A work of text in the age of Artificial Intelligence: In a post-ChatGPT world, do we still need human curators?, Ciar O’Mahony
June 2023
Image of Wolfgang von Kempelen’s ‘The Turk’. Photo: Wikimedia Commons.
Ever since ChatGPT was released to the public in November 2022, I’ve been asked with increasing frequency if I, as an independent curator/general arts person, feel threatened by it. One might wonder if curators risk being replaced by ChatGPT, were it to become adept at writing wall texts, artist bios, and catalogue essays.
The threat is suggested by the media's hyperbolic descriptions of ChatGPT and AI in general, with headlines like “Yes, ChatGPT is coming for your office Job” or “Will ChatGPT Take Your Job?” The 2023 Writers Guild of America strike is bringing up issues around how AI can be used in the creative industry of writing for film and TV. I’m reminded of Omar Kholief’s 2014 essay “The Curator’s New Medium,” in which he wondered if the role of a curator could one day be taken over by the “recommended for you” algorithms used by sites like Amazon and Artsy.1
I’ve gone to ChatGPT multiple times since it launched for help with writing. The first time, I was struggling to write an overview of artist and theorist SA Smythe's work for a gallery website. After going around in circles for a day or so, I prompted ChatGPT to “write an article about SA Smythe’s practice.” On a tight deadline and working remotely, I needed this text done fast, preferably without annoying my coworkers. ChatGPT gave me 500 words in nine seconds. Here’s a taste:
I was impressed by how quickly this text was spat out and its close resemblance to something a human would produce, particularly the second paragraph. But, I ultimately didn't use any of it in the draft I submitted. Weeks later, while developing a guided tour of Drink from the River, a solo exhibition by artist Brenda Draney, I prompted ChatGPT to “write a tour script for a Brenda Draney solo exhibition.” I was curious to see its response, as a tour script felt like a more complex request than an article. Here’s a sample of what it gave me:
This impressed me less, mainly due to its tonal weirdness, with phrases like "Welcome to the Brenda Draney tour!" I was, however, taken with linguistic gestures such as "emotionally charged works" and "she draws inspiration from," as they demonstrated how well ChatGPT could produce the niche conventions of bland art writing. Is it good? No. But passable? Sure.
The threat is suggested by the media's hyperbolic descriptions of ChatGPT and AI in general, with headlines like “Yes, ChatGPT is coming for your office Job” or “Will ChatGPT Take Your Job?” The 2023 Writers Guild of America strike is bringing up issues around how AI can be used in the creative industry of writing for film and TV. I’m reminded of Omar Kholief’s 2014 essay “The Curator’s New Medium,” in which he wondered if the role of a curator could one day be taken over by the “recommended for you” algorithms used by sites like Amazon and Artsy.1
I’ve gone to ChatGPT multiple times since it launched for help with writing. The first time, I was struggling to write an overview of artist and theorist SA Smythe's work for a gallery website. After going around in circles for a day or so, I prompted ChatGPT to “write an article about SA Smythe’s practice.” On a tight deadline and working remotely, I needed this text done fast, preferably without annoying my coworkers. ChatGPT gave me 500 words in nine seconds. Here’s a taste:
SA Smythe is a non-binary Black artist, scholar, and activist whose work explores themes of race, gender, sexuality, and identity. Their practice is interdisciplinary, incorporating performance, video, installation, and writing. Smythe's work is deeply rooted in their personal experiences as a queer Black person, and seeks to challenge and disrupt systems of oppression and marginalisation.
[…] Smythe's performances often involve the use of spoken word poetry, and incorporate themes of history, memory, and identity. In their work ‘Manifesto for a Black Future,’ Smythe addresses the legacy of slavery and the ongoing impact of systemic racism, while also imagining a future in which Black people are free from oppression and able to fully embrace their humanity.
