When it came to the time-consuming operation of identifying ceramic fragments, researchers discovered that a deep-learning model was just as accurate as four human specialists and significantly more efficient.
The arduous task of categorizing pieces of pottery into subtypes is an important part of an archaeologist’s profession. When archaeologists are asked why they placed a fragment in a particular category, they often struggle to explain how they arrived at that judgment.
“It’s kind of like looking at a photograph of Elvis Presley and looking at a photo of an impersonator,” said Christian Downum, an anthropology professor at Northern Arizona University. “You know something is off with the impersonator, but it’s hard to specify why it’s not Elvis.”
However, archaeologists have now proved that a computer can be programmed to perform this crucial task as effectively as they can. Researchers reported in the June issue of The Journal of Archaeological Science that a deep-learning model categorized photos of ornamented shards as accurately — and sometimes more accurately — than four expert archaeologists.
“It doesn’t hurt my feelings,” Dr. Downum, one of the study’s authors, said. Rather, he said, it should improve the field by freeing up time and replacing “the subjective and difficult-to-describe process of classification with a system that gives the same result every time.”
Tusayan White Ware, a type of painted hand-formed pottery used for serving meals and keeping water between 825 and 1300 in the canyons and mesas of northeastern Arizona, was the subject of the study. Tusayan White Ware pieces feature regular patterns based on the time period in which they were manufactured, archaeologists discovered in the 1920s.
The researchers enlisted the help of four of the most knowledgeable experts on this style of pottery. Each had spent 30 years or more studying ceramics and had identified tens of thousands of Tusayan White Ware fragments earlier.
They also spent nearly four hours sorting images of Tusayan White Ware using a neural network, a powerful mathematical system that can learn specific jobs by evaluating large volumes of data.
Humans and machines were given the task of categorizing thousands of photographs into one of nine categories and were graded on their accuracy.
According to the researchers, the neural network tied two of the human analysts for accuracy and outperformed the other two.
In addition, the machine was significantly more efficient. According to Dr. Pawlowicz, none of the human analysts wanted to go through all 3,000 photos without stopping because the process was tedious.
In just a few minutes, the neural network whizzed through thousands of photos.
In comparison to its living, breathing opponents, the computer program was not only more efficient and accurate than the archaeologists, but it could also explain why it had classified shards in a given way. In one case, the machine made a brilliant sorting observation that the researchers hadn’t seen before: According to Leszek Pawlowicz, an adjunct faculty member at Northern Arizona University and another study author, two comparable types of pottery with barbed line design components could be identified by whether the lines linked at right angles or were parallel.
The machine also outperformed humans in providing only one response for each classification; the participating archaeologists frequently disputed on how things were classified, according to the authors, which is a known issue that hinders archaeological studies.
The neural network performed as well as — or sometimes better than — the archaeologists, according to Phillip Isola, an electrical engineering and computer science professor at M.I.T. who was not engaged in the project.
“It’s the same story we’ve heard a few times now,” Dr. Isola said. In the field of medical imaging, for example, researchers have found that neural networks rival radiologists at identifying tumors. Academics are also using similar tools to categorize plant and bird types.
This isn’t the first time archaeologists have used artificial intelligence in their research. French academics used machine learning to classify medieval French pottery in 2015. In addition, a group of archaeologists and computer scientists from five nations are working on a digital tool to classify pottery fragments. Neither of these programs, on the other hand, openly sets man against machine.
Since the publication of the paper, some archaeologists have expressed concerns to the authors about being replaced by machines. Drs. Downum and Pawlowicz indicated they were unconcerned about such an occurrence.
“We’re the ones that decide what’s important to study,” Dr. Downum said.