Ancient Maya site uncovered with ‘vibe archaeology’ ignites firestorm over AI in scienceNEWS | 19 November 2026Contrary to online claims, AI did not discover an ancient Maya site. But it did spark a clash over how the technology should be used in science
On November 10 artificial intelligence researcher Pietro Miotti shared on X what appeared to be new evidence of a previously unmapped ancient Maya civic center at an undisclosed location in Mexico. Miotti said he had found the site using AI while on vacation in Peru. The post went viral, with other social media users cheering this “vibe archaeology” as another example of an AI-assisted breakthrough. Major accounts, including the online betting platform Polymarket, boosted his story. Some national media outlets picked it up, too.
But the site wasn’t a new discovery—it had been mapped at least once before. Archaeologists and other scientists jumped on the error as exemplary of why “vibe” science is bad science. The controversy was short-lived—but illuminating, exposing a deep divide over the role of AI in science as a whole.
In a blog post, Miotti described how he had prompted dozens of AI agents to sift through publicly available lidar (light detection and ranging) data from NASA to look for signs of now hidden structures across what was once the Maya Empire. One area appeared particularly interesting—the AI couldn’t identify any evidence that the site had been documented in academic papers. It suggested Miotti reach out to Mexican officials about the findings.
On supporting science journalism
If you're enjoying this article, consider supporting our award-winning journalism by subscribing. By purchasing a subscription you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.
Miotti did just that. Mexico’s National Institute of Anthropology and History (INAH) replied, saying the location wasn’t registered by the agency—a clue, but not confirmation, that it might be a legitimate discovery. (INAH didn’t respond to a request for comment from Scientific American.) “By analyzing NASA lidar data, my AI agents discovered a Maya site in Mexico that, as far as anyone can tell, no one had reported before,” Miotti wrote in the initial version of his blog post. (The post has since been updated.)
The enthusiasm for AI as an engine of discovery is not totally misplaced. Just this year, as Scientific American has reported, OpenAI announced that its large language model (LLM) had solved the Navier-Stokes problem, a fluid dynamics math problem that had gone unsolved for centuries. Myriad other math problems have crumbled under AI since then. Some experts have said that particle physics or neuroscience may be the next field to be upset by AI—but really, no discipline is likely to be untouched by this transformative tech.
The excitement over Miotti’s find was cut short, however. Less than a day after he shared his findings on X, Sarah Parcak, a professor at the University of Alabama at Birmingham, noted on social media that the site had been documented at least once before—by Francisco Estrada-Belli, a professor at Tulane University. Estrada-Belli and his colleagues published a description of the site in the Journal of Archaeological Science: Reports in 2025. “This is why peer review exists,” Parcak tells Scientific American.
In fact, their work built on lidar data that had been analyzed before, as early as 2016, says Marcello Canuto, a co-author of the 2025 study and a professor of anthropology at Tulane. The site is among “dozens and dozens” of others like it across the region, he says. Ivan Šprajc, another co-author of the 2025 study and head of the Institute of Anthropological and Spatial Studies at the Research Center of the Slovenian Academy of Sciences and Arts, helped conduct a visit to the site in May. It is a “peripheral group” of a larger center, he says.
For some commentators on social media, the apparent mix-up was evidence of AI’s shortcomings rather than its strengths. “I’ve been suggesting the term ‘artificial mediocrity,’ since what most of these programs do is to imitate what humans have said to similar queries,” one Bluesky user wrote in response to Parcak’s post.
AI, of course, misfires. It hallucinates. Its abilities rely on information that’s available online, not in the margins of an expert’s field notebooks. “I can’t tell you how many thousands upon thousands, millions, of articles, books, archeological datasets are not contained within any publicly accessible databases,” Parcak says. “Not everything is online.” That’s where human expertise—and collaboration—comes in, she says.
“Why would we rely on machines? We need other people to tell us we’re wrong,” Parcak says.
“I think it’s great that people want to be citizen scientists using AI and experiment with it,” Estrada-Belli says. “But I think they shouldn’t just blindly take the results without doing some background checks or due diligence before they go out and make a claim like that.”
Others have noted that the debacle overshadowed the work archaeologists are doing with AI. In recent years, for instance, AI has helped decipher ancient Greek texts, reveal design secrets of Etruscan tombs and reconstruct ruins. “We’re all pushing the limits of what machine learning can and can’t do,” Parcak says.
“All the AI bros think we’re luddites, putting our trowels in the ground, and we don’t do computer,” she adds.
Users on X added a community note to Miotti’s original discovery post, with references to the 2025 paper. He later posted his own correction on X and updated his blog post.
“Seeing the post go viral on X was completely unexpected; it was meant to be read mostly by my AI community,” Miotti tells Scientific American.
“As soon as archaeologists and the community pointed me to the study, I updated both the blog post and the X thread to credit their survey,” he adds. Miotti says he hasn’t been in touch with Estrada-Belli directly but is “glad his work was highlighted in the discussion and is getting the credit it deserves.”
Miotti maintains that the experience reflects a new frontier in scientific research, driven by AI. “Five years ago, screening 1,200 km2 of raw lidar data required very specialized knowledge,” he says. “Today a single person with a hypothesis can run AI agents overnight to triage candidate structures for experts to evaluate.”
“At the same time,” he adds, “this experience also highlighted where the real human contribution stands: literature review, field verification and other more specific domain expertise.”Author: Claire Cameron. Jackie Flynn Mogensen. Source