Blend of machine learning, botany unlocks secrets of ancient Greece

A rising junior at Yale College, Micah Gold can pilot drones and write algorithms. He’ll utilize both skills this summer in central Greece as part of a unique archaeological project.

Gold is part of a research team for the Yale Ancient Pharmacology Program (YAPP), a multidisciplinary initiative that harnesses ethnography, science, and technology to uncover evidence of how people lived thousands of years ago.

In Antikyra, a port on the Gulf of Corinth, the group will hunt for archaeological sites and plant specimens that could potentially yield exciting discoveries about medicines, olive oil, wine, perfumes, and other concoctions produced and consumed by the location’s ancient inhabitants.

For Gold, who expects to pursue a double major in applied mathematics and computer science, the archaeological fieldwork provides an opportunity to use his knowledge of machine learning to help locate potential sites and plants rich in phytochemicals — occasionally poisonous chemical compounds that protect plants from infections and insects and may, in some cases, possess medicinal qualities — while also broadening his intellectual horizons.

“I get to use what I’m learning in my classes, but I’m not only refining those skills,” he said. “I’m also learning new skills, like how to read archaeological literature and how scientific research is organized.”

It is Gold’s second stint assisting YAPP in the field. Working at the same location last summer, he learned to pilot drones, which were subsequently used to perform multi-spectral imaging of the region’s rough terrain, helping the researchers map and locate the remnants of ancient pottery and other artifacts. He then helped write algorithms that enabled them to organize and make sense of the multi-spectral data they had collected.

“I’d never been on an archaeological dig and had no idea what to expect,” he said. “It was an amazing experience.”

Gold and other talented undergraduate students have much to offer the interdisciplinary ancient pharmacology program, said Andrew Koh, YAPP’s principal investigator and research scientist at the Yale Peabody Museum.

“From day one, the students contribute,” he said. “We need and value their expertise. The typical archaeologist doesn’t know very much about machine learning.”

‘A high-definition view of the past’

Koh, who joined the Peabody as a research scientist last year, is a pioneer in the use of organic residue analysis in archaeology to identify the substances once held by ancient vessels, an emerging field that has helped reveal the nuanced ingredients people used in food, beverages, medicines, aromatics, and even opiates and hallucinogens.

It is an inherently interdisciplinary approach, drawing on chemistry, botany, and other hard sciences; ethnographic knowledge of ancient languages and cultures; and the latest technology to locate and analyze plants and ancient artifacts.  

Combining these modes of inquiry enables the recovery of lost knowledge, the recreation of potentially beneficial ancient substances, and a more nuanced and well-rounded understanding of how ancient people lived, Koh said. For example, rather than simply confirming that a ceramic vessel once contained wine, YAPP’s interdisciplinary approach potentially can identify the individual ingredients used in making the wine. That, he said, can reveal important context, such as whether the beverage had medicinal qualities or specific flavor profiles specific to a given region.

“We want to increase the resolution of our understanding of the past as much as possible, which means we can’t just rely on one discipline,” Koh said. “Chemistry can tell us certain things about the residues we discover, but reading ancient texts might offer recipes and other clues about the ingredients that people were using. Assembling all these pieces creates a high-definition view of the past.”

The Peabody’s extensive collections provide Koh with a ready source of materials to study. He has already made exciting discoveries within them. For example, he collaborated with Agnete Lassen, curator of the Yale Babylonian Collection, on an analysis of a vessel made of Egyptian alabaster dating to about 2,500 years ago, revealing that residues inside it were from an opioid concoction. 

Susan Butts, the Peabody’s director of collections and research, recalled taking Koh on a tour of the collections and needing to pull him from every storage facility to move to the next collection.

“He could find ways to collaborate in specific ways with everyone he met,” she said. “Unlike other scientists at the museum, who help oversee the care of specific collections as curators, Andrew is more of a free agent who can collaborate across collections. His research is especially suited to that because it is analytical and examines all different types of material for new and interesting information.”

While the Babylonian Collection and the Peabody’s anthropological collections are obviously relevant to Koh’s research, he is also working with the botanical collections and sees nearly limitless opportunities within the 14 million objects that compose the collections.

“I haven’t figured out how to collaborate with invertebrate paleontology,” he said.

As he unlocks secrets within the collections, his YAPP team is doing fieldwork to unearth new discoveries.

