To Remember, To Forget (2024)

“To Remember, to Forget” is a multi-modal self portrait blending the material process of creating casts
from a single mould with the digital process of creating two AI characters based on the same corpus
of data. Byung-Chul Han (referencing Freud) considers the process of remembering, and forgetting,
deeply human experiences and ways we make sense of the world. He contrasts this to the flattened
time horizon of a big-data modelling process, where all reference points become collapsed into a
“collective immanence”, and describes this as a perspective belonging to the undead.

Given a lot of my work with AI has explored how it may work as a store of memories, as an archive of
the not-yet (Hui), I felt challenged by Han’s provocation. In my experience, the hallucinations found in
machine learning “recall” (or inference, to use the ML term), are strongly reminiscent of the manner in
which memory is imperfect and a process of fictioning. This work explores how this may be so.

I have collated recent writings into a corpus of text, and two chatbots have been fine-tuned on this.
The characters are differentiated by their “system prompt” – precursor text that positions them in a
space / time / situation, “Pastlife” is from Australia in the 90s, “Nathan” is situated in contemporary

Their different responses can be considered different masks made from the same mould, with similar
shapes but infinite differences in the details. The emergent conversation constitutes a recursive mode
of self-reflection and a hallucinatory, speculative self-portrait. This work considers the “corpus”, the
data body, from multiple perspectives and in multiple materialities.


Goldsmiths University, London, May 2024