Salon with Linda Dounia Rebeiz
On 4 Decembre 2024 we discussed with Linda Dounia Rebeiz about AI, archiving practice and agency.
Linda Dounia is an artist, designer, and writer interested in the philosophical and environmental implications of technocapitalism. She is Senegalese Lebanese, and lives in Dakar. Her work mediates her memories as alternative realities and evidence of excluded ways of being and doing. It is formed through the dialogue and tensions between lived experience, code, and AI. In 2023, Linda was recognized on the TIME AI 100 list of most influential people in AI for her work on speculative archiving — building AI models that help us remember what is lost. In 2024, she was also the recipient of Mozilla’s RISE25 award for her work in AI, advocating for greater agency over algorithms – how we perceive them and are perceived by them.
Linda Dounia’s website: https://lindarebeiz.com/
Watch the video #
Look at the participation #
Notes #
- [notes from the presentation]
- Thinking about images (or artifacts) as data - deputies for reality, raw materials
- Data not only as a way to describe, but to also see
- Data and patches of invisibility
- “The forgotten continent that is at the start and the end of the AI value chain”
- Technology that does not know me - and carries harms with it
- Being unseen, or misrepresented
- Archives as ways to regain agency on data existence, act of remembrance
- Models reveals gaps (to fill?) -> inspire what to archive
- Synthesize the unseen?
- Evolution of synthesis technologies and styles. From GAN to SD
- Instruments of data capture
- “The parts that are broken ar the ones that are interesting to us”
- Access to resources to customize (retrain, finetune) models - is a barrier.
- About newer models black boxes, “I cannot see where my work begins and end” GANs give more agency
- Current model agency => $$$
Links #
- More about Linda Dounia’s work: https://lindarebeiz.com/
- Agbogbloshie in Ghana https://www.bbc.com/news/articles/c4gvq1rd0geo
Questions #
- [questions from the presentation]
- If data is a way of seeing and knowing, how do “capture them”, what are the instruments or the medium? Can data become the medium as well?
- Data and patches of invisibility - can you tell us more - is this bias, unknown unknowns?
- Is the archive creation an individual or collective process?
- Can archives be seen as acts of resistance, or influence training sets?
- Would you say that you are using AI as a “reflexive” instrument? A mirror of your process of creating archives? Or a mirror of collective creation of archives and memories?
- How do we navigate using historical data from colonial archives, such as images for example that have been adjusted to create a certain representation of history? For example work by Edward Curtis who would stage images of Native Americans in an attempt to preserve or capture “his perception” of their history, culture and traditions.
- What is the boundary between hallucinations, imagining unrealized possibilities, realism, and persisting misrepresentation?
- Would love to know your take on the role of AI as it is being used to preserve languages at risk of, or on the process of disappearing and how this might relate to your work.
- In addition to data - do you see ai-model’s architectures reflect colonialism, or other forms of structural bias. How much do you feel you need to know about the technology to recognize biases?
- Data is a way to represent and measure things/beings in the world, a model is a representation of these things/beings (always imperfect), how do you handle the simplification made by models in these kinds of projects?
- How do you document your decision making process around what an Image should look like? as you attempt to avoid creating mis-leading images/mis-representations through hyperrealism.
- More on the technical side, when you work on such a longitudinal project such as the project where you generate non-documented flowers, do you manage to keep the version from 2019, etc. ? Does it also become an archive of past technology?
- I think about efforts like OpenStreetMap related to the themes you discuss here - Are you aware of similar efforts when it comes to data and training datasets?