From the desire to position multivocality as the guiding principle of its activities, Het Nieuwe Instituut opts for an exhibition form that combines a multitude of entrances and perspectives with a diversity of forms of representation. Temporary House of Home manifests itself as a spatial collage in which the idea of a single theme is questioned from a variety of psychological, technological, economic and social perspectives.

Commissioned by Rotterdam's New Institute for Architecture, Design, and Digital Culture, Tivon Rice trained five artificial intelligences (A.I.s) to read images of interiors from the archive of Het Nieuwe Instituut. The A.I.s learned to write by analyzing works by authors – from anthropology, philosophy, economics, literary fiction and children's literature – in order to make a speculative comparison between different forms of language surrounding our private spaces. Each voice suggests a different perception of what is meaningful about an interior.

The voices of the child, the home, the space, the worker, and the other examine how COVID has changed our relationship with our home. How has our domestic experience been affected as private spaces have also become a place for education, work, and our virtual social life? How does language function in this digital guise, when it communicates between virtually linked spaces? What opportunities are there to work creatively through these channels?

Developed before large language models like Chat-GPT were widely available, this project explored natural language processing (NLP) through different voices by fine-tuning the GPT2 model on the following datasets:

The Child - Roald Dahl: complete works 
The Home - Georges Perec: Life: a User's Manual
The Space - Marc Auge, Michel deCerteau, Michel Foucault: various texts
The Worker - Karl Marx: Kapital I, II, III
The Other - Jakob von Uexküll, Donna Haraway: various texts

These narratives are contingent upon the underlying GPT model, the fine-tuning dataset, and the prompt: the question, the instruction, the provocation given to the system. But in this project, the prompt is also generated by an algorithm, an image captioner – DenseCap – which indicates the content and context of elements within the image frame. Such computer vision systems are clearly fraught with errors and biases in the underlying training sets. Thus, an additional layer of mediation is introduced into the system, amplifying the temporal and cultural disconnects between these images from the Dutch archive and our contemporary, post-digital notions of place.