure of the body as the reason. Needless to say, an algorithm trained on pornography would produce vastly different results than, say, one trained on Song Dynasty Chinese landscape paintings. Furthermore, curation and intent not only figure in the raw input of data, but also determine what images are included in the final showcase.

When it comes to AI art, an oft-cited milestone is the exorbitant price that Edmond de Belamy, an AI-generated portrait, fetched for at Christie’s in 2018. Obvious, the Paris-based art collective that claimed authorship over the work, sparked controversy when another artist, Robbie Barrat, accused them of stealing both the algorithm and dataset. While the open source nature of the process raises concerns over issues of originality and plagiarism, the object-oriented application of these art productions reveals an inherent problem at play. Not immune to market forces, generative networks are utilized as a means to an end, specifically as overhyped mechanisms used to generate closed-off, standalone artworks for sale. Why couldn’t it be valued as a tool that could give us insights onto other things.

When a technology is relatively new, it’s easy for it to be commoditized, especially when it works in the artist’s favor to keep crucial details shielded from the public. Perhaps Heiddeger’s conception of techne and poeisis can be illuminating: whereas technology is too often understood in terms of instrumentality, a way of getting things done without considering consequences, an alternative way to think of it as poeisis, which means a way of bringing forth something that had been previously concealed[3]. If the exhibition indeed aims to “investigate” portraiture in an active sense, what could this process say about the Western-centric focus of art history, gender biases, different classes of aesthetics, to name just a few?

In a recent paper, AICAN’s team writes about how a sample of AI-generated art mixed in with genuine paintings from Art Basel was able to fool viewers, who genuinely preferred the former over the latter. The uncanniness finds its parallel in the Trisolaran’s artmaking, in the sense that both entities have obtained enough data to formulated a grasp over human aesthetics. On the flip side, we can observe increasingly worrying trends of image-making. StyleGan, an algorithm released by chip maker Nvidia, made headlines by dropping hyperreal human portraits that are almost indistinguishable from authentic photographs. We are currently living in an age of disturbing DeepFakes, where AI superimposes existing images or videos onto another, often resulting in someone’s face appearing in pornographic or politically-incriminating content without their consent. We are hence forced to contend with these phenomena, blinded by our limited understanding of history and technology. Seen in this light, are AICAN’s portraits truly faceless and timeless?

[1] Liu, Cixin. 2018. Death’s End.

[2] Spivak, Gayatri Chakravorty, and Jacques Derrida. 1998. Of Grammatology. Baltimore: The Johns Hopkins University Press

[3] Heidegger, Martin. 2013. The question concerning technology, and other essays.

Banyi Huang is born in Beijing. She is now a free-lance writer, curator, and translator based in New York. She also sometimes dabbles in 3D printing, interactive design, and other technologies that produce new forms, functions, and modes of being. She is currently pursuing an MA in art history and curatorial studies at Columbia University. She is interested in looking at the human body and the formation of identity/performativity within our mediatized and technological landscape.