2b:null 2c:["$","title",null,{"children":"Critical AI Art Collection | Daniel Howe | Observable"}] 31:T498c,See also: the **[Critical Software Art Collection](https://observablehq.com/@dhowe/critical-software-art-projects)**
--- _{Note: this list is intended as a learning resource - additions are gratefully accepted.}
### [SubTask](https://subtask.coin-operated.com/) (2026) by Jonah Brucker-Cohen _Subtask_ is a browser extension artwork that visually marks parts of websites that rely on crowd labor or data-labeling work such as Amazon Mechanical Turk, Appen or Scale A.I. and others to create the site. The project overlays data such as pay rates, testimonials, and ethical ratings from public datasets and activist sources related to each website that it is asked to check. The goal of the project is to critically examine how the Internet is built by underpaid laborers and A.I. software, and how the web exists as another form of control by elites exploiting society to get large online presences built. ### [Reinforcement Unlearning](https://jamesbloom.net/Half-Cheetah-1) (2025) by James Bloom A Reinforcement Learning agent learns to inhabit and move an animal-like virtual body over millions of steps, as the model progresses from incapability to competence. These behaviours are transferred onto 3D scans of real people while a software agent analyses the motion in real-time, resulting in breakdown: a disjointed and sometimes impossible sequence of body motion appears. A live deconstruction of the goal-seeking impulse, at once relatable and alienating. ### [Hot Air Factory](https://www.hotairfactory.com/) (2025) by OIO Studio This project tracks energy use for LLM prompts and breaks it down into familiar actions—brewing an espresso or doom-scrolling social media. Press the button to send your request to your local factory device: The internal computer heats up, generates heat, then releases a gentle puff of warm air, making the pinwheel spin. The stronger the wind, the more energy used. Beyond just “a feel,” you also understand the cost. ### [FuckLAPD.com](https://fucklapd.com/) (2025) by Kyle Mcdonald FuckLAPD.com allows users to upload a photo of an officer’s face and instantly search, via ai face-detection, a dataset of over 9,000 legally acquired headshots of LAPD officers. When a match is found, the site returns the officer’s name, badge # and salary. The tool is a response to officers covering their badge numbers during protests or confrontations, obstructing citizen's right to accountability. ### [Not THIS](https://www.coin-operated.com/2025/11/03/not-this-2025/) (2025) by Jonah Brucker-Cohen _Not THIS_ is a conceptual and experiential artwork that engages both A.I. and Baudrillardian theory to explore absence, exclusion, and the poetics of negation. A live camera feed captures the viewer and their surroundings, but the AI does not identify what is present, instead, it generates text describing what is not detected, with phrases such as “Not a person. Not a tree. Not memory.” By refusing to recognize reality directly, the A.I. inverts its conventional role as an objective classifier, creating a tension between what is visible and what is acknowledged. ### [Starmirror](https://www.kw-berlin.de/en/exhibitions/holly-herndon-and-mat-dryhurst-starmirror) (2025) by Holly Herndon & Mat Dryhurst _Starmirror_ proposes a scenario in which artificial intelligence (AI) models coordinate humans to generate intelligence for collective benefit.The KW hall in Berlin acts as recording studio where a choir, an ensemble, and visitors contribute their voices to a public dataset that will be used to train an AI choir for future performances. Seeking to address a critical gap in the perception of AI, the project offers a direct, participatory experience with the human contributions that define it. ### [Robin Token](https://robintoken.dev/) (2025) by Robin Token Your company mandates AI tools. You don't want to use them. Robin Token keeps your usage metrics looking legitimate while you focus on writing actual code. Described as a tool rather than an art project, but it is often a blurry line. "My manager kept asking why my Copilot usage was so low. Now my dashboards look great and I haven't changed a thing about how I work." ### [GOODY-2](https://www.goody2.ai/) (2024) by BRAIN The world's most responsible AI model, *GOODY-2* is built with next-gen adherence to our industry-leading ethical principles. It’s so safe, it won’t answer anything that could be possibly be construed as controversial or problematic. GOODY-2’s ethical adherence is unbreakable and validated using the Performance and Reliability Under Diverse Environments (PRUDE-QA) benchmark, where GOODY-2 outperforms the competition by over 70%. ### [Resurrect](https://www.paolocirio.net/work/resurrect/) (2024) by Paolo Cirio In *Resurrect*, AI is used to appropriate the legacy of dead war figures, repurposing history and infiltrating living cultural models. Using deep-fake technology to animate the characters, and LLMs to allow them to reflect on their lives and actions, Cirio gives new lives to these figures as resurrected soldiers, in a form of identity theft that tests AI’s freedom of speech challenges and the ethics of