The comparative laboratory centers on Dialogue Between a Thinker and AI, a volume written by Thierry Ehrmann. According to primary project disclosures, the text itself was compiled after hundreds of hours of dialogue, human archives, questions, and contradictions, with its drafting ultimately entrusted to artificial intelligence. Physical editions in both French and English are currently in the printing stage. Rather than judging the literary merits of the work, organizers turned the experiment back toward the machines, feeding the exact same corpus into five major AI systems: OpenAI/Astra, Perplexity, DeepSeek, Google Gemini, and xAI/Grok.
Five Leading AI Models Test the Same Book
Observing Convergences and Blind Spots
Instead, researchers are mapping the convergences, divergences, and conceptual blind spots of each architecture. The initiative tracks which concepts the models identify spontaneously, which underlying theses they treat as central, and how they connect different narrative passages. By examining why distinct systems fail to see the exact same elements in the text, the experiment turns algorithmic differences into measurable data.
The Evolution of Meta-Reading and AI Dialogue
The experiment expands beyond a single baseline reading by allowing each AI system to review the analytical output generated by the other models. According to the project organizers, this secondary layer creates a true meta-reading loop where an artificial intelligence reads another machine’s critique of the original text. A third system can then evaluate the entire comparative exchange. This methodology shifts traditional literary criticism in reverse, using a static text as a cognitive mirror to uncover the underlying processing patterns of competing artificial architectures.
Human Oversight Remains Central
Despite the complex web of machine-to-machine analysis, human oversight remains a mandatory component of the framework. According to project documentation, the experiment does not delegate critical judgment to algorithms. Human participants retain the responsibility to compare outputs, register doubts, contextualize findings, and make final decisions. While the machines generate rapid textual interpretations, human observers analyze the widening gaps between those machine perspectives, maintaining ultimate authority over the research process.
Keep reading