Google SWE uses AI agents to identify potential Maya site in lidar data
AI-Assisted Discovery in Archaeology
A software engineer recently demonstrated the potential of agentic AI in scientific discovery by using 48 AI agents to scan 1,200 km² of public lidar data for previously unmapped Maya structures. The process, which spanned 72 hours from initial concept to an official response from Mexico's National Institute of Anthropology and History (INAH), highlights how AI can lower the barrier for non-experts to contribute to specialized fields like archaeology.
Methodology and Workflow
The project relied on public lidar datasets, which allow researchers to strip away dense jungle canopy to reveal ground-level features. The engineer employed 48 AI agents to process the data by converting lidar strips into images. The agents were instructed to act as archaeologists, manually scanning these images for geometric patterns—such as pyramids, plazas, and platforms—that deviate from natural topography.
Rather than using a specialized machine learning classifier, the agents followed a set of human-provided guidelines to identify potential structures. Once the agents flagged a promising hilltop, they performed a literature and archive search to verify if the site had been previously recorded. The final report, including coordinates and a 3D reconstruction, was then submitted to INAH for official review.
Verification and Community Feedback
While the initial report suggested the site was unmapped, subsequent community investigation and author updates clarified that the site had been previously documented in a 2025 academic paper by Estrada-Belli, Canuto, Šprajc, and Fernandez-Diaz. The author acknowledged this, noting that while the agents and the author missed the existing record, the experiment successfully demonstrated the efficacy of "vibe-archaeology"—using AI to rapidly scan vast datasets for anomalies.
Commenters on the project emphasized several critical takeaways regarding the use of AI in scientific research:
- The Importance of Verification: Critics pointed out that AI agents can be prone to hallucinations or overconfidence. As one user noted, "This post turned out to be a perfect example of the blind faith in LLM agents. Their outputs are not to be trusted and must be rigorously checked."
- Resource Accessibility: Some users questioned the scalability and cost of running 48 concurrent agents, noting that such workflows require significant compute budgets that are not currently accessible to the average user.
- Ethical Considerations: There is ongoing debate regarding the impact of "citizen scientists" using AI to discover sensitive sites. Concerns include the potential for increased looting, the burden placed on local authorities to verify and protect sites, and the ethics of using AI to generate official reports without subject matter expertise.
The Future of AI-Driven Exploration
Despite the caveats, the project serves as a proof-of-concept for how AI can democratize data analysis. By applying large-scale compute to public datasets, individuals can perform tasks that previously required specialized teams. As noted by one commenter, "I'm a software engineer, not an astronomer. Being able to explore an entirely different field and independently arrive at an existing discovery was pretty remarkable." The consensus among observers is that while AI will not replace expert archaeologists, it will likely become a standard tool for identifying candidates for field study, provided that human oversight remains the final arbiter of discovery.
Sources
Related
- Dispatch
- Dispatch
- Dispatch
- Dispatch
- Dispatch