Ghost Specimen

During our research period at Cambridge University we were given access to the Herbarium collections, including specimens collected by Charles Darwin and other explorers, many of which have not been seen publicly. Encountering these archived plants, some now extinct, stored in yellowed cardboard boxes and wrapped in newspapers from countries some of which have since disappeared, been renamed, or reclaimed, revealed not only botanical histories, but also the infrastructures of power embedded within systems of taxonomy, classification, and scientific observation.

Archival practices reveal how botanical knowledge is shaped by utilitarian forms of categorisation often rooted in colonial science that have historically transformed landscapes into legible, extractable resources, establishing authority over what counts as knowledge and how the natural world can be used. In this context, plants are treated as detached from indigenous ecological systems and transformed into portable scientific data.

The work begins with a pressed herbarium specimen of Hibiscadelphus wilderianus, the Maui hau kuahiwi, a flowering plant once native to Maui, Hawaii, and now extinct. First documented in 1911 by botanist Gerrit Wilder in the lava fields of Maui, the species disappeared as the surrounding landscape was transformed into cattle ranches. Grazing animals stripped the bark from its trunk, invasive rats consumed its seeds, and by the 1930s the last known wild tree had died. It was the only member of its species ever found.

Preserved within the Harvard Herbarium archive, the flower survives as a flattened scientific trace: an object shaped by processes of collection, classification, and preservation. Using AI image interpretation and 3D modelling, the artists translated this pressed specimen into a sculptural form, subsequently 3D printed in metal. In doing so, the extinct plant is transformed into an artefact suspended between biological record, machine hallucination, and memorial object.

The sculpture is installed with the flower turned back-to-front against the wall, exposing the side that was absent from the original dataset. Faced with incomplete information, the AI system was forced to speculate, generating a synthetic reverse based not on knowledge of the plant itself, but on statistical assumptions derived from training data.

The sculpture’s strangely flattened surface and unresolved textures further expose the limitations of machine vision and the instability of AI-generated representation. Its ‘face’ turned away is suggestive of both the loss of the flower’s original identity and the inability of computational systems to truly encounter the complexity and interconnectedness of the world. What emerges is not a reconstruction, but a hallucinated approximation – a form produced through prediction, interpolation, and bias.

Machine learning systems do not perceive environments directly, they interpret the world through datasets shaped by human decisions, historical inequalities, and inherited structures of visibility. In this sense, AI reproduces older ways of seeing embedded within colonial archives and scientific taxonomy: both transform complex ecosystems into isolated atomised units of information that can be organised, classified, and operationalised. Plants become detached from ecological relationships and reconstituted as portable data, in turn shaping collective perceptions of what nature is, how it should appear, and which forms of life become legible or valued by replacing material knowledge with predictive models and probabilistic representations.

By transforming an extinct pressed flower into an uncanny metallic object, Ghost Specimen probes the relationship between archival preservation, machine learning, and environmental abstraction. The sculpture asks what happens when nature is encountered primarily through systems trained to categorise, predict, and generate images of the world and how these technologies may reinforce forms of distance, bias, and disconnection from the living environments they claim to represent.