Traces

Creativity and the nonhuman

Photo by Shannon Potter on Unsplash

In a study on people’s aesthetic judgements of AI-generated text, three researchers presented 1,342 American participants with six poems addressing three topics—impermanence, regret, and memory—from either a first-person or third-person perspective. They told some participants the truth, that ChatGPT generated the poems, while others learned that a human poet wrote the poems.

The participants evaluated the poems on quality, creativity, and enjoyment. Participants who knew the poems were AI-generated rated them as less creative, lower in quality, and less enjoyable in comparison to participants who thought a human wrote the poetry. Their ratings were particularly low for poems that adopted the first-person perspective. The researchers posit that “consumers may have negative reactions to AI-generated content that enters the uncanny valley and mimics content that is viewed as evocative of the human experience.”

While I find the thought of being a mere “consumer” of “content” both grotesque and half-true, I’d like to focus on creativity as a doing word rather than an inherent (human) attribute, holding space for two seemingly contradictory ideas:

1. Human beings possess an incredible capacity for creation and remixing. Many human artists display an astounding level of craft and passion that generative AI, as a non-living non-thinking entity, can only imitate but not originate.

2. Nonhuman entities are an integral part of the creative process—insisting on a human monopoly on meaningful art-making limits how we think about creative practice.

I don’t fully believe in the argument I’m about to make, but I hope it’s at least interesting.

Reassembling Creativity

Photo by Florian Klauer on Unsplash

Perhaps we must expand the category of “creator” to include nonhuman entities, even those lacking intelligence and intent: whatever facilitates the creation of art becomes a creator.

In Reassembling the Social: An Introduction to Actor-Network-Theory, Bruno Latour argues that to study the social, that is to conduct the “sociology of associations” or Actor-Network-Theory (ANT), we should start from the position of naïveté, tracing how associations between heterogenous (human and nonhuman) actors constitute a collective. His understanding of the social expands the role of nonhumans within the social, such that:

there might exist many metaphysical shades between full causality and sheer inexistence. In addition to ‘determining’ and serving as a ‘backdrop for human action’, things might authorize, allow, afford, encourage, permit, suggest, influence, block, render possible, forbid, and so on. ANT is not the empty claim that objects do things ‘instead’ of human actors: it simply says that no science of the social can even begin if the question of who and what participates in the action is not first of all thoroughly explored, even though it might mean letting elements in which, for lack of a better term, we would call non-humans. This expression, like all the others chosen by ANT is meaningless in itself. It does not designate a domain of reality. It does not designate little goblins with red hats acting at atomic levels, only that the analyst should be prepared to look in order to account for the durability and extension of any interaction. The project of ANT is simply to extend the list and modify the shapes and figures of those assembled as participants and to design a way to make them act as a durable whole.¹

Our understandings of the social must thus be “collectively staged, stabilized, and revised” such that the social emerges from meticulous tracing of connections between actors, rather than as a fixed domain or the pre-determined bogeyman called “society.” Drawing on Latour’s work, I view nonhuman actors (like the computer, or generative AI) as part of the assemblage of creative work—and thus, as co-creators—regardless of the extent to which they are used.

Let’s shift to thinking about a creator’s environment in the broadest sense—including physical and/or digital infrastructure, nature, and culture—through the lens of a “conducive” environment for writing.

In the summer of 2022, I attempted to write a novel for the first time. Whenever I sat down to write, the blank screen mocked me. At the time, I lived in downtown Toronto and took the subway to work every other day. During the 30-minute commute, I played Sudoku on my phone; as a child of the internet, I am incapable of enduring a sliver of unmediated reality. One day, I saw someone reading a John Grisham novel on the subway. I switched from Sudoku to a notetaking app and wrote in a state of near rapture about my protagonist’s hunger, “She didn’t need the food; she didn’t even particularly want it. Her hunger was spiritual, a metonymic void.” Then I missed my stop.

Stumbling into my workplace ten minutes late, I mumbled about train delays and the unreliability of the Toronto Transit Commission (TTC). In the detritus of time between urgent tasks, I continued constructing my novel. I made my protagonist Kendi a self-conscious poet because I struggle with poetry, so that “the lines Kendi spits out are disjointed, wrong, misshapen, immutable: they fight her like she used to fight her parents—silently, not responding to tender ministrations, withdrawing just out of reach. The kind of fight she knows how to wage but not defend against.” For the first time, I felt excited about my novel-in-progress; it oozed raw potential.

However, when I got home at the end of the workday, embarrassment and fear slithered back into my mind. Something was fundamentally wrong with what I’d written—until the next subway ride, when my fingers danced over the keyboard as my phone screen swallowed the entire world.

I wrote majority of the (still unfinished) novel during my commute and at work, but also at 1 am when sleep proved impossible to locate so I stumbled towards the nearest subway station, seeking something underground—perhaps closeness to the Earth’s core, proximity to the planet’s secrets—only to find my novel patiently waiting.

Latour suggests that “[a]ction is not done under the full control of consciousness; action should rather be felt as a node, a knot, and a conglomerate of many surprising sets of agencies that have to be slowly disentangled.” I’m no Latourian (I can’t muster the deep, consistent surprise ANT demands or shake my belief in the inflexible bogeyman called “society”) but his ideas are useful for thinking through the connections between seemingly random things/entities.

Of course it wasn’t strictly necessary to write on the subway. But I like the idea of my story having its own story. I am comforted and entranced by everyday rituals—perhaps due to years of Catholicism:

The person reading the John Grisham novel, my workplace, the subway, and the TTC—among many other human and nonhuman actors—constitute the assemblage of my novel attempt. They are my co-creators; without them, the novel would not exist.

