artificielle ou humaine?
la mémoire collective au travers des prompts

frédéric clavert | frederic.clavert@uni.lu

24/04/2025

introduction

qui suis-je?

generative ai

AI systems trained to create new content (text, images, etc.) rather than classify or predict. They learn patterns from vast datasets and generate outputs that resemble the data (stochastic parrots) they were trained on.

text / image / video / multimodal

prompts?

prompts are text descriptions or questions that generative AI systems require will to produce content

Why Should We Care?

  • Widely used to obtain information, write content, etc.
  • Includes the production of ‘original’ content on history, hence mediating the past
  • mediation with the past with no sense of truth but many biases
  • might in the future shape popular (political) understanding of history

Vectors of Collective Memory

  • Training data are embedding collective narratives about the past
  • Training data carries biases, gaps, and silences

→ the vectorization of memory means AI outputs can reinforce certain recollections and erase others.

our cultural heritage is “morally and politically deeply flawed,” so an AI drawing on it might replicate those flaws (Kansteiner)

Prompts as a Negotiation with the Past

  • Prompts as questions about the past
    • prompts are then doors into collective memory
  • Prompts as User–Machine interaction / negociation
    • final content as a cyborg content
    • tensions can appear in this human/machine interaction

Examples

Europe

Joe Biden

Hello history

Conclusion