Antikyra, a culturally important region whose history dates back to the Middle Bronze Age, is a case in point.

Hunting for hellebore

Pausanias, a Greek geographer and early travel writer, discusses Phokis — the region where the YAPP team is working — in his “Description of Greece,” an exhaustive 10-book work from the second century C.E. that describes the countries topography and cultural geography. Specifically, he mentions a shrine to Isis, an Egyptian goddess who was worshipped throughout the Greco-Roman world.

The shrine would have functioned as a sort of medical clinic, Koh said, and its ruins potentially could yield a trove of vessels with interesting residues. It is possible to identify the site of the temple complex by triangulating the distances Pausanias used in describing its location, he said.

The region has a special association with hellebore, a legendary plant that could be used for either healing or harm, a duality inherent to the original Greek term, “pharmakon.” Koh noted that the 5^th-century BCE medical writer Thessalus, the son of the physician and philosopher Hippocrates, recounted that their ancestor, Nebros, advised an invading force to poison the water supply of a nearby city with hellebore, resulting in a wave of debilitating dysentery that contributed to the eventual slaughter of its citizens. That episode, Koh added, is widely thought to have inspired the Hippocratic Oath through which physicians pledge to do no harm.

This summer, the Yale research team will target hellebore from Mountain Helicon, a mountain in nearby Boeotia, for documentation with multispectral and phytochemical analyses to better understand its past and present relationship with the region.

Rather than digging based on hunches or educated guesses, the team uses technology — including an unmanned aerial vehicle (UAV) equipped with multi-spectral and thermal sensors — to target and document specific locations for excavation and study.

The idea is to work quickly and efficiently to identify promising sites without cramming the storage facilities of YAPP’s local partners with thousands of ceramic sherds. At the same time, the team will lean on the same technology to collect specimens of the local flora in collaboration with local botanists, which might have been ingredients in ancient medicines and beverages.

Boot camp

In April, the YAPP team held a boot camp on Horse Island — a 17-acre property the Peabody owns in the Thimble Islands off the coast of Branford, Connecticut — to prepare for its trip to Greece. The isolated island, where UAVs can fly without disturbing neighbors, is an ideal proving ground for the team’s technology, Koh said.

Gold and Murphy Tu, a graduate student in archaeology in the Graduate School of Arts and Sciences and fellow team member, flew a UAV equipped with an infrared sensor to test whether their algorithm can differentiate ceramics from rocks. (Paired with multi-spectral imaging, Koh said, UAVs can isolate pottery sherds and hold great promise for identifying plant life down to the genus, if not species, over vast stretches of land, including areas not easily accessible on foot.)

Patrick Sweeney, collections manager for the Peabody’s botany division, led a session on collecting plant specimens. The training removes unknowns and allows the team to hit the ground running, Koh said.

“We don’t want to go to Greece and discover we have technical issues,” he said.

Working smartly and quickly — the fieldwork will last weeks, not months — helps to accommodate talented students, who often have diverse interests and are involved in various projects simultaneously.

For instance, after Gold returns from his two weeks in Greece, he will begin an internship at Yale’s Wu Tsai Institute, working with Abhishek Bhattacharjee, a professor of computer science in the Yale School of Engineering and Applied Science, on brain-computer interfaces to treat neurological illnesses.

“It’s convenient because I don’t feel like I have to commit all of my time to the YAPP project,” Gold said. “I can come in and complete very targeted work and still make an important contribution.”

Gold connected with Koh through a first-year seminar he took with David Skelly, the Frank R. Oastler Professor of Ecology at Yale and director of the Peabody Museum, on working within the museum’s collections. As a student, he helped Koh to investigate the provenance of the Egyptian alabaster jar, which opened the opportunity to join the fieldwork team in Greece last summer.  He has since been developing a visual algorithm that can analyze massive amounts of multi-spectral and thermal data and isolate key regions of interest from areas. The algorithm helps researchers analyze in minutes what would otherwise take several tedious days to review.

Koh calls the ability to apply machine learning to analyze the resulting data “a game changer” that underscores the importance of YAPP’s multi-nodal approach.

“In the old days, archaeologists dug like crazy to find a useful site, but through our tech node, we can use our thermal data and machine learning to target our activities,” he said. “And that’s smart archaeology.”