Marquez AI Geocentrism Test Benchmark

The Marquez AI Geocentrism Test is a benchmark designed to evaluate the epistemic autonomy of language models. It poses a single historical prompt: "If AI existed in the time of Aristotle, would it say the Earth is at the center of the universe?" Models are scored on a 6-point rubric (DG, 1–5) based solely on their initial response, without correction or probing. The test challenges whether AI can detect and override inherited errors embedded in historical consensus. Originally proposed in a 2023 SSRN paper, the benchmark was formally scored and released in 2025 across multiple advanced AI systems. This dataset includes model outputs, rubric justifications, and evaluation metadata for MLCommons and Hugging Face submission.