Play Live Radio
Next Up:
0:00 0:00
Available On Air Stations

Archaeologists Beginning To Leave The Tedious Work Of Sorting To Computers

Leszek Pawlowicz
Northern Arizona University
The many designs of Tusayan-White-Ware in sherds.

Why do the eye-straining work of sorting pottery at an archaeological dig site when a computer will do it for you? Researchers at Northern Arizona University (NAU) say a kind of machine learning is beginning to catch on with archaeologists and has applications in other fields.

Rapidly and accurately sorting thousands of pottery designs in the Southwest is no easy task. But it's important work to help understand ancient societies.

Credit Leszek Pawlowicz / Northern Arizona University
Northern Arizona University
(from left) NAU Professor Chris Downum and Adjunct Faculty Leszek Pawlowicz apparently have found a much easier way to sort pottery pieces but now have to convince others.

One of the problems is that the people who could consistently identify the sherds are no longer living, according to NAU Anthropology Professor Chris Downum.

"The wizards of pottery are the ones who learned the types and they pass that along through kind of an apprentice learning system," he says. "And what's happened that is a problem is the people who created the typology have passed away and so have the people they have trained them have passed away." 

Simply Take A Picture And Upload It

Enter CNN, or Convolutional Neural Networks, a type of machine learning that emulates the human thought process using digital photographs.

Downum and NAU Adjunct Professor Leszek Pawlowicz gathered thousands of pictures of pottery fragments known as Tusayan White Ware.

Then they recruited four experienced archaeologists to identify every sherd. This was to train the computer.

"The beauty of this particular approach is that you don't actually teach the computer. It's actually set up to teach itself," says Pawlowicz.

The archaeologists must have done a pretty good job. The computer outperformed two of them and was comparable with the other two.

"Ultimately what we're thinking of doing is putting the model into a smartphone app," says Pawlowicz. He says this kind of artificial intelligence is not just useful for pottery but for anything that would have variation you can see in a digital photograph.

Now the challenge is convincing other archaeologists to use and trust the system. At an American Archaeology virtual conference in April, two of the presenters used similar approaches so the method appears to be catching on. But Downum says some are leery.

"We're starting to get some pushback from some of our colleagues. We have a marketing job to do because they're very skeptical about this approach." He says they are a stubborn bunch.

Their research is published in the Journal of Archaeological Science.

Ann Thompson has decades of journalism experience in the Greater Cincinnati market and brings a wealth of knowledge and expertise to her reporting.