From Ink to Algorithm: Linda Dounia’s Faux Letters

Faux Letters, Linda Dounia's recent print release with Avant Arte, combines hand-drawn letters and generative AI to dissolve the distinction between language and form. As letters coalesce into lively abstractions, Dounia demands we read between the lines to find new meaning within machine-generated writing.

Faux Letters 14, 2024

With a background in design, Dounia has always been fascinated with the aesthetic potential of language, the way the rhythmic fluidity of a letter or the intricate structures of a sign can evoke a range of emotions, and how, without replying on a particular lexis, they can convey a message to the viewer. Henri Michaux's Narration (1927), an ink drawing comprising eighteen rows of chaotic markings, is considered one of the earliest examples of asemic writing (writing that resembles language but is devoid of semantic meaning). This eventually evolved into a distinctive genre of modern art, perhaps most famously exemplified by the gestural, incoherent scribbles of Cy Twombly or the algorithmically plotted pseudo-writings of Vera Molnar. While visually distinct from Twombly or Molnar's cursive script, Faux Letters stems from the same legacy, and builds upon it by defamiliarizing language through a machine-learning model.

Henri Michaux, Narration, 1927, Ink on paper, 36 x 26 cm. Image Courtesy of ArtNet

Looking to the past as much as the future, Dounia began Faux Letters by researching over twenty-five written languages from five regions of the world, examining systems both ancient and new. Enthralled, for instance, by the Mende Kikakui script from Sierra Leone, she started emulating characters from each of these languages using pens, paint and scraps of paper. Whilst Roman letters came naturally to her, those recently discovered were much harder to recreate. As characters' capacity to convey meaning was negated in the languages she did not understand, Dounia focused closely on their formal configurations, discovering, in turn, what it means to read without language and write without meaning.

Linda Dounia in her studio, 2024

Moving to the machine, Dounia fed her drawings into a text-to-image diffusion model, a distinct type of AI that mirrors an artists' asemic approach to writing. Instead of weaving letters into words or words into sentences, diffusion models perceive language purely as form. The emergent compositions incorporated Dounia's letters but only according to their aesthetic attributes, fragmenting and layering them into a rhythmic synthesis of shape and texture.

Compiled into a dataset, her drawings were also used to train a GAN, an AI model comprising two networks: a generator that creates images based on a dataset, and a discriminator which decides whether they are real or fake. By learning patterns and structures from Dounia's drawings, the GAN created uncanny approximations of letters, allowing her to further interrogate the enigmatic space between text and image. Collaged together and arranged on grids, the final compositions exhibit the cold, precise nature of computers-led aesthetics without losing the delicate, gestural qualities of her hand-drawn elements.

Faux Letters 16, 2024

Like much of Dounia's work, Faux letters oscillates between celebration and critique. In her experiments, she noticed that diffusion models frequently output letters that resemble the Roman alphabet. In an attempt to redress this bias, she used Saki Mafundikwa's book Afrikan Alphabet to focus her drawings and dataset on non-western scripts. Yet, amidst the incoherent beauty of converging letterforms, the symmetrical and geometric sensibilities of the Roman alphabet remain apparent in the final outputs.

Viewed through this lens, Faux Letters is fundamentally a political project about the deep-set biases existing within AI today. As big tech companies advance toward generalized models – AI systems capable of surpassing human cognition across disciplines – Dounia urges us to recognize the inequalities they will perpetuate if pursued too quickly. As an artist from Senegal, she understands the necessity of her engagement with the medium, creating works like Faux Letters to shine a light on the marginalization of communities, whose histories and cultures are at risk of erasure in the realm of AI. Ultimately, through such projects, Dounia aims to provide insight into the inner workings of AI systems so that we may begin to redirect our trajectory towards more inclusive techno-futures.

Faux Letters AP 3, 2024

Elusive, yet uncanny in their familiarity, Dounia's letterforms draw us nearer like Twombly or Molar's indecipherable script, revealing the formal beauty within centuries of written communication. But far from meaningless, Faux Letters examines what machine-led asemic writing can tell us about AI. By traversing the analog-digital divide, playfully unveiling its coded prejudices, Dounia sets a precedent for critical participation in AI and invites us to envision alternative narratives for its future.