Guest Post: After the Dataset — An Introduction by Gazelli Art House

After the Dataset  examines how artificial intelligence mediates cultural memory, imagination, and visual knowledge. Bringing together artists working across machine learning, generative systems, and simulation, the exhibition considers AI not as a neutral tool but as a structure embedded within historical, political, and aesthetic frameworks.

Across the artworks archival material, mythological narratives, and ecological systems are reconfigured through algorithmic processes, revealing how datasets carry inherited assumptions while shaping new visual imaginaries.

Presented by Gazelli Art House, the exhibition brings together artists who have all participated in the gallery’s digital residency GAZELL.iO, reflecting its ongoing commitment to supporting critical practices at the intersection of art and technology.

Memo Akten & Katie Hofstadter

Memo Akten & Katie Hofstadter, Superradiance, stills, 2024. Courtesy of the Artists and Gazelli Art House

All Watched Over by Machines of Loving Grace (2021) by Memo Akten takes its title from Richard Brautigan’s 1967 poem, reworking its utopian vision through the lens of contemporary AI. Developed using custom machine-learning systems prior to the emergence of consumer image-generation tools, the work reflects on the cultural tendency to both mythologise and submit to technological systems. It proposes instead that the distinction between “natural” and “artificial” is illusory, suggesting that attempts to control one are inseparable from attempts to control the other.

This enquiry extends into Superradiance (2024), a large-scale video and sound installation developed with Katie Hofstadter. Across its chapters, Embodied Simulation and Embodying Earth, the work constructs dynamic, immersive environments in which forms emerge from interacting systems rather than fixed compositions. Drawing on ecological, physical and computational processes, Superradiance models intelligence as distributed and relational, situating perception and consciousness within a broader continuum of human, machine, and planetary systems.

Nouf Aljowaysir

Nouf Aljowaysir, Salaf: Ancestral Seeds, still, 2025. Courtesy of the Artist and Gazelli Art House

Salaf: Ancestral Seeds (2025) is a series of AI-generated images and videos trained on the Salaf dataset, compiled from colonial photographic archives of the Middle East. Constructed through a Western gaze that staged and codified the “Orient,” the dataset is systematically altered, with subjects and the visual codes used to define them deliberately erased.

From this process emerge recurring white figures, described as “seeds.” These Middle Eastern silhouettes echo the subjects of colonial photography, yet are stripped of their expected stereotypes. Generated through omission rather than addition, these forms are not neutral abstractions but residual structures shaped by what has been removed.

By foregrounding absence, the work opens a discursive space in which memory, erasure, and representation can be reconsidered as active sites of resistance. Yet even within these reduced forms, traces of orientalist visual logic persist. This persistence reveals how AI systems do not simply reproduce images, but inherit and extend the ideological frameworks embedded within their training data, reanimating the colonial gaze even through acts of erasure.

Morehshin Allahyari

© Morehshin Allahyari, Moon-Faced, What Models Make Worlds: Critical Imaginaries of AI at Ford Foundation Center for Social Justice, 2023. Courtesy of the Artist and Gazelli Art House

In ماه طلعت (Moon-Faced) (2021), Morehshin Allahyari reactivates Persian Qajar-era portraiture through AI, reconstructing a visual tradition in which beauty was historically understood as genderless. In classical Persian literature, ماه طلعت (“moon-faced”) functioned as an adjective applied to both men and women; in contemporary usage, it has become exclusively feminised, reflecting a broader shift in cultural and visual representation.

A parallel transformation occurred within Qajar painting (1786–1925), where earlier modes of gender ambiguity were gradually displaced through processes of modernisation, the influence of European realist aesthetics, and the adoption of photography. These developments contributed to the erosion of a visual language in which gender was fluid and undifferentiated.

Working with a multimodal AI model trained on archival Qajar paintings, Allahyari employs carefully constructed prompts to generate a series of video portraits that reintroduce this ambiguity. Rather than reproducing the archive, the work intervenes in it, using machine learning to reconfigure inherited visual codes and to propose a counter-history that restores non-binary representation within Persian visual culture. The score is composed by Mani Nilchiani.

Brendan Dawes

Brendan Dawes, Persian Dreams series, stills, 2023. Courtesy of the Artist and Gazelli Art House

The Persian Dreams series (2023) by Brendan Dawes brings together generative AI, motion-captured choreography, and algorithmic image-making to reinterpret narratives drawn from the Shahnameh, a 10th-century Persian epic poem recounting mythological, historical, and heroic cycles. Presented as a sequence of short, time-based works, Creation, Dynasties, Heroes, and Monument, the series translates these narratives into continuously shifting visual forms.

Across the works, figures do not appear as fixed representations but emerge through processes of transformation, oscillating between abstraction and figuration. Sculptural forms are generated, dissolved, and reconfigured in motion, shaped by the interplay between choreography and machine learning systems.

Rather than illustrating the source material, Persian Dreams rearticulates it through computational processes, establishing a temporal continuum in which ancient narratives are neither preserved nor replicated, but continually re-formed, suggesting how cultural memory persists through adaptation across technological contexts. With visuals choreographed to the movements of Charlotte Edmonds and scored by artist duo Madota, whose experimental soundscapes draw on elements of zoorkhaneh rituals, the works bring the Shahnameh’s verses and imagery into oscillation between figuration and abstraction.

Jake Elwes

Jake Elwes, Zizi & Me – Anything You Can Do (I Can Do Better), still, 2020. Courtesy of the Artist and Gazelli Art House

Terms & Conditions Opera: A Legalese Libretto (2024–2025) transforms the terms of service and usage policies of AI platforms into an audiovisual composition. By feeding systems their own regulatory frameworks, the work produces a recursive loop in which legal language is reinterpreted as sound and performance.

Rather than focusing on generated imagery, the work foregrounds the infrastructures that govern AI, policies, permissions, and labour, making visible the conditions under which these systems operate.

This approach builds on Elwes’ earlier work Zizi & Me – Anything You Can Do (I Can Do Better) (2020), part of The Zizi Project, in which machine learning is used to generate synthetic drag performances. Developed in collaboration with LGBTQ+ performers, the project reclaims AI systems by queering their outputs and challenging the biases embedded within training datasets. Together, these works shift attention from what AI produces to how and by whom these systems are constructed, used, and contested.

Matteo Zamagni

Matteo Zamagni, Horror Vacui, still, 2018. Courtesy of the Artist and Gazelli Art House

Horror Vacui (2018) by Matteo Zamagni layers aerial imagery, macro photography, and 3D scanning to construct composite landscapes that oscillate between natural and artificial forms. The work draws on the concept of “fear of empty space,” linking it to contemporary conditions of overdevelopment and environmental strain.

Through its dense visual field, the film reflects on the disconnection between human activity and ecological balance, situating technological production within broader planetary systems.

Conclude

Across these moving-image works, After the Dataset positions AI as a site where historical knowledge, cultural bias, and speculative futures converge. Working through time-based, generative processes, the exhibition shifts attention from discrete outputs to the conditions that produce them, foregrounding how images emerge, unfold, and are continuously shaped.

Explore the exhibition here: virtual.hek.ch.

Want to curate our online exhibition space or propose a project? Join Friends of HEK and take part in future open calls.