PRINT April 2021


Automatic Writing

Page from Anne de Vries’s Deep Scroll (Onomatopee, 2020).

Deep Scroll, edited by Anne de Vries. Eindhoven, the Netherlands: Onomatopee, 2020. 380 pages.

DEEP SCROLL is a book for this precise mediated moment. It’s a chaotic journey in which the reader walks a tightrope between the real and the simulated, in which the old is repackaged as the new and we are driven to furious speculation over a multitude of speakers’ intents and ideologies. Artist Anne de Vries offers a mad pastiche of dense theory layered atop digital collages of his installation work from the past two decades. He sourced classic writings by Hannah Arendt, Alain Badiou, Manuel DeLanda, Maurice Merleau-Ponty, Carl Sagan. He worked “with” a range of AI generators, many driven by machine learning, such as GPT-2, QuillBot, and TensorFlow. Sprinkled through these excerpts are pink bubbles of text marked “AI GPT-2 Response,” suggesting the passage in question was created by GPT-2 (an unsupervised language model that implements a neural network to learn from prose input in order to predict what words are most likely to follow future inputs, and then to generate new writing).

Before you can dive into these experiments, you are stopped cold. The book seems designed to be unreadable. My first sensation was revulsion. Deep Scroll’s aesthetic is blistering hardstyle as a Photoshop theme, reminiscent of a decade-old, choppy, handcrafted online landscape. The fonts are those of Hallmark cards, motivational posters, metal T-shirts, erotica forums. Grungy, frazzled type seems to be having a meltdown; letterforms look like they’ve come back from war. This is the ugly, chthonic internet. In contrast, to conduct a deep scroll of an online feed in 2021 is to be lost in it, chaining from horror to pleasure. We are seduced by refined interface designs and an ethic of calm technology that occludes awareness of the frame. That’s to say, we entrain to the smooth scroll thanks to sophisticated planning that has become its own field. The feed, its targeted curation, its typography, shapes and contours, all calculated with military precision, calibrate a deep psychoemotional impact. (Captology, the study of computers as “persuasive technology,” is supported by a whole Stanford research lab on behavioral design “for good.”)

Page from Anne de Vries’s Deep Scroll (Onomatopee, 2020).

Graphic design has ideological power; it can induce belief or dismissal. In Deep Scroll, even J. G. Ballard’s words (“The advanced societies of the future . . . will be driven . . . by competing systems of psychopathology”), here set in Avenir in neon-green bubbles, seem trite. Across a long stretch of pages, we gaze at UV prints of balding men in bomber jackets, their faces not visible to us. They are “thinking” thought bubbles filled with instructive Freud and Kant passages, but one turns away from these words.

Close reading has its limits. Even the most committed exegete might struggle with a book that mimics algorithmically driven movement across feeds bled of context. Deep Scroll reflects what reality feels like for the digital laborer, parked in an ergonomic chair 24/7, absorbed in reactive consumption of a deranged technosphere. We read associatively through sections designed as “scrolling pathways” around edgy keywords: BLACK ICE, SNEAKERNET, the Neolithic, HYDRA 5, Sonic Warfare: 1929, SUBMISSION. We ricochet between 1960s French phenomenology and hauntology-era Mark Fisher. Spam about MoMA jostles with special offers from Sweetgreen.

Bits of theories work until they don’t, until technological paradigms shift or the software updates.

Many of the big names associated with the post-internet wave, like De Vries, are now mature. Shrewder and slower, less reactive than in the past, the members of this group—Katja Novitskova and Harm van den Dorpel among them—assemble their archives, their troves of found stock imagery, their Plexi sculptures and referential mania. The reader of Deep Scroll can track the wax and wane of jargon across the whole genre’s life span: Noumenal and biodrome have the ring of discourses on their way out; identitarian feels ascendant.

Printed books are all history books now. Regardless of the ideas they contain, their form—the codex, one page following another in unchangeable sequence—preserves a dead episteme. Deep Scroll keeps this past alive only to kill it at every opportunity. Authorship barely matters here; epigrams lend their aura to De Vries’s artworks in the background. Context is effaced by very online jump cuts among texts, images, and visual detritus. Germinal texts, like The Xenofeminist Manifesto (2018), streak past with a kind of lexical Doppler effect.

Here is a primer on the supermarket approach to theory, which has both shaped and framed this entire field of artistic practice that appropriates and twists the aesthetics of networks, of a hypercapitalist surveillance state. The scroll vomits up theorists’ keen insights on humans-as-systems, on pattern recognition and pareidolia, on the increasing value of the hyperreal. Glittering concepts and metaphors appear as if to anchor our understanding of digital mediation: Marc Augé’s “non-places,” DeLanda’s nonlinear histories. Yes, “desiring machines” once helped explain the attention economy, and we can surely trace lesser AI back to the dream of homunculi. Bits of theories work until they don’t, until technological paradigms shift once more or the software updates.

