The Brooklyn Rail

Brendan Dawes triptych Altarpiece: The Divinity (2025) employs early modern Christian iconography to implicate vaguely held notions—about being, creation, consciousness, imagination, responsibility, self-determination—that are seeping into the discourse of AI and AGI (artificial generative intelligence). The arts’ increasing reference to the early modern invites comparison to a similar period of flux and disorientation, but also an assessment of concepts, histories, values… I can’t unpack all of it here, but what follows is my plodding attempt towards an overview of the topics and problems, associations and situations, of AI for the arts.

The outcry surrounding the recent Christie’s auction, Augmented Intelligence (February 20–March 5, 2025), reveals confusions about what AI is or does. It’s been applied to robots and chatbots; search engines and social media filter bubbles; voice assistants like Siri or Alexa, and now ChatGPT; and text-to-image/sound/video/text generators, but also a bunch of other things. Similarly, the auction presents an assortment of practices, but I’m not sure that clarified the term or what the arts, in particular, mean—what kind of “it” are we dealing with when presented with such a broad range?

Well-intentioned conversations collapse. The frustration of such flailing undermines the very efforts to manage the problems we see in, or forecast for, AI. I start with seven common critiques raised around generative AI, which leak into anything associated with any form of AI. All are relevant to the arts, but helpfully disentangled, even though tied to one another. Just listing them reveals a wicked problem. That leads me to lay out seven issues faced by the arts in trying to engage AI.

The arts operate within ambiguity, which is a strength we wield because we know how to define and connect, compose and deconstruct. Yes, we should be worried, but that was true a while ago, and doing something necessitates understanding what part of our worries we wish to address. This list isn’t meant to be the final word, but some touchpoints that could orient what we mean or want to wrangle.

 

Part I: Seven Common Critiques of AI
For the sake of expediency, I have presented each critique in alphabetical order with a simplified common claim and limited response to introduce the complexity that unfolds.

Bias
Claim: AI is sexist and racist, with limited perspectives.
Response: This is not new. For over a dozen years, data studies scholars and artists have expounded upon the bias in filter bubbles, search engines, content moderation, and now datasets undergirding foundational models. What exists online and within datasets is what gets processed, regardless of its accuracy or significance. Additionally, how code is written—its efficiencies, parameters, constraints, language, stack, and more—reveal values, preferences, assumptions, and thus ideologies. The problem is systemic and exacerbated by value systems designed into networked culture (let alone society).

Creativity
Claim: AI isn’t creative because it’s just a machine rehashing existing work or simply churning numbers.
Response: Such arguments revisit the rejection of early photography, but also appropriation, conceptual, and code-based art, even before resistance to digital photography or current tech applications. Sometimes aesthetic appreciation depends on understanding the creative approach to underlying processes. Beyond that, “creativity” itself flounders (scholarly interpretations jostling since Alfred North Whitehead established the term, linking it to innovation and novelty). This introduces metaphysical quandaries. Is creativity only human? Is it transcendental, some drive of the cosmos? Does it appear spontaneously or via discrete processes we can cultivate and perhaps teach? Is it a spectrum? Is it the only value for art? These arguments become highly personal and pretty clearly lodged within complex worldviews.

Economic/Labor
Claim: AI will cause widespread job loss.
Response: Probably. Automation aims to streamline work, so is associated with job loss. People rightly worry that social infrastructures won’t successfully address the consequences. Also, the machinery enabling such computation depends on a global workforce with human and labor rights issues apparent at every stage, from mining and manufacturing to corporate offices. If you doubt that last one, then I recommend the book by photographer Mary Beth Meehan and media studies scholar Fred Turner, Seeing Silicon Valley: Life Inside a Fraying America. Also, access to public generators has subscription costs that may seem inconsequential in the Global North, but rout usage elsewhere.

Environmental
Claim: AI will destroy the planet.
Response: Wide scale planetary change is certainly apparent, and it’s not improved by the supply chain enabling AI. Fossil fuels generate the electricity necessary to charge our devices, cool data centers that store information, let alone train foundation models or produce outputs. Recognizing that carbon costs are always proximate, no one doubts that querying a search engine requires less than submitting a prompt to ChatGPT.

