A Google researcher says a group of 48 AI agents has flagged what may be an unmapped Maya site in Mexico, and the country's national archaeology institute is now looking into it.
Pietro Miotti described the project as a "random thought on vacation": could AI agents sift LiDAR data for signs of buried structures? LiDAR (Light Detection and Ranging) bounces lasers off the ground to build topographic maps, and it can pick out shapes under forest canopy that the eye misses. Miotti pointed 48 agents, drawn from different AI apps, at a NASA LiDAR dataset. They returned multiple candidate structures. A second pass narrowed those to one site that, according to Miotti, has not been mentioned in scientific literature.
The agents also handled the follow-up. They told him to contact Mexico's National Institute of Anthropology and History (INAH), then drafted a Spanish-language email with the find and its map location. INAH replied, and its experts are investigating. In his November 10 post, Miotti counted 72 hours from idea to reply. He also had Gemini generate an image of how the site might have looked in 700 CE.
The site has not been confirmed as Maya, and the account comes from Miotti himself. The 72 hours also measures how fast an institute answered an email, not how long verification takes. That still lies ahead, with INAH's experts.
The more interesting part is the division of labor. Spotting candidate shapes in a huge dataset is tedious for people and well suited to software, while deciding whether a shape is a Maya building falls to specialists. Miotti's call to "LiDAR the entire planet" assumes the first step will yield real finds more often than false leads. One unconfirmed site cannot show that, but INAH's verdict will be an early data point.