Pietro Miotti says a group of AI agents flagged a hilltop in Mexico that appeared to contain Maya structures. On November 11, he updated his first-person account, saying archaeologists had already mapped the site in a 2025 regional lidar study. Miotti reports that Mexico’s National Institute of Anthropology and History found no entry for that location in its register when it responded. That response concerned the register. The earlier study had a separate record.
Miotti says 48 agents examined terrain images derived from about 1,200 square kilometers of public lidar strips, looking for shapes such as plazas and platforms. His October diary describes roughly 72 hours from the initial idea to the institute’s response. The correction came later, after he identified the earlier research.
Test detection and novelty independently
A terrain reviewer can ask whether a shape in an image is worth checking in the underlying point cloud. A records reviewer can check whether a public authority has a matching entry. A literature reviewer can compare the candidate with earlier surveys, including papers whose coverage is described in maps or supplementary material. Success at one stage does not automatically translate to success at the next.
A research team could measure the screening method with a blind set of tiles containing known structures, similar natural shapes, and empty terrain. Report the share of known structures flagged and the number of false leads per 100 tiles. Keep a separate count of how often the literature check correctly recognizes a previously mapped location. That would show whether the agents are useful for finding candidates and how much independent checking their novelty claims still need.
Abdessalam Alaoui
I'm Abdessalam Alaoui, founder of Musthave.AI. I write practical guides to AI models, tools, generative media.
