Thea Pitman

A great deal has been written in a very short period of time about the impact of AI tools, and most recently AI image-generation tools, on art and on visual culture more generally. In relation to art, where definitions tend to privilege creativity and originality over ‘intelligence’ and mimicry – and value the artistic process and the artist’s intentionality over the end product, whatever it looks like – many critics have argued that there is no such thing as ‘AI art’ and/or that what passes for it is of questionable value as such.¹ At the more generous end of the spectrum, Joanna Zylinska, in AI Art: Machine Visions and Warped Dreams (2020), says that she has chosen to use the term ‘as a proposition, not a typological designation’ (14) in order to be better able to explore the politics around AI and its relationship to art.² At the less generous end of the spectrum, Simona Chiodo (2024) argues that AI images are simply devoid of the ‘meta-sensemaking’ that is core to the function of art (2), and, for most of the critics in this category, AI-generated images presented to the world as art are little more than wallpaper, illustration, screensaver or lava lamp (if animated). They evidence aesthetic but not artistic features, even if they have also proved quite engaging for the general public. And indeed, the references to wallpaper are prescient: this kind of visual production may be exactly that – a deliberate semblance of art as cover-up. As Antonio Somaini (2023) argues, AI-generated images may be part of an effort by Big Tech to engage in ‘art-washing’ to ‘help disguise its most problematic implications as a technology of surveillance, prediction, discrimination, and job replacement’ (78).³

However, leaving definitions of art to one side, the wider field of visual culture is awash with AI-generated imagery. While some critics, such as Lev Manovich and Emanuele Arielli in their Artificial Aesthetics: Generative AI, Art and Visual Media (2024), are balanced and open-minded about the impact of generative AI on both art and visual culture in general, exploring the potential of an ‘extended aesthetics’ of co-creation between humans and AI systems, others are more damning in their assessments of the impact of generative AI on visual culture. Eryk Salvaggio, in his article ‘How to Read an AI Image’ (2023), defines the output of generative AI tools as simply ‘infographics for their underlying dataset’ (84) that quickly reveal its ‘categories, biases, and stereotypes’ (87) and argues that we should endeavour to get over the ‘magic spell’ cast by these ‘bias engines’ (97). Hito Steyerl (2023) goes further, theorising these outputs as ‘mean images’ – mean because they are based on probabilities and tend towards the average, but also mean because they correspond to the dynamics of what is ‘common’ or ‘popular’ – and condemns them as nothing more than ‘hallucinated mediocrity’ (82). Building on this, Roland Meyer (2025) discusses what he calls ‘generic visual content’ and what is popularly referred to these days as ‘AI slop’,⁴ arguing that generative AI imagery is nostalgic and conservative in nature because it is nothing more than a recycling of preexisting content resulting in ‘algorithmically powered aesthetic populism’ (7). And indeed, Gareth Watkins (2025), writing in the New Socialist, goes a step further and identifies this as ‘the new aesthetics of fascism’.⁵ Finally, moderating between these polarised visions of whether AI can generate art or just slop, Jan-Noël Thon and Lukas R. A. Wilde’s most recent anthology, AI Aesthetics: AI-Generated Images Between Artistics and Aisthetics (2025), explores both the artistic and more instrumental potential of AI image-generation tools as well as their ‘aisthetic’ potential – that is, the potential that ‘a situated interaction between humans and AI-generated or AI-augmented outputs (or, indeed, the interfaces of generative AI platforms more broadly)’ has in terms of how users’ ‘sense perception, embodied experiences, and affects are addressed, negated, or modulated therein’ (12). This latter perspective sits well with the situated nature of the AIAI project.

Much has also been written about the ‘representational harms’ of AI. As Tarleton Gillespie (2024) argues, ‘AI systems can deny people the opportunity to self-identify; they can reify social groups; they can traffic in stereotypes; they can demean social groups; and they can erase them entirely’ (2). In relation to visual culture specifically, many critics have lined up to expose racial, gender and other biases that are hard-baked into generative AI tools because of the nature of their training data, the way those data have been labelled (and in which language) and opaque features of the algorithms used.⁶ Some authors have also pointed out that ever greater fine-tuning of data, metadata, parameters and so forth, with regard to questions of representation, is not the answer (Elam 2023, 253–54; Steyerl 2023, 90) and that things can go laughably wrong when the Big Tech companies step in to try to increase diversity through ham-fisted ‘shadow prompting’, where the tool itself adds hidden additional parameters to a user’s prompt (Salvaggio 2023b). Thao Phan (2024) explores the anti-woke backlash against the appearance of ‘Black Nazis’ and ‘Asian Vikings’ as a result of Google Gemini’s 2023 attempt to improve racial diversity in the images it generated through shadow prompting, and Jenka (2023) observes the rise of the ‘American smile’, typical of selfies, in AI-generated photographs of people from different eras and cultures (for example, late nineteenth-century Native Americans, Spanish conquistadores or contemporary Eastern European soldiers) where the historical record indicates that such a smile would not be expected.

Nonetheless, critics have also tended to argue that it is artists who are best placed to critique and ‘stress test’ such tools and their impact on visual culture and on society more generally (Elam 2023; Ivanova 2025). As Michele Elam (2023) argues, ‘the increasingly influential rise of AI artist- technologists, especially those of colour, are among those most dynamically questioning and reimagining’ the ways AI is impacting society (254). On more than one occasion, critics have turned to the creative work of artist and researcher Eryk Salvaggio as an example of this kind of critical AI art. Salvaggio himself, working in collaboration with Caroline Sinders and Steph Maj Swanson (2025), also notes the danger that ‘image generation tools … direct the user toward the production of commercial images that strip away much of artmaking’s critical potential’ and that their simple interfaces ‘encourage … passive consumption – product not process’. However, they also argue that artists need to ‘engage with these platforms from a critical position – without extending the spectacle and reinforcing the emphasis on stereotypes, technical proficiency, and the looting of other artists’ work’. It is precisely in this field – that is, where artists, in the broad sense, critically engage with AI image-making tools to reveal the deficiencies and hidden agendas in those tools – that the AIAI project situates itself.