Now at the SCHIRN: The World Through AI

The tour begins in Hall 1 with Timo Arnall’s “Internet machine” (2014) and draws attention to the massive data centers where AI technologies are housed. In his video installation, the artist reveals the reality of digital infrastructure, which—contrary to what the rhetoric of the intangible “cloud” might suggest—requires vast physical structures and enormous amounts of electricity and water to run.

The material and functional structures underlying AI systems are explored by several artists. In his series “Metamorphism” (2016), Julian Charrière fuses computer components such as motherboards and processors (CPUs) with soil to create sculptural mineral formations, raising questions about the toxicity of digital waste. The video installation “Computational Compost” (2023) by Marina Otero Verzier examines the ecological impact and limitations of technological data storage and other forms of collective memory. Agnieszka Kurant addresses alternative, nonhuman forms of collective intelligence in works such as “A.A.I. (System’s Negative) No. 6” (2016), a negative cast in zinc of an abandoned termite mound.

A further section is dedicated to machine vision and the new culture of a nonhuman gaze determined by algorithms. The SCHIRN presents, among other works, “Eye / Machine I–III” (2001–2003) by Harun Farocki, which unfolds across three separate video installations. At the beginning of the twenty-first century, the filmmaker and theorist was already engaging with technologies of automated detection, identification, and classification of objects, places, bodies, and faces, used in areas as varied as warfare, industrial production, traffic management, and robotics.

Among the works dedicated to face and emotion recognition is Trevor Paglen’s video work “Behold These Glorious Times!” (2017), which places images of objects, faces and their features, and emotions drawn from large training datasets alongside their machine analyses. The interactive installation “Faces of ImageNet” (2022) provides a live, first-hand demonstration for viewers of how a system based on potentially discriminatory datasets recognizes and categorizes them.

The exhibition also explores the algorithmic reconfiguration of the relationship between words and images, between the verbal and the visible. Generative AI models produce complex linguistic descriptions of images or, conversely, translate written instructions (or prompts) into visualizations. The collective Taller Estampa uses translation via object and face analysis programs as well as image-to-text models for its video installations, in which paintings and film segments are rendered as language. “What Do You See, YOLO9000?” (2019) and “Ekphrasis” (2025) reveal the limitations of these narrowly trained systems as well as the poetic potential that arises from the attempt to translate images into words.

Kate Crawford and Vladan Joler’s large-format diagram “Calculating Empires: A Genealogy of Technology and Power Since 1500” (2023) charts the deep genealogies and intertwined developments that have led to the current state of AI, making visible a history spanning more than five centuries in which technology and automation have been used to gain power and influence over the body, the mind, societies, time, and space.

The following sections explore the largely unseen human labor behind seemingly autonomous AI systems. Hito Steyerl’s work “Mechanical Kurds” (2025) references an eighteenth century chess-playing automaton constructed by Wolfgang von Kempelen, which was actually operated by a hidden chess master, as well as Amazon Mechanical Turk, an online platform for outsourced microtasks. The video installation documents the work of Kurdish “clickworkers” living in refugee camps, where they train unmanned drones that may eventually be deployed against the same Kurdish population.

A further section is dedicated to hallucinations and psyops. Generative AI is said to “hallucinate” when it produces outputs that do not align with the intentions of the prompts or instructions given to them. Such malfunctions may be random, statistically possible outputs, but they can also be deliberately engineered, serving as a means of psychological manipulation. Using two works from the series “Adversarially Evolved Hallucinations” (2017), Trevor Paglen demonstrates how generative AI models known as GANs (Generative Adversarial Networks) can be actively induced to hallucinate by training them on complex and heterogeneous images depicting allegories, metaphors, and symbols, described by keywords drawn from literature, philosophy, poetry, folklore, and spiritual traditions. With the video series “Cyclops” (2023), Paglen situates the current development of AI within a longer tradition originating in the psychological warfare and mind-control experiments of the Cold War.

Another room presents the interactive work “xhairymutantx” (2024–25) by Holly Herndon and Mat Dryhurst. Countering technology whose development has been shaped by corporate interests, the work proposes an alternative conception of AI as a medium for artistic imagination and collective creation. Taking as its starting point the question of how our image is encoded by and embedded within AI models, and how we can take back control over our representations, the artists developed their own public text-to-image model, complete with a prompt interface, that continually generates variations of a red-haired, mutated figure who resembles Holly Herndon. The work forms part of a broader, ongoing body of work in which the artists aim to envision and actualize an AI that is open, transparent, and collective.

Another of the exhibition’s commissioned works relates to its venue—the former Dondorf printing factory—and to the theme of AI and history, of material traces and latent archives. Ania Szczepańska’s film “Reinscribing the Dondorf” (2026) explores the history and memory of the Jewish Dondorf family, the founders of the Dondorf printing factory, while also reflecting on the implications of using AI in the context of historical and archival research.

In the next section, Jacques Perconte’s video installation “Better Mont Blanc” (2024) addresses the visual effects and theoretical questions raised by using generative AI models to compress or upscale digital images. While reducing an image’s resolution results in the loss of information, upscaling algorithms do not restore missing pixels but instead create new ones that are statistically plausible based on learned patterns. “Better Mont Blanc” uses original footage of places such as Mont Blanc or the Port of Rotterdam—locations undergoing significant changes due to climate change or the growing trade of goods—and highlights the differences between the richness of sensory experiences and the statistical approach of AI-generated images.

In the space next, Grégory Chatonsky explores a new kind of memory produced by generative AI models, which assemble potential alternative content from the vast amounts of human memories that have been stored digitally, thereby creating multiple iterations of a narrative. The multimedia installation “The Fourth Memory” (2025) takes the form of a grave on an uninhabited earth, where various versions of the artist are brought back to life by a machine, continually being resurrected as if in a dream.

In the final room, the exhibition brings together video works dedicated to visions and ruins, hopes and fears emerging from our contemporary AI culture. Inès Sieulle’s “The Oasis I Deserve” (2024) explores the relationships between human users and Replika, the controversial dialogue-based chatbot released in 2017, which oscillates between being a confidant, an oracle, and an object of desire. Gwenola Wagon’s “Chronicles of the Dark Sun” (2023), composed of images generated or transformed by the model Gen-2, is based on a text by Pierre Cassou-Noguès and inspired by Chris Marker’s “La Jetée” (1962). It envisions a post-apocalyptic future in which humans attempt to reconstruct a lost world with the help of AI.