A critical and aesthetic case study on the capacities of AI models to reconstruct missing fragments of historical documents, incl. those of classical, art and natural histories. Through generated quasi-historical objects carved in stone, this series questions the epistemological qualities of synthetic historiography, akin to AI driven reverse archaeology.

A critical and aesthetic case study on the capacities of AI models to reconstruct missing fragments of historical documents, incl. those of classical, art and natural histories. Through generated quasi-historical objects carved in stone, this series questions the epistemological qualities of synthetic historiography, akin to AI driven reverse archaeology.

2023. ckpt-format file of a collapsed Stable Diffusion Ai model, film-essay.
Images of chairs were generated by a Stable Diffusion deep learning model from the prompt a single chair on a white background. The dataset was reused to retrain the model, as if to enhance its understanding of the given prompt. After six iterations, data cannibalism led the figurative chair image to become abstract digital noise, exposing susceptibility to corruption of computational ontology due to data feedback loops within Ai model echo chambers.

A text-to-image AI model was given the prompt “A single chair on a white background.” After six iterations of retraining on its own outputs, the autocannibalised model lost ability to produce realistic chair images, but non-figurative digital noise instead. Thus revealing model collapse and fragile computational ontology corrupted by recursive data feedback loops in AI architectures.

Fact Caching protects investigative journalism. Censored archives are safeguarded against blocking by being hosted as tiny data bits across thousands of nodes on decentralised networks.

Hashd0x is an artist initiated dApp addressing disinformation via on-chain evidence. The work Proof of War uses it to 3D-scan a bombed house in Borodyanka, storing drone images on-chain as an uncensored record of war crimes.

Installation Chinese ink comprises real-time GAN-generated imagery displayed on e-ink screens. Trained on inkblots, the system produces hundreds of variations per minute via a liquid-cooled GPU infused with ink. Unlike light-emitting displays, e-ink mimics pigment on paper. The work reflects on whether ink wash traditions can endure industrial transformation and whether computational simulation can reproduce the depth and subtlety of ink on rice paper.

A critical and aesthetic case study on the capacities of AI models to reconstruct missing fragments of historical documents, incl. those of classical, art and natural histories. Through generated quasi-historical objects carved in stone, this series questions the epistemological qualities of synthetic historiography, akin to AI driven reverse archaeology.

Kickback is a series of interventions in global fast-fashion retail stores in which modified goods are returned for resale to offer critical commentary on current consumption–production ethics.

An on­line mystification concerning a parascientific break­through discovery of a radically new colour. The digital myth took the form of a fictitious company's website, video adverts & mockumentary interviews to become a viral sensation attracting mass attention online.