There’s a fine line between aleatory composition and sheer sloppiness. Consider the stunning 2018 Ariella Azoulay essay “Unlearning the Origins of Photography”—here stuffed into mean little boxes, all line breaks collapsed and original stresses (bolded and italicized lines) lost. Affect is flattened. To butcher a scholar’s emphasis and intent through chance procedure is not to reject editing; it’s just to edit poorly. Further, outdated framings are left intact. For instance, sharing space with Azoulay is De Vries’s Katanga Bub, 2011, a press photo of workers in the Democratic Republic of Congo, where tungsten and coltan are mined for mobile devices. The photo remains captioned as largely the story of “the problematic social and political infrastructures” of the DRC. Overlapping directly with Azoulay’s theory of unlearning the imperialism embedded in photographs, a more nuanced caption might have noted the “problematic” ways in which technological production and the extractive capitalism underpinning it constitute the driving neoimperialist force of our time.

Two pages from Anne de Vries’s Deep Scroll (Onomatopee, 2020). Ariella Azoulay, “Unlearning the Origins of Photography,” 2018.

De Vries’s disclaimer on the back cover notes that neither “AISSystem,” “the artist nor other entities involved take responsibility for any possible lack of narration and/or discourse, and/or any questionable layout choices.” This amusing, absurd disavowal reminds us, however, that while both De Vries and “AISSystem” are the book’s editors, we do not know how the edit actually reflects their work in collaboration, or what “AISSystem”—also listed as an AI generator—does, aside from having artworks and writing “stacked, tagged, and compiled.” De Vries cedes all responsibility for interpretation to readers, as long as they don’t judge the book as “regular media.” For an author taking “no responsibility,” De Vries makes plenty of demands on us to read the book the right way. As artists continue to experiment with AI as a tool, such facile evasions about the tool as author will be less convincing. Algorithms are human-trained and shaped. In the case of Deep Scroll, the typesetting, the layout, the choice of text order are all at least partly human choices. Technological frames, however futuristic, do not obliterate history or the author.

“AISSystem” seems to be De Vries’s personal shorthand for an obscure set of operations cobbled together by human editors and different AI generators, some of which learn from text while others rearrange lines or edit at will. The specter of human hands, of human choice, is strong, but we can’t know the rules of the book’s programming without any legible process, order, or constraints for its computational-tool long list. Take the GPT-2 responses, which offer easy, oracle-like appeal. In response to Sam Jacob’s “Rendering: The Cave of the Digital,” a lovely 2017 essay on the “tyranny” of perspective and CGI rendering, GPT-2 responds: “We were not so lucky today. But, I am confident that I am the reality that was created.” The cryptic, evocative lines give a brief ping of pleasure, of both recognition and defamiliarization. But since we don’t know if GPT-2 is responding to the whole essay or to one passage as input, our analysis of the output’s significance is limited.

Page from Anne de Vries’s Deep Scroll (Onomatopee, 2020).

The output of any AI “writer” will depend on many factors: the whole corpus of texts with which it was trained or taught; whether it is reading an essay or excerpts; whether it is deployed in tandem with other generative AI processors that have taken up the text in turn. Each of these choices would have an aesthetic, literary meaning that would complicate the book. A few times, we are told that QuillBot and CS Generator—two paraphrasing AIs—were used, as on Joshua Sokol’s 2017 “The Thoughts of a Spiderweb.” Sokol’s text was dropped into two websites. That paraphrased text was run through AI GPT-2. But why this order? What is the author’s logic? If this approach was thoughtful, considered, shared, the “suite” of AI generators as De Vries uses them—as collaborators, not as mere tools—would start a meaningful conversation. It would push the use of AI here past novelty.

As Oulipo and Alison Knowles showed us decades ago, knowing the constraints under which an author composes a text is instructive. Billions have been and will continue to be poured into machine learning; humanlike is used as a standard for natural language processing research to serve an overriding market imperative, not just to provide computational artists with fresh experimental frames. Nevertheless, in the field of computational creativity, adventurous researcher-poets mobilize GPT-2 in myriad ways. Notable creators of AI novels, poems, photographs, paintings, and texts, such as Allison Parrish, Casey Reas, and Anna Ridler, all train AI models from the ground up. They have made a practice of using machine learning, NLP, and other tools to create new types of visual and written language. These authors tend to be quite transparent about their process, and their descriptions of those processes frame their texts clearly. (Take K Allado-McDowell’s Pharmako-AI, for instance, published earlier this year by Ignota Books, in which the author and the language model GPT-3 produce a stunning novel through a back-and-forth recursive loop that the author describes in detail.) Their narration of process adds a meta-layer to the experience of their work, inviting us to think about how we read.

Page from Anne de Vries’s Deep Scroll (Onomatopee, 2020).

De Vries offers little to none of this transparency, echoing precisely the ways AI has been black-boxed in technology and in culture more broadly for decades. This relegates AI to the status of just another artistic trick to add to a bag of many. The book may be the perfect mirror of our grotesque moment, as we barrel toward loss of agency under a regime of Valley-controlled algorithms that “listen”—reading our speech, learning from our impassioned critiques. Created, shaped, illustrated by algorithms for which everything is just more material, it misses an opportunity to suggest how we might navigate the post-post internet ahead.

Nora N. Khan is a critic, a curator, an editor, and a professor in Digital + Media at Rhode Island School of Design.