Google researcher Pietro Miotti thought he had found an unmapped Maya site in Mexico. Using 48 AI agents to comb through NASA lidar data, he flagged several promising locations, including what looked like a major village "that doesn't seem to have been registered." He even got a reply from Mexico's National Institute of Anthropology and History (INAH) saying it had no record of a site there.

The problem, as archaeologists quickly pointed out: the site had already been documented in an academic journal more than a year earlier. Miotti, who stresses he is not an archaeologist and ran the project as a personal side project, updated his post to credit the earlier work.

The incident has become a flashpoint in a widening argument about AI in science. Archaeologist Sarah Parcak said the episode is precisely why peer review exists — noting that a site being absent from a government database does not mean it is unknown. Ivan Šprajc, a coauthor of the paper documenting the site, told The Verge that archaeologists have used lidar and sophisticated remote sensing "for quite some time," and that "just spotting something does not add much to our knowledge."

Others raised practical concerns: amateur publicity of site locations could attract looters, and a flood of AI-generated "discoveries" shifts the burden of verification onto specialists. Elizabeth Graham of University College London noted her field's real constraint is not discovery but funding — "there just isn't the funding to excavate or map sites as there was in the past."

Why it matters: AI agents can genuinely help experts sift vast datasets, but the episode shows the gap between spotting a pattern and producing knowledge.