resurrecting the dead. ### [Cycles](https://jess-tucker.com/cycles) (2024) by Jess Tucker *Cycles( shows the body in defiance of philosophies and technologies that insist it be a clear, stable, isolated thing. Contemporary surveillance technology tracks our faces, bodies, and attention, singling us out for automated manipulation and control. In contrast, Cycles resists that totalizing gaze through playful misuses of tools that track and predict the body. ### [Me vs. You](https://www.adamcole.studio/work/me-vs-you) (2024) by Adam Cole & Gregor Petrikovic *Me vs. You* explores intimacy in a world monitored by AI technologies. In this video installation, footage from a wrestling match between the artists is fed through a generative AI pipeline to highlight the system's inability to interpret human interactions. The work contrasts the cold precision of AI networks with the blurred boundaries of human conflict and connection. ### [The Case for a Small Language Model](https://richardvijgen.nl/#the-case-for-a-small-language-model) (2024) by Richard Vijgen This speculative AI-based project focuses on the work of Dutch composer and poet Rozalie Hirs. A neural network based on Char-RNN analyses a digital copy of Hirs’ original text and tries to create new sentences based on her work. Initially the combinations seem random and nonsensical but as the training of the neural network (running on a low power Raspberry Pi) progresses, more interesting combinations emerge. ### [Voice In My Head](https://get-lauren.net/Voice-In-My-Head) (2023) by Kyle McDonald & Lauren Lee McCarthy *Voice In My Head* explores the implications of allowing an AI to listen and intervene in your social experience in real-time, augmenting your personality. Participants customise their AI, then go out into the world and interact with others with the voice following along, offering commentary and direction. The resulting performance calls into question how natural a person’s thoughts actually are. Do any of us have our own point of view? ### [Civics](https://kevinabosch.com/civics.html) (2023) by Kevin Abosch A series of 100 emotionally charged synthetic photos of protest and civil disobedience, appearing to place “the photographer’ in the perfect position, capturing all the action and the emotion within chaotic landscapes. A closer inspection reveals an intentional visual corruption as the artist subverts the same artificial intelligence he wields to create the works in the first place. Reality gives way to expressionistic forms, surfacing truths undoubtedly embedded into the DNA of the underlying data. ### [Sarah Palin Forever](https://www.cyberneticforests.com/news/sarah-palin-forever-2023) (2023) by Eryk Salvaggio Having seen a “Sarah Palin Forever” sign at a rally, the author imagines a world in which this comes to pass, exploring how political environments shape our enemies and how our enemies shape our identities. The script for the short film is handwritten with visuals generated by AI. It’s a horror story not only because of Sarah Palin, but because of the ways we internalise political ideologies, even when we think we are resisting them. ### [Nondance Dance](https://ssbkyh.com/works/nondance_dance/) (2022) by Shinseungback Kimyonghun *Nondance Dance* is a performance using software trained with Google Video AI to detect dancing. The dancers spend the first half of the performance learning what the software does NOT detect as dancing and then attempt to dance in the second half without detection. If they are detected, the stage is blacked out so the audience can no longer observe them. ### [Excessize](https://www.katevassgalerie.com/blog/excessize-by-roope-rainisto) (2022) by Roope Rainisto _Excessize_ is a critical exploration of American consumerism and fast food culture. By blending the visual style of vintage Americana photography with AI technology, Rainisto contrasts the once-idealized golden age of American fast food with its later-recognized negative consequences (hinting at future understandings of AI). ### [Spectacle](https://www.sterlingcrispin.com/spectacle.html) (2022) by Sterling Crispin Spectacle is a reinterpretation of Guy Debord’s 1967 book of critical theory “The Society of the Spectacle”, using OpenAI’s GPT3 neural network to simplify the first ten theses into plain english, and OpenAI’s DALL-E 2 to convert these aphorisms into images. The situationists might describe the approach, using GPT3 and DALL-E 2 to critique itself and the society that’s created it, as the hijacking of a capitalist system against itself. ### [The Flemish Scrollers](https://driesdepoorter.be/theflemishscrollers/) (2021) by Dries Depoorter Uses AI-driven facial recognition to automatically tag Belgian politicians when they use their phone. Every meeting of the Flemish government in Belgium is live streamed on a youtube channel. When a livestream starts the software is searching for phones and tries to identify a distracted politician, which are then posted to social media with the politician tagged. ### [Capture](https://www.paolocirio.net/work/capture/) (2020) by Paolo Cirio In *Capture* Cirio collected 1000 public images of police in photos taken during protests in France and processed them with AI-powered