In 1981, then–graduate student Melinda Zeder was sorting through animal bones from a Paleolithic cave in southwestern Iran when she came upon a fragment she couldn’t identify. “When you can’t tell stone from bone, you place your tongue on it,” says Zeder, now an archaeozoologist at the Smithsonian Institution’s National Museum of Natural History. “If it’s bone, it will stick.”

The object didn’t stick. In fact, it began to dissolve on Zeder’s tongue. Puzzled, she turned to her more experienced colleague and asked what he thought it was. “Oh,” he smiled. “That’s hyena poop.”

Such ancient feces can hang around for thousands of years, even retaining their original shape and color. And archaeologists can typically differentiate human from animal poop, based on size and other attributes. But doggy dung, it turns out, is remarkably hard to distinguish from the human kind—something that can stump researchers trying to reconstruct what ancient people ate.

“When I give talks, I ask audiences to guess,” says Christina Warinner, a molecular archaeologist at Harvard University. “They always guess wrong.”

Now, Warinner and colleagues have developed a tool based on artificial intelligence that they claim can accurately tell human and dog “paleofeces” apart. And after analyzing more than a dozen samples spanning thousands of years, they’ve come to a surprising conclusion: The archaeological record is full of dog poop.

“There’s a lot of really great things you can do with this,” says Zeder, who calls the new work a “leap forward.” If refined, she says, the method could help reveal key milestones in dog domestication.

None of that was on Warinner’s mind when she started to ask archaeologists around the world for samples of ancient human feces. She studies how the human microbiome—the vast populations of bacteria that inhabit our intestines—has changed over time. Such changes are influenced by where we live and what we eat and have been linked to diseases including arthritis and obesity. They also leave traces in our feces.

The samples Warinner received baffled her, however. “We thought all of them were human,” she says. “But the data we got back were really strange.”

Human paleofeces can contain traces of canine genetic material because some people ate dogs. And dog excrement can contain traces of human DNA, because dogs sometimes eat human poop. But when Warinner’s team analyzed genetic material from the excrement—some of it in a fossilized form known as a coprolite—a few of the samples contained so much canine DNA that they could only have come from dogs.

Seeking a better way to distinguish the two, Warinner turned to one of her graduate students, Maxime Borry, who’s working toward a Ph.D. in bioinformatics at the Max Planck Institute for the Science of Human History. Borry amassed all the DNA from the fecal samples, which included not only human and dog genetic material, but sequences from microbes, plants, and anything else in the owner’s intestines. He then trained a machine learning program—which learns to make correlations among massive amounts of data—on modern samples of human and dog excrement.

The researchers applied the resulting program—christened coproID—to 13 samples, ranging from dung recovered from a 7000-year-old Chinese farming village to a 400-year-old home in southern England. They also tested seven control samples: sediments that did not contain feces but were from places feces might be found, including ancient garbage piles and the pelvic cavities of human skeletons.

The program classified all the control samples as unlikely to be feces. It also confidently identified seven of the ancient poops—five as human, two as canine—the team reports today in PeerJ. The genetic profiles of three other samples suggest they came from canines as well, Warinner says.

One of the most surprising finds was from the 17th century British home. During a renovation in the 1980s, workers had come across a chamber pot—complete with its “deposit”—near the original roof. They sent it to a local museum, where it sat for decades with the label “Three Human Coprolites.” But coproID reported that the feces came from a dog.

“How it got there, who can tell,” laughs Kate Britton, an archaeologist at the University of Aberdeen, who sent the chamber pot specimen to Warinner. She suspects that either some postmedieval owner was too lazy to take their dog for a walk, or the renovators were playing a prank.

Zeder hopes the new approach will offer insight into the evolution of the human-dog relationship. We domesticated dogs more than 15,000 years ago, but exactly when, where, and how this happened remains a mystery. At some point, she says, our canine pals evolved from carnivorous wolves to omnivorous dogs as humans began to feed them table scraps. Using feces to mark how the dog microbiome—and then genome—evolved to process these new foods could reveal milestones in the human-canine relationship. “The ability to track this through time is really exciting,” she says.

Still, Ainara Sistiaga, a molecular geoarchaeologist at the University of Copenhagen, says the approach isn’t quite ready for prime time. Sistiaga, who has studied the feces of everything from dinosaurs to Neanderthals, notes that the canine data used to train coproID came exclusively from Western dogs that ate pet food—hardly a diet of ancient times. That may be why the program struggled to identify some of the dog excrement. “The more data we put in, the more useful this tool will be,” she says.

In the meantime, Borry is getting used to his new identity as “the guy working on dog poop.” At a recent departmental retreat, he says, Warinner gave everyone a pop quiz, and Borry’s team lost. His consolation prize: a plastic dog that poops Play-Doh. “I don’t really think it was a prize for losers,” he says. “She just really wanted to give it to me.”