Person descending into Queen Subway Station, Toronto, Canada. Photo by LinedPhoto on Unsplash

Beyond the Individual

My definition of “creator” may be too expansive. If I ever finished and published that novel, I wouldn’t list the TTC as a coauthor (although I find the idea amusing). Nonetheless, the myth of the creator as a solitary genius is clearly unrepresentative.

Art—and thus, creativity—has always been a collective practice, considering the guardians who shape children’s artistic and intellectual development, librarians who introduce children to life-affirming stories, teachers who acknowledge and encourage students’ creativity, staff at magazines and presses who foster literary community, editors who provide incisive but kind feedback, intriguing strangers who act as inspiration for characters, and everyone who advocates for a piece of art even before the creator believes in it. Acknowledging the role of nonhuman actors within the vast network of creativity builds upon the necessity of acknowledging human actors.

While I don’t dispute the uniqueness and aesthetic value human beings bring to art, art originates beyond the individual. As Roland Barthes suggests in “The Death of the Author” (transl. Stephen Heath) a text “is not a line of words releasing a single ‘theological’ meaning (the ‘message’ of the Author-God) but a multi-dimensional space in which a variety of writings, none of them original, blend and clash. The text is a tissue of quotations drawn from the innumerable centres of culture.”² Nonhuman actors—including generative AI—form part of the complex tapestry that makes art possible.

However, generative AI is not a neutral tool. I’m persuaded by this argument in AI Snake Oil against the current wave of AI criti-hype: focusing on the long-term, existential, “sci-fi” risks of theoretically all-powerful AI makes the models seem more advanced than they actually are, and obscures concrete risks in the near-term. I think the comparisons of AI criti-hype to recently failed hype cycles for cryptocurrency and the Metaverse are extreme—generative AI at least has varied utility and clear(-ish?) pathways to mass adoption. Regardless, rather than worrying about the existential risk of large language models like ChatGPT “replacing” writers, it may be more prudent to worry about AI-generated work flooding an already over-crowded literary arena or AI as a tool to accelerate the concentration of wealth and deepen inequality.

Since generative AI models are trained of data extracted from the internet—majority of the data used to train GPT-3 originated from the Common Crawl dataset, a massive repository of data scraped from the internet from 2008 onwards, as well as two datasets of books, and Wikipedia—most models exist as a result of theft. For instance, Sudowrite, a model for creative writing based on GPT-3 (marketed as the “AI writing partner you always wanted”), demonstrated the presence of fanfiction in the GPT-3 training data.

The fervent desire of some generative AI enthusiasts to sever art from the artist, going as far as copying specific artists’ styles without compensation, reflects the ongoing devaluation of art and artists.

I’m also partial to the argument that the rise of generative AI may be analogous to the Industrial Revolution, due to the contraction of already precarious jobs in creative industries and “knowledge”/white collar professions. I won’t stake out a definite position on the future of generative AI; we shall see. As for what AI-generated art (perhaps “AI-assisted art” depending on the level of human involvement) is worth: we shall also see.

I’m worried about how “disruptive” technology, however harmful/unethical, is presented as an inevitability we must learn to embrace like a friend.

Photo by oktavianus mulyadi on Unsplash

The Theft of Joy

I have never used generative AI for creative purposes, except when conducting research for an independent study course on AI and creativity. Despite my arguments in favour of nonhuman creativity, I intend to resist using AI in my creative practice for as long as possible.

I view writing, particularly creative nonfiction, as a confrontation on the page—an iterative process of working with complexity to facilitate self-actualization. As bell hooks writes in Remembered Rapture, “That moment when I whirl with words, when I dance in that ecstatic circle of love surrounded by ideas, is a space of transgression. There are no binding limitations; everything can be both held and left behind.”³ The ecstatic “whirl with words” drives me to seek improvement on every level, from experimentation with form, to the placement of every punctuation mark and the rhythm of every phoneme. (I exaggerate, but only a little.) This almost sickly love of detail means my writing takes longer than I would like.

However, it’s life-affirming to linger over my work, hoarding time to write purely for the joy of creation. While I enjoy charting my progress from one piece to another, I cherish even more the potential for continuous improvement: the titillating certainty that I will always be (un)learning how to write, making mistakes, stumbling into beauty. Programs like Sudowrite—marketed as “autocomplete on steroids”—aim to take away the discomfort I find productive and joyous.

(Using generative AI at work will likely be a necessity—but creative writing is not necessary, therefore I feel no pressure to be “competitive,” only as good as I can be, whatever “good” means. I finally understand the writers who favour longhand over using computers.)

Perhaps widespread usage of generative AI will open up new ways to think about creativity, and create new norms for what we consider “good” and “great” art. Maybe someone is already co-creating the next paradigm-shifting masterpiece with ChatGPT or Midjourney.

Nonetheless, I find the idea of subordinating any part of my writing—and thus, any part of my joy—to generative AI grotesque. I’d rather not write at all.

---

ICYMI: Early last month I wrote a post for

Inner Life

on Yvonne Adhiambo Owuor’s novel Dust—

Dust: Archiving Haunted Silences

Thanks to

<Mary L. Tabor>

for the invitation to post on Inner Life. You can also check out her newsletter

Mary Tabor "Only connect ..."

. I recommend her piece on ageless creativity.

1. Latour, Bruno. Reassembling the Social: An Introduction to Actor-Network Theory. Oxford University Press, 2005.

2. Barthes, Roland. “The Death of the Author.” Image Music Text, translated by Stephen Heath, Fontana Press, 1997, pp. 142-148.

3. hooks, bell. Remembered Rapture: The Writer At Work. Henry Holt and Company, 1991.