Artificial intelligence could help archaeologists quickly identify new sites to investigate from satellite imagery or aerial photos, and even spot ones we might otherwise have missed.
Archaeology is a painstaking, often slow, process. It involves looking for sites that are buried and these aren’t always obvious from surface studies. Researchers currently use satellite photos to decide where to dig, called remote sensing.
To speed things up, at the University of Bologna, Italy, and his colleagues have trained an AI model on existing, labelled satellite images of around 66,000 square kilometres of the historical region of Mesopotamia, in what is now the Middle East. This involved locations where archaeological surveys had taken place, , with a few in Iran.
Advertisement
Once the model was trained, the researchers tested it with other satellite images of areas where there have been known archaeological surveys and asked the AI to identify those worth digging.
The test areas covered parts of Mesopotamia and what seemed to be topologically similar sites in Uzbekistan. In Mesopotamia, the AI was able to identify dig-worthy sites with 80 per cent accuracy. This would allow an archaeologist to then assess a much smaller selection of areas to decide where to dig, says Casini.
However, with the Uzbekistan images, accuracy was less than 30 per cent.
Casini believes this is because the model was trained on Mesopotamian land and the more urban and vegetation-rich images from Uzbekistan flummoxed the AI.
“It’s providing value even if it’s not perfect,” says Casini, “because the last step that actually needs to be made will always be made by a human.” He believes it will speed up filtering potentially fruitful dig sites from those likely to be a waste of time. “Now you can do it in minutes with the system,” he says.
Archaeologist at Newcastle University, UK, sees the potential too, and says it reduces the need for archaeologists to manually comb through thousands of images of potential sites.
However, as the Uzbekistan result shows, the system won’t work everywhere. “The success of using AI depends on several factors,” says Brandolini, including the quality of the source images, the way the model is trained and challenges of false positives or negatives, because AI isn’t always able to distinguish between natural features and human-made structures.
Despite those limitations, Brandolini sees a place for AI like this in archaeology. “AI can help identify patterns in the data that might be difficult for humans to detect, leading to the discovery of sites that might have otherwise been overlooked,” he says.
Reference: arxiv.org/abs/2302.05286
