Lidar‑derived 3D view reveals a straight‑edged plaza and an 8–12 m oval pyramid on a cliff‑side hilltop, the features flagged by AI agents.
AI AGENTS FLAG UNREGISTERED MAYA SITE IN MEXICO AFTER 72-HOUR SPRINT
*In less than three days, a suite of AI tools turned public NASA lidar into a claim that Mexico's heritage agency had never recorded. The find sparked a rapid exchange with INAH and raised questions about how fast digital archaeology can move.*
On 18 October, while traveling in Peru, the author imagined what Hiram Bingham might have achieved with modern AI tools. The thought sparked a midnight command to a quartet of language models to hunt for lidar data that could pierce the dense Maya jungle canopy. By the next morning, the agents had processed 1,200 km² of NASA‑collected point clouds, turning raw laser returns into clear images of the forest floor. Within 48 hours they identified a hilltop settlement with straight‑edged architecture—a plaza, an 8–12 m oval pyramid and a possible rainwater reservoir—none of which appeared in Mexico’s official archaeological register. The author compiled a bilingual report, sent it to INAH, and received a confirming reply on 21 October, all in under 72 hours.
The AI‑driven Lidar Hunt
The author prompted four large‑language agents—Gemini, ChatGPT, Claude and Grok—to search for publicly available lidar over the Maya lowlands. NASA’s 2013 G‑LiHT campaign had already stripped the canopy from roughly 1,200 km² of jungle, leaving bare‑ground elevation maps. The agents converted the strips into raster images and instructed 48 parallel instances to scan each picture for geometric anomalies. No custom classifier was used; the bots followed a step‑by‑step guide that taught them to ignore natural hills, sinkholes and roads while hunting for straight edges, plazas and pyramidal platforms. Within hours they flagged several candidate villages, one of which displayed a levelled plaza, an oval pyramid 8–12 m tall, a raised courtyard and a flat‑bottomed hollow that could be a rainwater reservoir.
How did AI agents locate the Maya site?
After the overnight scan, the agents cross‑checked the flagged locations against existing literature and the archives of Mexico’s National Institute of Anthropology and History (INAH). The check returned a single major village that did not appear in any published register. The agents then verified the raw point cloud, confirming that the brown points—ground returns—traced the plaza, pyramid and courtyard beneath 10–17 m of forest. They compiled a report, added confidential coordinates, and generated a Spanish translation on Claude’s suggestion. The report was sent to INAH on 20 October, completing the AI‑human workflow from data ingestion to official submission in under 72 hours.
Official response from INAH
INAH replied on 21 October, stating that their records contain no registered archaeological site at the supplied coordinates. The agency’s reply highlighted the absence of any entry in its public register, effectively confirming the AI‑flagged location as unmapped in official databases. The author notes that the site later appeared in Estrada‑Belli et al. 2025, meaning the AI agents had rediscovered a location already documented in academic literature but not yet entered into INAH’s system. The exchange demonstrates how rapid AI analysis can surface gaps between scholarly publications and governmental inventories.
Implications for digital archaeology
The episode shows that publicly released environmental lidar can be repurposed for heritage discovery in record time. By automating the visual inspection of millions of square meters, AI agents compress a task that would normally require months of fieldwork into a matter of days. The speed of the workflow—thought on 18 October, report on 20 October, reply on 21 October—suggests a new model for collaborative discovery between independent researchers and state agencies. However, the later citation of Estrada‑Belli et al. 2025 warns that AI outputs must still be cross‑referenced with academic sources to avoid false claims of novelty.
Frequently asked questions
Did the AI agents actually discover a new Maya site?
The agents flagged a major village not listed in INAH’s register, but the location later appeared in Estrada‑Belli et al. 2025, meaning they rediscovered a site already documented in academic literature.
How long did it take from the initial idea to a government reply?
The author thought of the project on 18 October, sent the report on 20 October, and received INAH’s reply on 21 October—under 72 hours total.
What technology allowed the AI agents to see through the jungle canopy?
Public NASA lidar data from 2013, which fires laser pulses through canopy gaps to map bare ground, was processed into images that the agents examined.
The rapid AI‑driven discovery underscores a shift: heritage agencies must now monitor open‑source geodata and AI outputs as part of their surveillance of cultural resources. As more lidar archives become public, the bottleneck will move from data collection to verification and integration. If institutions can keep pace, the pace of archaeological mapping could accelerate dramatically; if not, they risk being outrun by the very tools meant to aid them.
Sources: Hacker News post “My AI agents found an unmapped Maya site” (https://pietromiotti.com/latest/maya-site)