I was impressed by how quickly this text was spat out and its close resemblance to something a human would produce, particularly the second paragraph. But, I ultimately didn't use any of it in the draft I submitted. Weeks later, while developing a guided tour of Drink from the River, a solo exhibition by artist Brenda Draney, I prompted ChatGPT to “write a tour script for a Brenda Draney solo exhibition.” I was curious to see its response, as a tour script felt like a more complex request than an article. Here’s a sample of what it gave me:
INTRO:
Welcome to the Brenda Draney tour! Brenda Draney is a contemporary artist from Canada known for her captivating and emotionally charged works. Today, we will take a closer look at her life, her art, and the stories behind her paintings.
ART STYLE:
Draney's art is characterised by her use of vivid colours, bold brushstrokes, and layered imagery. Her works often explore themes of memory, identity, and place. She draws inspiration from her personal experiences growing up in northern Alberta, and her Indigenous heritage.
This impressed me less, mainly due to its tonal weirdness, with phrases like "Welcome to the Brenda Draney tour!" I was, however, taken with linguistic gestures such as "emotionally charged works" and "she draws inspiration from," as they demonstrated how well ChatGPT could produce the niche conventions of bland art writing. Is it good? No. But passable? Sure.
Ciar leading a Tour of Brenda Draney exhibition Drink From the River. Photo: Hyerim Han, courtesy the Power Plant Contemporary Art Gallery.The experience of recognizing the voice of your subculture in ChatGPT’s texts is described by artist Matt Nish-Lapidus in his 2021 article “An Infinity of Traces”:
The Mechanical Turk was a famous fake automaton from the 1800s, which toured Europe tricking audiences into believing it could play chess, when in reality, a man was hidden inside.2 While the Mechanical Turk was a vessel hiding just one person, ChatGPT is a vessel for millions, and I was calling on them with my prompts.
I didn’t end up using any of the Brenda Draney text, either. I’m not against using AI as a writing tool; many artists and writers openly do so.3 My gripes with each text, like the tonal weirdness in the one about Brenda Draney, or the listing of cultural and identity markers so bluntly in the one about SA Smythe, scratched the surface of my larger issue: both texts comprised extended lists of statements, “Brenda Draney is a contemporary artist from Canada” and surface-level descriptions, “Smythe’s performances often involve the use of spoken word poetry,” gathered from extant writing about Smythe and Draney, without expressing an idea. Per Nish-Lapidus: “[GPT 3’s language model] doesn’t produce new ideas, words, or meanings. At best, language-generating systems produce idiosyncrasies that a human projects meaning onto.” My experiments with it confirmed his observation.
And yet, I continue to remain uncertain about my future. Even if neither of the texts it generated for me were actually usable, my goal wasn’t just to string facts together, but to articulate a perspective, and ChatGPT can’t do that. What makes me nervous is the pervasive devaluation of processes undertaken by curators, artists, writers, and other creative professionals that result in thoughtful work. One tweet about the previously mentioned WGA strike exemplifies this devaluation: "Oh no. A writer's strike. Whatever shall we do. If only there was some kind of machine that could endlessly pump out textual content so we didn't have to rely on these flaky humans."
If mainstream culture doesn’t care about ChatGPT’s limitations, inability to form new thoughts in writing, or the fact that it can only generate “textual content” by drawing on millions of text works written by human beings (when put through plagiarism software Turnitin.com, Chat GPT’s text on SA Smythe came back ⅓ plagiarized), then it doesn’t matter that I, a human, can do more interesting, original work, and so can you. With major film studios opening the door for AI to become a regular part of scriptwriting, I don’t see why museums, commercial galleries, auction houses and any other art space you can think of wouldn’t one day go in the direction of AI “textual content” production if it’s cheaper, legal, and their audiences are willing to engage with it. Leaving me with the feeling that it’s not ChatGPT I’m afraid of, it’s my continued precarity under capitalism.
Since GPT-2 and 3 are trained on a massive collection of human-written text, we encounter ourselves and our culture reprocessed and modelled through its output. What appears more like a technical wonder, a speaking machine, might actually be a Mechanical Turk with all of us squeezed inside.
The Mechanical Turk was a famous fake automaton from the 1800s, which toured Europe tricking audiences into believing it could play chess, when in reality, a man was hidden inside.2 While the Mechanical Turk was a vessel hiding just one person, ChatGPT is a vessel for millions, and I was calling on them with my prompts.