Legal
Claim: Generative AI is stealing people’s work.
Response: The online scraping of content to amass large datasets necessary for training foundation models enraged individuals and companies whose language, sound, and image works were culled without permission. Great legal minds are trying to figure out what existing structures or novel arrangements may be necessary. The fair use doctrine allows limited use, particularly if it is “transformative” (making questions of creativity reappear). The author Cory Doctorow argues such scraping is not a violation. On the other side of this process, artists of all kinds want to know their works using generative systems are protected. The Copyright Office recently reversed its former refusal to accept such works by assessing the degree of human involvement in the process of making.

Realism
Claim: AI is ridiculous and insufficiently realistic, but also producing seemingly true content that confuses audiences.
Response: Early image-making processes are often derided for their lack of “realism” (VR has this same problem) as evidenced in the “uncanny” surrealistic quality of content (many fingers, hybrid creatures, etc.). The drive for realism deserves to be queried. Last year, three philosophers argued these are not “hallucinations,” but bullshit as defined by Harry Frankfurt’s 2005 treatise.

Social
Claim: AI is ruining society.
Response: AI seeks to expedite industry processes, decision making, and solution finding. General acceptance of online content, whether generated or shared on social media sites, and the abandonment of the fairness doctrine (at least its ideals) contributes to a culture of credibility; the US Surgeon General “Advisory on Misinformation” (2021) warned as much. Skills and employment needn’t oppose civic education or liberal arts. Automating socio-political decisions rather than valuing deliberation and devising compromise negates democratic process (see Eryk Salvaggio on this coup).

 

Part II: Problems Situating AI
Those critiques are entangled. Discussions of AI dissolve because this thing exemplifies a “wicked problem,” an idea from the 1960s, defined by design theorists Horst W. J. Rittel and Melvin M. Webber in “Dilemmas in a General Theory of Planning” (1973). Wicked problems are symptomatic of other problems and hard to define or delineate because their qualities (constraints and resources needed to solve them) change over time; their solutions can’t be perceived as merely experimental, since every applied effort has a significant impact, nor can those efforts be understood as right or wrong, but better or worse in terms of impact. A wicked problem has many stakeholders, often with different worldviews, which means their framing of the problem alters the value system by which a solution might be accepted.

AI represents a wicked problem by exemplifying how divergent worldviews struggle over the proper functioning of society. AI also becomes a wicked problem because it is proposed as an answer to some of the questions within those debates, often without considering the impact on others. Solutions for its effect(s) on the world depend on what stakeholders think it is, does, or might enable. What if “it” isn’t just one thing?

Terminology
What do we mean by “AI”? The term is widely used though its vagaries confound, even as anxiety, fear, and repulsion imply a sense of knowing. The historian Thomas Haigh argues it is a branding concept more than anything else. It means different things at different times and in different contexts.

These days, general usage suggests generative AI (outputs from publicly accessed large language or diffusion models, like ChatGPT or Stable Diffusion, etc.). But conversation then enfolds other systems and problems. The panoply of processes represented in the Christie’s auction Augmented Intelligence include relatively simple algorithms from many decades ago as well as this recent batch of products.

In 2022, Emily Tucker, the Executive Director of the Center on Privacy & Technology at Georgetown Law, argued to specify model and process in “Artifice and Intelligence.” Amelia Winger-Bearskin did that for Nearest Neighbor (2025), a digital acrylic print on canvas. The title points to her use of a nearest neighbor interpolation algorithm for the underlying design of the work, the proximity that AI has to every area of our lives, and who/what we ignore as neighbors. Specifying nomenclatures can initially feel like terminology soup or techno-fetishism, but confrontation breeds familiarization and starts the necessary work of learning differences among the types of models or algorithms subsumed under “AI.”

