Google Earth and a Bar Conversation Yield an Archaeological Breakthrough and ‘Times’ Coverage for Alumnus Pablo Crespo

Pablo Crespo helped archaeologist Gino Caspari map Scythian burial sites across a huge swath of territory in present-day Russia, Mongolia and China using Google Earth images. (credit: Pablo Caspari)

Pablo Crespo (Ph.D. ’19, Economics) is the co-author of an archaeology paper that is receiving significant attention, including coverage by The New York Times. The story of how an economics Ph.D. collaborated on a breakthrough discovery involving 2,000-year-old burial tombs in the Altai Mountains of Mongolia involves a chance conversation in a bar and a willingness to try something new.
 
Crespo spoke to The Graduate Center about the discovery and its backstory.
 
The Graduate Center: The New York Times is covering your paper. That’s a big deal. What are the main takeaways of your paper that you want people unfamiliar with your field to know?
 
Crespo: We tackled the problem of finding Scythian tombs in the Altai mountains. These tombs are about 10 meters in diameter and the area in which they are suspected to be is over a million square kilometers. Since not much is known about this nomadic culture other than from their crafts, generally found inside the tombs, it is important to allow archaeologists to get to them. The area also covers militarized and conflictive areas and tombs have been known to be looted. However, they are easily distinguishable from aerial photography. Using known locations for tombs near China, we were able to design an AI classifier that could look at large areas of aerial photographs to identify what patches of land actually contained tombs. We reached remarkable accuracy even with relatively limited information with our method, providing archaeologists with a tool that could help them easily access sites without exhaustive expeditions nor research assistants looking at photographs for hours on end.
 
GC: What do you expect will be the impact of your findings? 
 
Crespo: Hopefully save a bit of that part of history! We used Google Earth pictures and a gaming computer to design the algorithm, as well as free software. This is a cheap, effective alternative to using satellite data that might be proprietary, and the efficiency of the model can make this type of research easily accessible to anyone who might need it or might want to undertake it themselves. We were cited as part of the revolution in “cyborg archaeology” in this paper. We hope that we have helped the new wave of interdisciplinary work between modern statistical modelling/machine learning and archaeology.
 
GC: You teamed up with Gino Caspari, an archaeologist at the University of Sydney, while you were pursuing your Ph.D. in Economics at The Graduate Center. That seems pretty unusual. How did you get involved in this research and this area of study? How does it pertain to your Ph.D. in Economics?
 
Crespo: My dissertation at the GC involved using machine learning to expand econometric methodology, which had components that could benefit from predictive modelling that was numerical rather than based on specific assumptions. Gino and I used to live in the same building while he was doing his master’s at Columbia University. We were friends, and one night at a bar he started talking to me about the difficulties of finding these tombs and methods he had considered to find them. I jokingly said, “I bet a data scientist could figure something out.” He then said, “How?” and this started us on the path of working together.
 
GC: What’s next for you in this area?
 
Crespo: We would like to expand the research from simple classification to detection and geolocation with a single model, and create hopefully a tool that could be used with other objects. We plan on using non-proprietary data for this purpose. The power of machine learning should help us keep things cheap and accessible for other researchers.
 
GC: What did you learn from publishing this paper that you can share with graduate students who are looking to publish their research and establish themselves in their fields? 
 
Crespo: It is okay to feel like one is out of their comfort zone. It is a good idea to make friends from other fields and listen to their research ideas. We were rejected from a journal before we got this paper published and the reviews were very harsh. But the revisions gave us a version we really liked. So, never underestimate the value of negative feedback, it can definitely help one make a product that one likes more!
 
GC: I see that you’re a senior data scientist at Etsy. That sounds interesting too. What do you do in that role and how did you land it? 
 
Crespo: I work for the Experimentation Science team. We work on developing methodology and tools centered around state-of-the-art causal inference for improving our experimental analysis and speed. This work involves research in both machine learning and traditional statistical inference. Etsy had posted a job for a brand-new team within their data science/machine learning group in 2019. The position seemed highly academic and exciting. I sent in an application to it thinking it a long shot and was fortunate enough that my particular research interests seemed to align with their work. It’s an incredible team and company; it has allowed for great opportunities for intellectual growth and impact.
 
GC: Is there anything else you’d like to add?
 
Crespo: Yes! Just because a project seems to be very odd, it does not mean you should abandon it. I am learning those can be the most fun!
 

Submitted on: NOV 24, 2020

Category: Alumni News | Economics | GCstories | General GC News