Meet Devin Guillory, a Staff Data Scientist at Etsy whose career in AI today started with a deep interest in robotics as a child. Learn about what a day at Etsy looks like for him, his proudest career moment, and more in this week’s installment of Role Models in AI.
We interviewed Devin as part of AI4ALL’s Role Models in AI series, where we feature the perspectives of people working in AI in a variety of ways. Check back here weekly for new interviews.
As told to Nicole Halmi of AI4ALL by Devin Guillory, edited by Panchami Bhat
NH: Can you describe what you do as a Staff Data Scientist at Etsy? What does a typical day look like for you? What kind of projects do you work on?
DG: As a Staff Data Scientist at Etsy my role is to take machine learning research and get it into production to affect our buyer-seller community in ways that align with our business goals. At a high level, I work on search and promoted listing ranking algorithms to help our sellers get their items in front of more people.
Depending on the stage of the projects I’m working on, I might be doing research or literature reviews, scoping out future work, conducting experiments, tweaking algorithms, or pushing validated offline approaches into production.
How did you decide to get a bachelor’s and master’s in electrical engineering? Were you interested in the field at a young age, or did you discover it in college? And how did you come to focus on computer vision, machine learning, and natural language processing?
From a young age, I was very interested in robotics. I knew I wanted to work in robotics, but I didn’t have a clue as to how one goes about doing that. We didn’t have robotics clubs or anything like that in high school, so I just continued to take math and some computer science courses in high school. When I got to college, I learned there are three pillars of robotics: a computer science element, an electrical engineering element, and a mechanical element. I knew the least about electrical engineering, so I figured I would try to learn it.
When I finally reached upper-level robotics courses — where we had projects like building a robot that played a board game — it became clear that computer vision was a hard and central problem. I dove into it, and my interest just kept growing. I also became interested in how natural language interacts with computer vision. Now, I’m using language and images for information retrieval, for powering search.
You’ve said that you “believe that diverse representations are central to machine learning, as incorporating various perspectives and sources of truth has been consistently shown to outperform homogenous systems.” What do you think are some of the barriers that prevent technologists from incorporating diverse perspectives?
I think one of the biggest issues [in AI] is not having diverse people in the room. Different people will be more in tune with different components, so they’re more likely to catch problems. It’s something that you get for free when you have a diverse group of collaborators or people working on things.
Machine learning algorithms learn whatever they’re exposed to. A central issue in machine learning is that want to be able to generalize, as you often want algorithms to be more broadly applicable. The best way to be able to do that is to capture more diverse representation in your data.
Using diverse data is one very effective approach to making algorithms more broadly applicable. It’s not just about the quantity, but how representative the data is that you’re collecting.
For example, think about Wikipedia as a dataset. You might assume a dataset as large as Wikipedia has enough information to build whatever system we want to build. Wikipedia won’t be representative of how long-form fiction will be written, though. It won’t account for how tweets, text messages, or emails are written. You need to be intentional in collecting a variety of forms of data instead of saying, “hey, this dataset already exists and it takes just one click for me to download and interact with it.”
What are some of the important things people should be doing to create a positive, inclusive, and ethical future for AI?
It gets very dangerous when we start to separate math from the social sciences. We should be pushing for more expansive education for people involved in the field, as well as more accountability — particularly as some of these AI systems are starting to affect millions, billions of people.
Part of the solution is inviting more people to the table and having more open discussions about the potential impacts of AI systems.
This might mean more education for the people in AI, or it might mean including people with more holistic backgrounds and different perspectives.
Who were your role models growing up? Do you have any role models now?
My parents and my older brother had a tremendous impact on me were great role models for growing up. Outside of that, there were people whose careers or actions I admired, like the long history of Black inventors. Garrett Morgan was one of the people I was particularly interested in.
One person who sticks out to me now is Paul Judge. He’s a scientist-slash-entrepreneur, running a couple of companies. He also happens to be from my hometown of Baton Rouge. I think the work he’s done in his career has been really exciting.
What has been the proudest or most exciting moment in your work so far?
One of the things I’m most proud of is the work that I did at Blackbird before coming to Etsy. We were a really small AI start-up, and we were dealing with computer vision, language processing, and search recommendation, all strung together. I was given challenging projects and didn’t know how to do them. Instead of backing down, I jumped in and learned a tremendous amount. We were able to produce products that people found useful. I was proud of my ability to get in over my head and come out on the other side.
What advice do you have for young people who are interested in AI who might just be starting their career journeys?
It’s very easy to get overwhelmed by the amount of material out there. I advise you to pick something you’re interested in and learn as much about it as possible. Along the way, you’ll continue to pick up broader skill sets. Don’t focus on trying to know everything in the beginning. Just run with something you’re interested in.
Don’t believe in pre-requisites. If there’s something you want to do, go after it. Try not to believe that you have to read certain things first, or take certain classes before you can get involved. You can do all of that while you’re pursuing whatever it is you want to do.
About Devin Guillory
Devin Guillory is Staff Data Scientist at Etsy where he architects machine learning systems to power search, recommendation, and advertising products. He believes that diverse representations are central to machine learning, as incorporating various perspectives, and sources of truth has been consistently shown to outperform homogenous systems. His research in the overlapping fields of Computer Vision, Natural Language Processing, and Information Retrieval has led to publications and presentations at prestigious AI conferences such as KDD and RSS. Devin received a patent on his novel approach to generating hierarchical classifiers for image and text data and his work designing AI systems for ecommerce companies as a founding engineer at Blackbird eventually led a high profile acquisition by Etsy.
Follow along with AI4ALL’s Role Models in AI series on Twitter and Facebook at #rolemodelsinAI. We’ll be publishing a new interview with an AI expert on Wednesdays this winter. The experts we feature are working in AI in a variety of roles and have taken a variety of paths to get there. They bring to life the importance of including a diversity of voices in the development and use of AI.