Art writing in the age of ChatGPT
As one who writes about art for a living, it is important for me to be aware of the rapidly changing cultural, political and technological context in which I operate. Perhaps the most transformative development of our times is the advent of artificial intelligence (AI), which is already a source of great anxiety and controversy, but which is also unstoppable. Artificial intelligence is already impacting the making of art itself, with AI processes entering mainstream art production in ways that will surely be as revolutionary as the invention of photography, if not more so.
Artificial intelligence also has implications for art writing. In recent weeks, I have done some writing experiments with ChatGPT v3, using the OpenAI platform, and the results have been instructive. According to the online Brittanica, “ChatGPT, in full Chat Generative Pre-training Transformer, [is] software that allows a user to ask it questions using conversational, or natural, language”. In responding to these user questions, or prompts, ChatGPT can produce texts that read like a formal essay or article. The generation of these texts is almost instantaneous, and it literally just takes a few minutes and a few well-placed prompts to produce something usable. The technology is rapidly evolving and will inevitably become more sophisticated, over time.
My first experiment was to request a review of an (unspecified) exhibition at the National Gallery of Jamaica. The initial results were very generic and referred to a fictional exhibition on Jamaican art history, but it was clear that the software was drawing from established online sources and therefore came across as somewhat “knowledgeable”. The generated text became far more credible, however, when I added a few prompts that encouraged ChatGPT to add more depth and provided it with some basic information. Had it been about a specific exhibition, the resulting review might have been “publishable”, albeit in contexts where the bar is not placed very high.
My second experiment was to ask ChatGPT to produce a short text on the work of the Jamaican artist Phillip Thomas. The initial result was again very generic, so I asked it to cite specific examples and to discuss the prevalence of realist painting in contemporary African diaspora art. The examples it cited were fictional, even though Thomas’ work has a decent internet presence, so clearly ChatGPT does not access all online sources. The broader commentary, however, sounded alarmingly like the sort of generalizing, truistic language and narratives we routinely see in artist statements and exhibition press releases and wall texts. Here is an excerpt: “Phillip Thomas is a Jamaican artist whose work celebrates the cultural landscape of the Caribbean and explores themes of identity, memory, and history. His paintings are known for their vibrant colors, stylized figures, and surreal landscapes, which draw on the rich visual traditions of the region. Realist painting has played an important role in contemporary African diaspora art, and Thomas is no exception. By depicting people and communities in a realistic and nuanced way, he challenges dominant narratives and stereotypes, and asserts the value and dignity of Black lives”.
I was left to wonder who, had I not disclosed its origins, would realize that this text was, in fact, machine generated.
I next asked ChatGPT to review the current Art of Reggae exhibition at the National Gallery of Jamaica. In refining my prompts, I added some specifics, namely that the exhibition features the top 100 winners of the 2022 International Reggae Poster Contest and that the theme was “women in reggae.” Here is an excerpt of what I ultimately got.
“One of the major positives of the exhibition is the way it highlights the contributions of women to the reggae music genre. Through the posters, the exhibition celebrates female artists, musicians, and other figures who have made a significant impact on reggae music and culture. This is an important step towards addressing the underrepresentation of women in the music industry and recognizing their achievements.
“Additionally, the exhibition also addresses larger societal issues related to gender and social justice. Many of the posters feature themes related to women’s empowerment and the fight against gender-based violence, providing an important platform for artists to engage with these important topics. However, there are some potential negatives to consider as well. One criticism of the exhibition might be that it overly romanticizes the role of women in reggae music. While it is important to celebrate and acknowledge their contributions, there is a danger in portraying women as solely positive and heroic figures without acknowledging the systemic issues of gender inequality and sexism that persist in the industry”.
Most of this “review” is just verbiage and obviously based on broad, routine assumptions related to the exhibition theme. ChatGPT has not “seen” the exhibition and is thus unable to make independent assessments. It, for instance, does not “know” that there are no obvious references in the posters to “gender-based violence”. Paragraph three of the excerpt, however, gave me pause, as it rightly argues that the exhibition “romanticizes the role of women in reggae music … without acknowledging the systemic issues of gender inequality and sexism that persist in the industry”. I had, frankly, not expected such an astute observation, and assume that ChatGPT picked it up from some online discussion, but some critical depth and nuance may thus be possible.
So, what did I learn from my initial encounter with ChatGPT? One thing is that the quality of the results depends greatly on the quality of the prompts and that, with more time and savvy on the part of those who pose the questions, more credible results can be produced in record time. That is, in itself, rather alarming, with all sorts of scope for dishonest abuse.
The most commonly cited concern is, indeed, that ChatGPT is a major threat to education and academic integrity. Educators should familiarize themselves with the technology and learn how to recognize AI-generated texts. They should also take care to work closely with their students, so that they know how well their write, what their writing voice is, and how they think, as ChatGPT is not (yet) able to capture and reproduce such nuances. For those who know what to look for, it is still quite easy to identify AI-generated text, as the tone is flat and impersonal and the platform generally dispenses predictable, risk-averse generalities, without any strong commitments to opinion.
The experiment, however, taught me a lot about the sort of meaningless, routine generalities that often appear in art writing, as these are on full, embarrassing display in what ChatGPT produced for me. What we get in ChatGPT reminds us that good art writing requires a distinctive voice and a clear and informed point of view, which the platform is (as yet) unable to provide. Perhaps, the advent of AI will indirectly have a positive effect on the quality and originality of art writing, and encourage a stronger focus on the production of opinion.
The question arises, however, whether platforms such as ChatGPT could be used as a legitimate, ethically acceptable writing tool that helps to generate texts that are worth reading, and how that would work in practice. For art criticism of any substance and credibility, it is obviously too early, as the technology is not yet able to process sophisticated analytical thought or complex information, but that will likely change in the not-so-distant future (a fourth, more advanced version of ChatGPT was launched while I wrote this article).
What is certain, however, is that AI is deeply transforming our world, with far more to come. We should therefore familiarize ourselves with the technology, rather than to be intimidated or dismissive of it, so that we are empowered to use it judiciously and learn from it as it evolves, while being able to identify and avoid its pitfalls. The art world ought to be no exception.
Dr Veerle Poupeye is an art historian specialized in art from the Caribbean. She works as an independent curator, writer, researcher, and cultural consultant. The second, revised and expanded edition of her best-known book “Caribbean Art” was recently published in the World of Art series of Thames and Hudson. Her personal blog can be found at veerlepoupeye.com.