Blinkylights Blog

, ,

AI Agents Can Find the Lost City. Somebody Still Has to Google It First.

A man with no archaeology background pointed 48 AI agents at NASA lidar of the Maya jungle and got back a hilltop site with a plaza and a pyramid. Then he learned it was already on the map. Search is cheap now. The part that’s still ours is checking whether anyone got there first.

Here’s the story as Pietro Miotti tells it on his own site. He asked Gemini, ChatGPT, Claude and Grok how you’d find hidden ruins, and they all pointed him to lidar, which fires laser pulses through gaps in the canopy to map the ground underneath. He pulled about 1,200 square kilometers of public NASA lidar over the Maya region and had 48 agents look at the images for pyramids, platforms and plazas, while rejecting natural hills, sinkholes and roads. No special model, no classifier. Agents looking at pictures.

They flagged a hilltop in Mexico: a plaza, an oval pyramid at least 8 meters tall, a raised courtyard and a possible water reservoir. The agents guessed a Classic-period civic center, roughly AD 250 to 900. That’s their guess, not a finding. Miotti checked the point clouds, wrote a report, translated it into Spanish and sent it to INAH, Mexico’s national anthropology institute. The reply came the next day: no registered site at that location.

On November 11 he added an update. The site had already been mapped in a 2025 regional lidar study by Estrada-Belli and colleagues. His agents had searched the literature and missed it. So it was a re-discovery, and the headline on his post says so.

An oval pyramid at least 8 meters tall is a hard thing to hide. That should have been the first clue that somebody had already met it.

There’s a long tradition here. Hiram Bingham gets the credit for Machu Picchu in 1911, but a local named Melchor Arteaga led him up the mountain, people lived in the neighboring valley, and Agustín Lizárraga had scratched his name on a temple wall nine years earlier. Miotti’s agents just did their rediscovering in about three days.

The part that’s actually new isn’t the rediscovery. It’s that a non-expert ran the whole pipeline in 72 hours. Professionals use the same trick, but slowly. A 2024 paper in Antiquity went back to lidar collected in 2013 for forest monitoring, never meant for archaeology, covering about 122 square kilometers of Campeche. The team found Valeriana, a previously unknown site with two monumental hubs about two kilometers apart, and counted 6,764 structures across the survey. Every feature was digitized by hand and checked by at least three analysts. The data was collected in 2013 and reanalyzed for archaeology in 2024. Agents turn “nobody had time” into “somebody had a free weekend.”

Agents turn “nobody had time” into “somebody had a free weekend.”

That density is the catch. The same survey works out to about 55 structures per square kilometer. In a landscape that crowded, “new” is a long shot. The jungle is full, and your pyramid has neighbors.

How to not rediscover something

Here’s the due-diligence checklist I’d run before emailing anyone. Twenty minutes, no machete.

  1. Start with open data. NASA’s G-LiHT, Goddard’s airborne lidar, hyperspectral and thermal imager, lets you browse and download its data through a web map. Forest-monitoring lidar is another source; the Antiquity team reused a 2013 Mexican survey that was made freely available.
  2. Give the agents a job description with a reject pile. Ask for pyramids, platforms and plazas, and tell them to throw out natural hills, sinkholes and roads. That’s the filter Miotti used.
  3. Check the raw point cloud, not just the picture. Miotti sent INAH point-cloud checks showing the features were on the ground. A blob that doesn’t survive the raw data was never a pyramid.
  4. Do the literature check yourself. Search the region plus “lidar” in Google Scholar and read the last two or three years of papers with the maps open. An agent will tell you there’s nothing there with the confidence of a old man from Amarillo giving directions to Palo Duro Canyon. Miotti’s did.
  5. Then report it to the people who own the question, and leave the shovel at home. Write a short report in the local language and send it to the national authority, as Miotti did with INAH. Notice that INAH had no registered site at that location while a published paper had already mapped it. Those can both be true.

Miotti argues that lidar-mapping the whole planet is the next step, and he’s probably right. The first thing it will produce is a flood of rediscoveries. Agents will keep finding things. The job that’s left is knowing whether it’s new.

Sources

← previous

cd ~ — back to all posts