By Mike Toner
Archaeology has traditionally been a slow, methodical science, accumulating knowledge of past cultures, a spadeful of earth or a cache of stone tools at a time. But times are changing. New technologies—airborne laser scanning, ground penetrating radar, magnetometry, and side-scan sonar—now gather gigabytes of information in hours that would once have taken archaeologists months or years to collect. As the deluge of data has grown, archaeologists are turning to new tools—veritable digital trowels—to cope with the torrents of new information. As the demand grows for ever more sophisticated methods of crunching and analysis, researchers are discovering new promises, and potential new pitfalls, in the use of artificial intelligence (AI).
In the Maya Lowlands of Mexico and Guatemala, for instance, AI analysis of aerial surveys is helping archaeologists identify previously undiscovered Maya ruins beneath the jungle canopy. Along the Atlantic Coast, AI is aiding the discovery of 4,000-year-old shell rings before they are washed away by rising seas. In Michigan, AI is being used to predict the location of ancient hunting sites by mapping the migration of computer-generated caribou. In the desert of southern Peru, AI-assisted surveys have identified hundreds of new glyphs, doubling the known number of features in the enigmatic Nazca Lines.
A transfer-printed plate identification tool is currently in beta testing. One of the goals of the project is to deal with inconsistencies in the designs that make more mechanical identification possible. Archaeologically recovered examples are often not only fragmentary or otherwise damaged, but often the manufacturing process produces slightly inconsistent, though still recognizable, patterns. Courtesy of John Chenoweth.
Popular awareness of AI, inspired by the explosive growth of user-friendly large language models like ChatGPT, Claude, and Gemini, is only a few years old. But the use of other forms of artificial intelligence in archaeology is not new. Pattern-recognition algorithms were used to classify artifacts and simulate them as early as the 1970s. Computer simulations were used to model Ancestral Puebloan demographics in the 1980s, for example.
Despite the fact that popular AI models have access to billions of words and millions of scientific abstracts and papers, when it comes to archaeology, they still have limitations. Just ask AI: “Most archaeological data sets are incomplete, inconsistent, or recorded in incompatible formats,” explains ChatGPT 5.1. “Field notes, sketches, artifact logs, and survey data vary by site and decade. AI models need large, clean, labeled data sets, which archaeology rarely has. AI thrives on scale. Archaeology deals in fragments.” But archaeology’s accumulation of fragments—digital fragments, not pot sherds—is growing by leaps and bounds. And as early versions of machine intelligence have evolved into “deep learning” neural networks capable of “teaching” themselves when archaeologists train them on existing data, AI is becoming a potent part of archaeology’s high-tech toolbox.
This is an excerpt of ‘When Data Outruns the Dig’ in American Archaeology, Spring 2026, Vol. 30, No. 1. Subscribe to read the full text
