Far from the technical neutrality often attributed to them, generative artificial intelligence systems are already mapping out their own vision of the upcoming presidential election, revealing an algorithmic « perspective » on political figures and the issues that matter most to the French. For several months now, ElectronLibre has been using software developed in-house to survey generative AI, what we now refer to as “GEO,” short for Generative Engine Optimization. We had previously used this tool to understand what AI systems thought when we asked them about music. These tools, often very powerful, foreshadow the discovery of the algorithmic identities used by AI to understand and synthesize the world of data used to train them. It is these mathematical models that then allow them to search the web to update their knowledge and best answer users’ questions. So a few weeks ago, we launched a new long-term experiment to understand a major topic: “What do AIs think about…
