An agent archaeology experiment illustrates a literature-search failure
Pietro Miotti says 48 agents inspected images derived from public lidar data and flagged a Maya site absent from a government register. His corrected account now says archaeologists had already mapped it in 2025; the agents missed that literature. Finding a promising structure and establishing that it is a new discovery turned out to be separate tasks.
Pietro Miotti
Why it matters
Parallel analysis can find candidates quickly while a weak novelty check still produces a misleading conclusion.
More context
The author describes reviewing roughly 1,200 square kilometres and then contacting Mexico’s INAH. This card covers his November 11 correction; the original post date is unverified. The useful workflow lesson is to independently review prior work before presenting an agent-generated candidate as a discovery.
Keep in mind
A personal account by a non-archaeologist, not an audited benchmark or a verified new archaeological discovery. Interpretation of the site’s historical purpose remains speculative.
Original read