Claims that AI agents discovered unknown Maya site are false

Earlier this week, Pietro Miotti, a researcher at Google, asserted that he had discovered a potentially unmapped Maya site in Mexico using a team of AI agents. However, the site had already been discovered and documented in an academic journal last year. This incident has triggered criticism from archaeologists online and highlights the growing tension in science as amateurs use powerful AI tools in specialist fields without the necessary expertise.

Miotti explained that the idea to use AI for archaeological searches occurred to him while traveling in Peru. He considered how historic explorers might utilise modern technology and decided on lidar, a laser-based mapping system that creates detailed three-dimensional images. He deployed a team of 48 AI agents to scan NASA lidar data for signs of undocumented sites. As public lidar coverage in Peru was insufficient, he shifted his focus to the Maya region.

The agents identified various promising locations, including one significant village that appeared to be unregistered in existing records. Miotti, who described this as a personal side project conducted without Google compute or resources, contacted Mexico’s National Institute of Anthropology and History (INAH) to confirm if the discovery was known. Using his agents to translate his correspondence into Spanish, Miotti said INAH informed him they had no record of a registered site at that location.

“It took less than 72 hours to go from a random thought while traveling to an official reply from the Mexican government. This is by far the best time ever to be alive,” Miotti stated on his blog. He emphasised that he is not an archaeologist and noted that any finding would require verification through fieldwork. Nevertheless, he hailed the experiment as proof that “we’re on the brink of a revolution across every science” and suggested that the “entire planet” should be mapped with lidar to allow AI agents to “help us find the rest.”

Had Miotti been an archaeologist familiar with the region, he might have realised the site was already known and documented in an academic journal more than a year prior. Archaeologists quickly highlighted this oversight on social media platforms. Sarah Parcak, an archaeology professor at the University of Alabama at Birmingham, posted on Bluesky regarding the importance of verifying findings against existing research.

“This is why we have peer review,” Parcak wrote, noting that an absence in a government database does not mean a site is unknown or unmapped. Other experts criticised the project for spreading misinformation and diminishing the work of archaeologists who have used digital tools for decades.

“This was a personal project that I was excited to share,” Miotti told The Verge. “I didn’t expect it to get this much attention.” He admitted that, as a non-archaeologist, he failed to realise that a site not being registered is different from it not being mapped, adding that expert knowledge “is needed to verify what the agents find.” He also acknowledged the limitations of his AI, noting that archaeologists do not typically publish site coordinates. Therefore, checking if a site is known requires comparing images against figures in papers, which is difficult for agents. He stated that he updated his blog as soon as people pointed out the error to credit the original work.

Despite the criticism, Miotti argued that this demonstrates how agents could assist experts in sorting through large volumes of data. He claimed to have been encouraged by interest from archaeologists and PhD students who reached out to him, excited about the workflow and potential applications for their own survey data. However, Elizabeth Graham, a professor of Mesoamerican archaeology at University College London, told The Verge that the problem is not discovering new sites.

“Most of us working in the Maya area know about tons of sites not explored or excavated or mapped, most of these found on foot,” she said. Graham also critiqued the term “discover,” noting that local communities are often aware of sites in their area.

While Graham credited Miotti with doing a decent job of using AI to map mounds from lidar data, noting that “the interpretation of lidar data can be tricky,” she argued that the primary obstacle is a lack of research funding. “The fact is, there just isn’t the funding to excavate or map sites as there was in the past, so in a way it doesn’t matter how many more are ‘discovered.’”