Effects
Use of algorithms, softwares, or models in the arts needs differentiation from their adoption in other industries like finance or medical research. For example, Photoshop fill tools “do” different things in journalism versus the arts.

A nearest neighbor algorithm might be a k-nearest neighbor (KNN) sequence that predicts the value of a new data point and so contributes to the design of an artwork. Alternatively, it might be an approximate nearest neighbor (ANN) algorithm used to identify objects in a large database, frequently applied in computer vision. KNN will search and sort an entire database, but ANN can terminate as soon as a match is found, illuminating how these systems in government watchlists erroneously flag people. This leads to distress but also darkly humorous art, like Hasan Elahi’s Thousand Little Brothers (2014), which uses neither of these algorithms but is a response to the problems such cause.

Geoffrey Hinton—who just won a Nobel Prize in Physics for work done in 1983–85 that is now considered foundational to current computational neural networks—continues to warn about the unregulated development of AI (familiar with it given his former work at Google), despite its potential within medicine (for which he has great hopes). A deep neural network helped scientists source some molecules possessing bactericidal activity in their search for new antibiotics, given increasing resistance to existing ones. A recursive neural network was used to identify political ideology bias within sentence structure, which also foreshadows worrisome applications that will reinforce existing self-censorship and cultivate official-speak.

Just as oil painters turn to refined linseed oil, stand oil, or Liquin for different purposes and effects, even similar algorithms (like the various neural networks) do different things.

Creative Practice
There is no such thing as “AI art.” I say this to emphasize Haigh’s point that “AI” is a branding tool. “AI art” does not clarify a kind of practice, either in terms of process or content.

When we hear “AI art,” it might mean art produced using publicly accessible generators like Midjourney that don't require knowledge of coding, mathematics, or engineering. It might mean a kind of quasi-surrealistic visual, in the same way that NFT came to mean, in the popular imagination, some monkey picture. Many of the works in the Christie’s auction include smart contracts (the blockchain function for NFTs), but none of their marketing invoked NFTs, as might have occurred three years ago (to Haigh’s point).

Some might think of “AI art” as something involving algorithms, which occurs when situating “pioneer” digital artists, like Vera Molnar or Herbert W. Franke, now celebrated for their generative art (to be distinguished from generative AI). Such predecessors raise questions for contemporary artists with coding practices, who would not readily identify as AI artists.

Calling something “AI art” might mean work developed from datasets. But when artists use large or small datasets, they are also saying something with that choice. Small datasets are often a rebuttal to the extractive, consumer-based values linked to public generators. The personal sorting and coding seeks to tend a local garden, as in the photographs taken by Sofia Crespo and Anna Ridler that went through different models to produce Long Short Term Memories (sketches from a garden in Argentina) (2025), a video work accompanied by twenty-four Polaroids. But those prints are already synthetic images: their respective photographs from time in Argentina were translated and retranslated using various models, moving from analog to digital and back to create the prints, which were then used for training a model that delivered the digital image content in the video.

If coding or this kind of machine learning process is excluded, then what of modifications to the outputs of generators? Efforts to eliminate anything evincing artistic alteration leaves us with the relatively flat suggestion that “AI art” is an output from an unrefined prompt lacking subsequent modification. These might be easily derided except for highly conceptual projects like Spectacle (2022) by Sterling Crispin, which situates Guy Debord’s ideas from Society of the Spectacle (1967) for this current moment.

Histories
Though seemingly academic, the different genealogies for this thing, “AI,” alter how the conversation unfolds. In the arts, do we think about history as starting with automata—that is, the appearance—or with its computational and algorithmic developments, focusing therefore on its underlying processes? The choice introduces different issues, interests, and identities.

Automata link to mythic desires for self-generation and can connect to mechanical robots, bio-engineering efforts like cloning (CRISPR), as well as computer vision enabled drones and machine learning robots. Sougwen Chung’s painting Study 33 (2024) uses an EEG headset and computer vision system to track her body movements, feeding that data into a custom robot, D.O.U.G._4 (Drawing Operations Unit: Generation_4). This lineage focuses on a three-dimensional, self-propelling entity, though currently one tethered more clearly to a human operator.

