Approches sensibles des archives du web was a course designed by Emmanuelle Bermès, Marina Hervieu, and myself.
As part of the SkyTaste project, we are organizing a transdisciplinary course involving students from PSL’s graduate arts program (history, art history, archives, design, fine arts) in a research-creation initiative using generative artificial intelligence. We aim to develop a sensitive approach to the skyblog archive, held at the BnF, in order to explore its various dimensions: emotions, images, texts, sounds (particularly music), colors, forms, performative qualities, themes, and patterns of navigation within the network. The collective nature of this project plays a key role in fostering a diversity of approaches, discoveries, and experiences when engaging with a massive corpus rich in personal and intimate content. Artificial intelligence will be used here as a generative tool (both consumer-grade and specialized tools) to interweave the concepts of archive and fiction, and to resonate them with various issues in today’s digital environment. Thus, the idea is to complement the sensory approach to archiving with a sensory approach to creation, involving the production of a set of hybrid documents that students will be free to design and exhibit as they see fit.
During this workshop, I tried to find ways to use AI to explore an archive in a similar way than the sensitive approach of Emmanuelle. Of course, the possibilities offered by the natural language processing are tremendous and would allow to index everything (at the cost of what it takes financially and ecologcally). My idea was born on the combination of three aspects:
- would it be possible to rely on the generative abilities of the model to generate on-demand skyblogs, to provide the same feeling but without legal & ethical issues?
- For the students, I wanted to make an interactive object that would allow to, at the same time, explore how AI behaves with non-usual data, but also to describe “negatively” the skyblogs : what is missing in the generation? And, for my interest, how we would feel in this experiment about the hanthologic perspective, really closed to the problem of archives (Mal d’archive, Spectres de Marx)? Will it be disgusting, touching, boring? Could we qualify the shift between real & fake ones (especially with the increasing panic of fake image generations)
- Artistically, I was intending to pursue my “synesthesia” interest of The Cloud : investing the poetic emergences that can arise from the neural methods of multi-modal learning. I designed a custom representation & dataset for the skyblogs and trained an LLM in order to be able to generate an article from an image, an image from an article, an image from a comment, etc. I wanted to push the farest possible the “interconnection” of all the data to, paradoxically, allow the model to generate freely.