I didn’t end up using any of the Brenda Draney text, either. I’m not against using AI as a writing tool; many artists and writers openly do so.3 My gripes with each text, like the tonal weirdness in the one about Brenda Draney, or the listing of cultural and identity markers so bluntly in the one about SA Smythe, scratched the surface of my larger issue: both texts comprised extended lists of statements, “Brenda Draney is a contemporary artist from Canada” and surface-level descriptions, “Smythe’s performances often involve the use of spoken word poetry,” gathered from extant writing about Smythe and Draney, without expressing an idea. Per Nish-Lapidus: “[GPT 3’s language model] doesn’t produce new ideas, words, or meanings. At best, language-generating systems produce idiosyncrasies that a human projects meaning onto.” My experiments with it confirmed his observation.
And yet, I continue to remain uncertain about my future. Even if neither of the texts it generated for me were actually usable, my goal wasn’t just to string facts together, but to articulate a perspective, and ChatGPT can’t do that. What makes me nervous is the pervasive devaluation of processes undertaken by curators, artists, writers, and other creative professionals that result in thoughtful work. One tweet about the previously mentioned WGA strike exemplifies this devaluation: "Oh no. A writer's strike. Whatever shall we do. If only there was some kind of machine that could endlessly pump out textual content so we didn't have to rely on these flaky humans."
If mainstream culture doesn’t care about ChatGPT’s limitations, inability to form new thoughts in writing, or the fact that it can only generate “textual content” by drawing on millions of text works written by human beings (when put through plagiarism software Turnitin.com, Chat GPT’s text on SA Smythe came back ⅓ plagiarized), then it doesn’t matter that I, a human, can do more interesting, original work, and so can you. With major film studios opening the door for AI to become a regular part of scriptwriting, I don’t see why museums, commercial galleries, auction houses and any other art space you can think of wouldn’t one day go in the direction of AI “textual content” production if it’s cheaper, legal, and their audiences are willing to engage with it. Leaving me with the feeling that it’s not ChatGPT I’m afraid of, it’s my continued precarity under capitalism.
Ciar O'Mahony is an artist, curator, and artist-curator based in TKaranto, Canada. They draw on their hybrid practice to create projects exploring education, labor, and the greater economy through the lens of contemporary art. Variously, they have exhibited work, developed public programs, and curated exhibitions for South London Gallery, UK, The Goldsmiths Centre for Contemporary Art, UK, Cubitt Artists', UK, the plumb gallery, TKaranto, The Aird Gallery, TKaranto, and Archer Beach Haus, US. Their work is carried by Art Metropole, TKaranto, and archived by Xpace, Tkaranto. They have been published by e-flux journal, Eater, Liminul Magazine, The Blackwood Gallery, and The Power Plant Contemporary Art Gallery. For more information, visit ciaromahony.space.
¹ Kholief, Omar. “The Curator’s New Medium” in You are here, Art After the Internet, edited by Omar Kholief, 78-85, First published in the United Kingdom in 2014 by Cornerhouse and SPACE
² Nish-Lapidus, Matt. “An Infinity of Traces”. The Society for the Diffusion of Useful Knowledge. October 2021. Accessed April 2, 2023, <https://www.blackwoodgallery.ca/publications/sduk/pronouncing/an-infinity-of-traces>
³ Parrish, Allison. “Language models can only write poetry”. August 13th, 2021. Accessed April 12, 2023. <https://posts.decontextualize.com/language-models-poetry/>
¹ Kholief, Omar. “The Curator’s New Medium” in You are here, Art After the Internet, edited by Omar Kholief, 78-85, First published in the United Kingdom in 2014 by Cornerhouse and SPACE
² Nish-Lapidus, Matt. “An Infinity of Traces”. The Society for the Diffusion of Useful Knowledge. October 2021. Accessed April 2, 2023, <https://www.blackwoodgallery.ca/publications/sduk/pronouncing/an-infinity-of-traces>
³ Parrish, Allison. “Language models can only write poetry”. August 13th, 2021. Accessed April 12, 2023. <https://posts.decontextualize.com/language-models-poetry/>