Searching For is a series of real-time simulations of infinite duration that follow Argus, a distributed, fictional satellite intelligence, as it perpetually attempts to find a collection of phantom islands: a grief process, where Argus tries to understand the fictions which might have poisoned the otherwise truthful dataset it had been trained on.

The first work in the series, Searching For (Bradley Land), follows the satellite toward an island Frederick Cook charted on his 1907–09 polar expedition and named for the man who financed the trip, casino owner John R. Bradley. As no later expeditions were able to find this island, it is believed Cook invented it as a way of funding his later pursuits.

Like all phantom islands, Bradley Land survives as language:

The lower coast resembled Heiberg Island, with mountains and high valleys. The upper coast I estimated as being about one thousand feet high, flat, and covered with a thin sheet [of] ice.
– Frederick Cook, My Attainment of the Pole, 1907–1909

Like all phantom islands, Bradley Land survives as a fiction in the training data that taught the satellite to see. With decommissioning imminent, the satellite is set on finding it. The satellite performs a latent dérive, a continuous stream of imagined islands, slowly evolving to match Cook's original description. As the search unfolds, brief dialogues between two nodes of the constellation give voice to the passage, approaching the island as if the only form of arrival were a search that never completes.

The work consists of a search process. A generative system produces rough sketches of island geometries. A diffusion model, fine-tuned on over 100,000 satellite images, renders each sketch as a satellite-like image of the resulting island. A vision-language model acts as judge, comparing each imagined island against Cook's textual description quoted above, and returning a score for their similarity. The fourth component, an optimization algorithm, adjusts the entire search process with the goal of raising the similarity score, such that successive islands converge toward the one Cook described.
The stream of islands shown in the work is the output of this process.

Searching For is part of End of Signal, a research-based body of work which examines the epistemic implications of AI deployed at scale, staged through a speculative future in which a satellite becomes an unreliable narrator of the reality on the ground. The project investigates what I term "technological underdetermination": a condition enabled by generative artificial intelligence, in which the unknown is produced by a surplus of plausible explanations rooted in existing ground truths, rather than by a scarcity of information.
A comprehensive presentation of the project can be found in this PDF document.

Production Credits
     Sofía Enríquez G., curator
     Laenz, sound design
     Andreea Iliescu, 3D environment design
     Yannis Siglidis, machine learning engineering
     MK Skitka, creative technology
     Sofía Enríquez G., print graphic design
     Katelyn Rebelo, digital graphic design
     Lindo Serviço, Estudios Durero, fabrication
     Fine Print, printing
     Elena Carbajal, exhibition coordination & production
     María Castillo, Alessandra Gargiulo, exhibition design & production
     Milena Fernández, communications
     Special thanks: Alasdair Milne, OFFICE IMPART (Anne Schwanz and Johanna Neuschäffer), Artemis Gallery, Aisling Murray

End of Signal is the winning project of the SOLO AI ’25 AWARD, presented by SOLO Contemporary, and part of the exhibition programme of Onkaos. I would like to acknowledge the support of SOLO Contemporary and STUDIOTOPIA in the development of this project.