The Infinite Talk
Installation with text generation neural nets – 2024
Every day for a month, an artificial intelligence model trained with the content of the CCCB’s archive compulsively writes a talk for eleven hours uninterrupted. The installation is activated every morning by a title suggested by a thinker, scientist or creator.
«The Infinite Talk» uses a text generation network trained with some 2,300 videos and audios from the CCCB’s archive of debates and talks. The installation uses this body of text to propose an experiment with AI and its uses. What value do we give it? How are we to understand it? What possibilities does it offer? What does AI do with everything it swallows up? The experiment sets out to highlight the strong points and limitations of statistical writing when exploring singular archives.
“Reflex Actions. Generative AI Tools and the Archive” article on the installation for CCCBLab. Cultural Research and Innovation
First foto: Alice Brazzit
Description of the installation at CCCB’s website:
An artistic installation by the Estampa collective that plays with the possibilities of artificial intelligence. Every day for a month, an artificial intelligence model trained with the content of the CCCB’s archive compulsively writes a talk for eleven hours uninterrupted.
You can follow it on the CCCB’s website and in the Pati de les Dones courtyard on a 32-metre screen that reproduces the text generated by the artificial intelligence model. The installation is activated every morning by a title suggested by a thinker, scientist or creator with the aim of provoking reflection on the issues that will concern us for the next 30 years.
«The Infinite Talk» uses a text generation network trained with some 2,300 videos and audios from the CCCB’s archive of debates and talks. Over the course of 30 years, the talks at the CCCB have addressed multiple issues with all manner of voices, coming to form an extensive cartography of the reflections, concerns and discoveries of contemporaneity.
The installation of the Estampa artistic collective uses this body of text to propose an experiment with AI and its uses. What value do we give it? How are we to understand it? What possibilities does it offer? What does AI do with everything it swallows up? The experiment sets out to highlight the strong points and limitations of statistical writing when exploring singular archives.
«The Infinite Talk» will be active for a month. Every day, AI writes compulsively and nonstop a talk that it generates, drawing on the themes and statements proposed by the 30 authors, trying to imitate the CCCB’s archive.
How was «The Infinite Talk» created?
To make «The Infinite Talk» work, the Estampa artistic collective has trained a text generation network with the transcription of talks from the CCCB archive. These networks are statistical imitation systems that, provided with a series of examples, look for probable solutions to generate other, similar ones. What a model generates always echoes the training dataset. Topics, themes, semantics, expressions… are all mixed by a combinatorial tool that consolidates the training data, but can also give rise to unexpected twists.
Unlike the big AI models currently being developed, such as Chat GPT, Estampa’s project is designed as a concrete, specific model—to imitate the Centre’s archive–and has been developed using high-performance hardware that is accessible at user level. This way of working responds to reflections on the need for a clearly outlined use of these tools, not only to reflect on what we want to do with them, but, first and foremost, for reasons of ecological expenditure. It is also the product of an interest in the results of smaller models, which, trained on small bodies of data, are more prone to unlikely combinations. Large models mimic the average and produce standard results; small models force combinatorics and, if we are lucky, new ideas appear.