Google researcher claims to have discovered a Maya site using AI agents; archaeologists point out it had long been recorded
Google researcher Pietro Miotti used 48 AI agents to analyze NASA lidar data and claimed to have discovered an unrecorded site in the Maya region of Mexico, but archaeologists pointed out that the site had been published in an academic journal more than a year earlier. Ivan Šprajc, an archaeologist and one of the paper's authors, said that merely identifying surface structures does not significantly increase knowledge of ancient civilizations; Miotti admitted that he is not an archaeologist and has revised his blog to add the contributions of previous researchers.
Why read this
AI score 52
This false-alarm incident illustrates the boundaries of AI agents in scientific discovery, and why literature checks and professional verification still cannot be omitted.
Topics
Body · AI translation
Google researcher Pietro Miotti recently claimed that he used a group of AI agents to discover a possibly unrecorded Maya site in Mexico's Maya region. However, archaeologists quickly pointed out that the site had in fact long since been discovered and had been published in an academic journal more than a year earlier.
Miotti first came up with the idea of using AI to find archaeological sites while traveling in Peru. He imagined what historical explorers could have achieved if they had been able to use modern technology. So he chose LiDAR technology, which can generate three-dimensional information about the surface through laser scanning, to look for traces of ancient buildings and settlements that might be covered by vegetation.
Miotti then assembled a team of 48 AI agents and had them analyze LiDAR data provided by NASA to look for possibly unrecorded archaeological sites. Because publicly available LiDAR data for Peru was insufficient in coverage, he ultimately shifted the scope of his research to the region where the Maya civilization was located.
These AI agents identified several locations worthy of further investigation, including a relatively large settlement. Miotti believed that this settlement did not appear to be in existing registry records. He said this was only his personal hobby project and that he did not use computing resources or other support provided by Google.
To verify the discovery, he contacted Mexico's National Institute of Anthropology and History (INAH) to ask whether the relevant location had already been registered, including whether there were unpublished research results. Miotti said he also used AI agents to translate the communications into Spanish. The institute replied that its registry records contained no information about the location.
Miotti then excitedly announced on his blog that the entire process, from coming up with the idea to receiving a formal reply from a Mexican government agency, had taken less than 72 hours. He believed this proved that AI agents could bring revolutionary changes to many scientific fields, and he called for more comprehensive global surface mapping using LiDAR technology so that AI could help find as-yet-unidentified historical remains.
However, problems soon emerged. Archaeologists found that the site Miotti claimed was not a new discovery. Archaeologist Ivan Šprajc of the Research Centre of the Slovenian Academy of Sciences and Arts was one of the authors of the earlier research paper on the site. The site had already been recorded in an academic journal, and the paper was published more than a year before Miotti's so-called discovery.
Sarah Parcak, a professor of archaeology at the University of Alabama at Birmingham, also criticized the episode on social media. She pointed out that researchers must carefully check existing academic work and must not conclude that a location had never been discovered or mapped simply because it did not appear in a government database.
Miotti admitted that he is not an archaeologist and had not realized that the absence of a government registry record does not mean a location has never been discovered or mapped by other researchers. He said that professional expertise is crucial for verifying AI agents' discoveries and acknowledged that both he and the agents had overlooked existing research. Because archaeologists usually do not publicly disclose the precise coordinates of sites, determining whether a location is already known often requires comparing LiDAR images with illustrations in academic papers rather than simply querying a database. This kind of work remains relatively difficult for existing AI tools.
In the view of the archaeological community, this episode reflects a fundamental limitation of AI-assisted scientific research: discovering some kind of surface structure does not amount to actually completing a scientific discovery. AI can rapidly process massive amounts of data and identify terrain features that are not easy for the human eye to notice, but it may not possess the research results already accumulated in the relevant field, and it may also fail to understand what academic significance a discovery actually has.
Elizabeth Graham, a professor of Mesoamerican archaeology at University College London, pointed out that archaeologists in the Maya region actually already know of many sites that have not been excavated or systematically mapped, and that quite a few of these locations were even found by researchers through on-foot field surveys. She also stressed that local communities often have long known of these sites, so the very use of the word "discovery" requires caution.
Graham acknowledged that Miotti's work using AI to process LiDAR data and identify surface mounds has some value, but she believes that the bigger obstacle facing current archaeological research is not a lack of sites waiting to be discovered, but a lack of sufficient funding for follow-up mapping and excavation. Even if AI can continuously identify more potential sites, without funding support it would still be difficult for researchers to further confirm their nature and historical value.
Šprajc likewise believes that archaeologists have long used LiDAR and other advanced remote sensing technologies extensively, so merely identifying certain surface structures cannot significantly increase humanity's understanding of ancient civilizations. He also pointed out that querying Mexico's National Institute of Anthropology and History is not sufficient for verification, because the institution does not hold all possible archaeological information.
He worries that as more and more amateur researchers use AI to announce so-called new discoveries, professional archaeologists may receive large numbers of insufficiently verified leads and inquiries, which would instead increase their workload. In addition, if amateur researchers publicly release the precise locations of sites, they may attract looters to go and damage the sites. However, as long as these discoveries do not lead to the looting of cultural relics, he believes that AI's participation in related research is not entirely without value.
This episode also reflects an increasingly obvious problem after AI entered the field of scientific research: AI tools may perform excellently in data processing and pattern recognition, but they do not necessarily fully understand the historical context, existing results, and academic norms of a research field. Even if a phenomenon identified by AI does indeed exist, without fully checking the existing literature and respecting the work of previous researchers, it is very easy to package existing results as new breakthroughs.
Miotti ultimately said that after archaeologists pointed out the problems, he promptly revised the content of his blog and added the contributions of previous researchers. He believes that this AI workflow may still help professional researchers process large amounts of survey data, and said that archaeologists and doctoral students have contacted him hoping to learn whether this method can be applied to their own research projects.