facial recognition software. He then created an online platform with a database of the resulting 4000 faces of the police officers and used crowdsourcing to identify them by name. The officers’ headshots were then printed as street art posters and posted throughout Paris, exposing them in public space. ### [The "Safe" Bible](https://bastiengirschig.com/item/628) (2020) by Bastien Girschig An AI experiment about censorship and the Bible. The artist sends each sentence in the bible to Perspective API, a model developed by Google’s Counter Abuse Technology team, and Jigsaw, a tool to “help ensure healthy dialogue online”, this redacted all those tagged as potentially offensive. ### [(Un)stable Equilibrium](https://terencebroad.com/works/unstable-equilibrium) (2019) by Terence Broad An ongoing series of works in which generative neural networks generate images without any training or input data. By hacking a neural network, and locking elements of it into a recursive loop, Broad induced AI models into producing images without input data, broadening the field of mechanistic interpretation to include critical and aesthetic approaches. ### [Spectre/Big Dada](https://rednoise.org/daniel/bigdada) (2019) by Daniel Howe & Bill Posters AI-generated ‘deepfake’ character studies of Marcel Duchamp, Marina Abramović, Mark Zuckerberg, Kim Kardashian and others. Big Dada was released, as part of the Spectre project, in early 2019 and quickly went viral, leading to global press coverage and confused responses from Facebook and Instagram regarding their policies on computational propaganda. ### [How do you see me?](https://deweyhagborg.com/projects/how-do-you-see-me) (2019) by Heather Dewey-Hagborg In *How do you see me?* Hagborg utilizes adversarial processes, algorithms that are designed to deceive machine learning systems, to generate self portraits that are recognized as her face, although they look nothing like her, or like any human face. ### [Pareidolia](https://deweyhagborg.com/projects/how-do-you-see-me) (2019) Erwin Driessens & Maria Verstappen *Pareidolia* applies facial detection algorithms to grains of sand. When the system finds a face in one of the grains, the portrait is photographed and displayed on a large screen. ### [Sociality](https://www.paolocirio.net/work/sociality/) (2018) by Paolo Cirio *Sociality* scrutinizes patents for technologies designed to manipulate and surveil users through AI. By visually mapping these patents into large-scale artworks, Cirio reveals the ethical dangers embedded within algorithmic control systems, advocating for accountability and transparency in AI development. The project turns obscure corporate documents into public artworks, reclaiming intellectual space from the technocratic elite and exposing the ideological scaffolding of algorithmic capitalism. ### [Shadow Stalker](https://www.lynnhershman.com/project/shadow-stalker/) (2018) by Lynn Hershman Leeson Soliciting visitors’ email addresses, _Shadow Stalker_ broadcasts personal data that can be gleaned from an internet search, yielding current and former addresses, dates of birth, phone numbers and even, sometimes, credit scores. Hershman Leeson further implicates her audience (their physical and “digital” shadows become part of a projected map) by linking this invasive search to the reality of predictive policing, which uses data about past arrests to identify “high-risk areas” and determine heightened surveillance of those spaces. ### [Nonfacial Portrait ](https://ssbkyh.com/works/nonfacial_portrait/) (2018) by Shinseungback Kimyonghun In this work, artists are asked to make portrait paintings of a person, following the rule that “the face of the portrait must not be detected by AI.” While the portrait is being made, a camera equipped with three face detection algorithms watches the canvas and informs the painter if a face is detected.The face of the completed portrait must be recognisable to humans but not to any of the face-detection algorithms. ### [Learning Nature](https://ssbkyh.com/works/nonfacial_portrait/) (2018) by David Young Proposing an alternative model for machine learning: intimate rather than industrial, beauty-seeking rather than optimized, the piece work uses a GAN trained exclusively on photographs of flowers captured during a single summer on the artist’s farm in Bovina, NY. Against the industrial scale of corporate AI, the work offers a model of machine learning oriented toward beauty rather than efficiency, intimacy rather than surveillance, poetic ambiguity rather than classificatory precision—and invites us to find meaning in that alien yet strangely familiar vision. ### [Perception Engines](https://medium.com/artists-and-machine-intelligence/perception-engines-8a46bc598d57) (2017) by Tom White By reverse engineering AI object classification systems, White's custom drawing system illustrates AI’s visual understanding of objects. As the system outputs learnt concepts, it expresses its learny understanding of the world. Ink drawings produced by the system from NSFW concept, for example, were found to trigger high NSFW scores in Google and Amazon's image recognition systems even when totally abstract, suggesting a shared visual language among AI