Turning to the computational and algorithm narrative invokes cybernetics, military-industrial relations of the twentieth century (apparent once more), and a mind-machine analogy that disregards embodiment. It focuses on intelligence (however that might be defined) and computational prowess for expedited reasoning, with less concern for the shape/form of the entity.

Other disciplines may introduce alternatives to these two delineations, while painting, printing, photography, theater, and film inform and form expectations, as do evolutionary biology, psychology, and more. Any genealogy includes metaphysics, ethics, and epistemologies with similarities to, but also meaningful differences from, another one.

Cultural
Nowhere is there such fear expressed about AI than in the wealthiest, most privileged country in the world. This light paraphrase came from a panel Q&A at the recent College Art Association (yes, academic conferences can be polemical and fun). Such sentiments are not uncommon among those situated outside the USA (or parts of Western Europe). Anxieties and adoption values vary in different regions and cultures.

Many creatives with access to fewer resources express interest and curiosity about the potential uses of this “tool” (remembering, hopefully, that the tool is already many different tools). Most don’t assume that tech devices, platforms, or services will be “user friendly” or designed with them in mind. The need to hack or alter what they get to suit their infrastructure or cultural expectations is common. This is true for technology and other systems (for example, aid organizations’ expectations of resilience, or capitalist market practices, but I digress).

This is also part of a conversation on un/marked nature of the term, notably addressed by the Indigenous Protocol and Artificial Intelligence Position Paper (2020). DeepSeek made some talk about Chinese AI and how national attitudes permeate that software. Can we recognize ChatGPT as “USA-AI”? Scholar Genevieve Bell discussed the place of AI back in 2017. Of course, such marking potentially ignores the global entanglement of these systems, but it also asks whether unlikeable things in corporate operating systems reveal ties to situated ideologies, too.

Archival
Just as many artists track shifts in different layers of Photoshop to remember where effects are, so do curators, conservationists, and collectors often need to know the stack of an AI artwork. There are no norms yet for identifying the use of algorithms or models as a medium in captions (note the awkward effort I imposed even within this article).

Christiane Paul spoke eloquently to the Baer Faxt in September 2024 about digital art’s collection management challenges, which need “to identify a work’s acquisition components and all the media, materials, and technical specifications involved, for example commercial or open-source software, custom software etc.” Third-party dependencies like “corporate software components—from AI models to data sources or networks—necessary to keep other elements of the work functional” can’t be acquired.

Institutions’ uncertainty about long term maintenance discloses the difficulty around caring for and about these systems. Care is necessary to sense them as something to be socialized by us rather than imposed upon us.

 

The End is Just the Beginning
Some suggest finding kinship with machines, oceans and plankton, polar bears and bees, let alone other people. That’s the challenge staged by Amelia Winger-Bearskin’s Nearest Neighbor work, which donates the proceeds from the sale to Community Solutions, an organization that has helped fourteen communities achieve “functional zero” homelessness. Doing so requires data analytics, which is to say AI softwares designed to help house different homeless populations (truly, an incredible story).

Neither AI nor the unhoused have a defined place in our society, though they are certainly present among us. We largely ignore both, but complain about a crisis. At the start of COVID, the playwright Kat Mustatea launched a performance project on Instagram retelling Dante’s Inferno for the contemporary through language from GPT-2 with stock photography and image editing software, all linked to what we call “AI.” The guide through Voidopolis (2023) was a hobo.

What might we gather in looking and thinking more carefully? As we proceed with vital concerns for our planet, global and immediate communities, and the arts in all their forms, a critical engagement with AI may disclose ideas and practices for other causes that matter to us.

The temptation to rectify the prevalent “move fast and break things” or “muzzle velocity” ethos with quick assessments undermines the astute ambiguity that the arts offer. We may worry less and do more if we slow down, discern and specify, ask questions and consider disconcerting proposals. Not only tech moguls and engineers, but also we, with our participation or absence, are designing what will be.