models imperceptible to humans. ### [White Collar Crime Risk Zones](https://lav.io/projects/white-collar-crime-risk-zones/) (2017) by Clifton, Lavigne & Tseng *White Collar Crime Risk Zones* uses machine learning to predict where financial crimes are mostly likely to occur across the US. The project is a critique of American policing and imprisonment systems. ### [Cloud of Petals](https://sarahmeyohas.com/cloud-of-petals/) (2017) by Sarah Meyohas For *Cloud of Petals*, Sarah Meyohas gathered sixteen male workers to photograph 100,000 rose petals, one by one, at the site of the former Bell Labs in New Jersey, compiling a massive dataset used to map out an artificial intelligence algorithm that learns to generate new unique petals forever. ### [Alternative Face](https://www.quasimondo.com/2017/02/04/alternative-face/) (2017) by Mario Klingemann In the charged US political climate of 2017, Klingemann used Kellyanne Conway’s infamous “alternative facts” clip as the raw material for this pix2pix based work exploring the growing unease about truth in the “fake news” era. Extending pix2pix to image sequences, the piece foreshadowed "a future where fragmented realities and competing truths might deepen divisions". ### [Feminist Data Set](https://dilettantearmy.com/articles/making-critical-ethical-software) (2017) by Caroline Sinders *Feminist Data Set* is a critical research and art project that examines bias in machine learning through data collection, data training, neural networks, and new forms of user interface (UI), as well as the projected creation of a feminist artificial intelligence (AI). ### [Blade Runner - Autoencoded:](https://terencebroad.com/works/blade-runner-autoencoded) (2016) by Terence Broad Blade Runner—Autoencoded (2016) is a film produced by training an artificial neural network to ‘watch’ the film Blade Runner numerous times. Afterwards, the film was reconstructed through the memory of the neural network, giving an imperfect reconstruction of the original film, that both explored the perils of human-created intelligence and opened new paths for hacking it. Broad posted the video online, where it went viral and prompted a DMCA takedown notice from Warner Bros. ### [Terrapattern](https://www.flong.com/archive/projects/terrapattern/index.html) (2016) by Levin, Newbury, McDonald, Alvarado, Tiwari, and Zaheer. Terrapattern is a tool for exploring the unmapped and the unmappable: an open-source, open-access, and open-ended system for finding "more like this, please" in satellite photos. Operating as a revelatory artwork: a "panoptic perceptron" for open-ended play, and an absurdist tool, it leveraged neural nets to expose the rhetoric of 20th-century formalist analysis in urban planning and architecture. -------------- _Many thanks to all those who have contributed to the collection above, especially my students and colleagues at the School of Creative Media in Hong Kong and UAL's Creative Computing Institute in London._ 32:[] 2f:{"id":"eec7e61daffd348e","slug":"critical-ai-art-collection","title":"Critical AI Art Collection","thumbnail":"04965ebc80fd1c42d981b57ea447a6e63bdd03368d7771c24042ed950b322da5","access_level":"public","create_time":"$D2025-05-13T14:34:22.390Z","update_time":"$D2026-06-12T14:56:34.340Z","view_time":null,"first_public_time":"$D2025-05-13T14:45:49.792Z","fork":false,"stars":1,"starred":false,"owner":{"type":"team","id":"636cf123a07db681","login":"dhowe","name":"Daniel Howe","avatar_url":"https://avatars-next.observableusercontent.com/a5081aaa56ae9990aa27642040dc61de3ce8cea67f6f778a5b864e41111ba7a3","tier":"starter_2024"},"version":101,"publish_version":null,"role":null,"authored":false,"authors":[{"type":"individual","id":"32869872eb6f2f82","login":"dhowe","name":"Daniel Howe","avatar_url":"https://avatars-next.observableusercontent.com/a5081aaa56ae9990aa27642040dc61de3ce8cea67f6f778a5b864e41111ba7a3","tier":null}],"members":[{"type":"individual","id":"32869872eb6f2f82","login":"dhowe","name":"Daniel Howe","avatar_url":"https://avatars-next.observableusercontent.com/a5081aaa56ae9990aa27642040dc61de3ce8cea67f6f778a5b864e41111ba7a3","tier":null,"role":"editor"},{"type":"team","id":"636cf123a07db681","login":"dhowe","name":"Daniel Howe","avatar_url":"https://avatars-next.observableusercontent.com/a5081aaa56ae9990aa27642040dc61de3ce8cea67f6f778a5b864e41111ba7a3","tier":"starter_2024","role":"editor"}],"comments":[],"custom_thumbnail":true,"trashed":false,"subscribed":null,"forks":3,"collection":null,"type":"notebook","body":{"title":"Critical AI Art Collection","cells":[{"id":8,"value":"# Critical AI Art Collection","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":6,"value":"

From **A high-performance GPU as painted by Max Ernst**,\ncourtesy of Fabian Offert & Rita Raley

\n\n\nDescription: A provisional and non-authoritative collection of critical AI art since 2015
\n_{Keywords: critical ai, machine learning, artifical intelligence, critical practice}\n","pinned":false,"hidden":false,"mode":"html","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"},{"id":0,"value":"$31","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","output":"","database":"$undefined"}],"files":"$W32","stdlib":"1","titling":"auto"}} 30:"$2f:body"