Meet the Team: AI2’s Young Investigators, Pt. 2
The postdoctoral Young Investigator program at AI2 allows those within a year of completing their PhD, or those who already have a PhD, to spend one to three years diving deeply into their area of research interest before moving into the next phase of their career. These creative and passionate people are provided with a dedicated AI2 mentor and have access to a team of world-class researchers and engineers to support their work.
We spoke with a group of incredible Young Investigators (YIs) currently working on our Semantic Scholar team at AI2 to learn about why they chose this experience, what their most memorable moments were, and where their career is taking them.
If you’re interested in a Young Investigator role with AI2, learn more about the program here, and have a look at all of the roles currently available across AI2.
Please introduce yourself, tell us a little about the work you’ve done at AI2, and where you’re headed next.
Tom Hope: I started as a Young Investigator on the Semantic Scholar team in early 2020. As of this month, I’m an assistant professor at the Hebrew University of Jerusalem’s School of Computer Science and Engineering (joining previous AI2 YIs Roy Schwartz and Gabriel Stanovsky), while remaining affiliated with AI2 as a research scientist. In my work at AI2, I develop artificial intelligence and natural language processing methods to augment and scale scientific knowledge discovery, by harnessing vast and diverse repositories of scientific knowledge (e.g., literature, knowledge bases, electronic medical records). I’ve done work on methods that mine scientific literature and knowledge bases to help discover new directions and solutions to problems, generate hypotheses, and build connections across different ideas and areas. Recently, we developed a method that predicts clinical outcomes of hospital patients by mining the literature for patient-specific medical literature, NLP-powered models for predicting off-label drug uses, and exploratory search engines for discovering biomedical causal relationships, challenges and directions in COVID-19 research, and authors who inspire novel directions.
Lucy Lu Wang: I joined the Semantic Scholar team as a Young Investigator in mid-2019. My research focuses on extracting useful information from scientific and medical texts, and more recently, on document understanding and accessibility for blind and low vision audiences. Over the course of my time here, I’ve had a chance to collaborate with tons of people, both on Semantic Scholar and other teams, which has been a great experience. I’m also looking forward to the next stage of my career — starting as an assistant professor at the University of Washington Information School right here in Seattle! I’ll be focusing a little more on NLP applications in healthcare since I’m forming more collaborations directly in that domain at UW.
Lucy, what led you down the path of accessibility work?
Lucy: My introduction to accessibility research started with a collaboration with colleagues at UW. The project was led by several graduate students, Kelly Mack, Emma McDonnell, and Dhruv Jain, and the PIs on the project were Leah Findlater (HCDE) and Jon Froehlich (CSE). We were doing a literature review on accessibility research and how the field had changed over the past decade, and I was using Semantic Scholar’s data to help enrich that work. Subsequently at AI2, we started pursuing the question of, okay, there’s significant research into accessibility, but how does that intersect with scientific NLP and research publishing? I was working with former Predoctoral Young Investigator Isabel Cachola (now at John Hopkins), and a research scientist on our team, Jonathan Bragg, on this problem, and we found that outside of the immediate fields of accessible computing and human-computer interaction, papers from other fields were woefully inaccessible to blind and low-vision researchers who use screen readers. That led to the building of tools like SciA11y and Paper to HTML, and new models like VILA to determine whether state-of-the-art NLP models could be used to address the problem. And that has led to a whole line of related research. The deeper you go, the more problems you uncover.
What made you decide to be a YI at AI2?
Tom: I’ve always been fascinated by the origins of innovative ideas across scientific and technological disciplines. The stories behind significant inventions are often interesting in their own right. In my PhD I worked on augmenting creativity and innovation with machine learning and NLP. The Semantic Scholar group at AI2 is a world leader in extracting knowledge from scientific papers and developing systems for scholarly discovery, so it was a perfect place to continue exploring my research agenda.
Lucy: As I was finishing up grad school, I knew I wanted to pursue academic research. I had previously interned at AI2, and the YI program was just starting at the time of my finishing my PhD. I knew I’d enjoy the environment based on the positive experiences I had during my internship. I also knew I would gain a lot of experience in NLP. Having the support, and the freedom and flexibility to pursue ideas within that supporting structure — it’s led to many successful and unexpected projects, like working on SUPP.AI, or working in the accessibility space. I don’t think I’d have been able to do that in another program.
What’s the most surprising, interesting, or memorable thing that happened with your work?
Tom: Well, undoubtedly a global pandemic coinciding with the early days of my postdoc could be considered surprising and memorable. Aside from the general tribulations of COVID-19, it also had a dramatic effect on my research. In response to COVID-19, AI2 worked in collaboration with The White House Office of Science and Technology Policy and released a large resource of scientific papers on COVID-19 and related coronavirus research. My interest in pursuing the use of the literature for AI-assisted scientific discovery rapidly found a very immediate application. I began by developing a novel visualization engine of connections across this diverse literature, in a system that was later further developed and named SciSight. Since then I have developed several other discovery engines over COVID-19 literature, evaluated in user studies with researchers, including MDs in a large hospital in Seattle.
Lucy: I think the most surprising thing definitely has to be what happened with SUPP.AI. SUPP.AI started as a Hackathon project back in my first year, and we put the initial prototype together in just a few days. We released it with a slicked-up UI in the fall of 2019, and it’s grown to support an amazing number of users (over 25,000 a month)! I didn’t expect the idea that we brought from conception to model to prototype over the course of a couple of weeks to be so useful to a broad audience. The goal of the tool is to help people discover information about dietary supplements and how they interact with medications. There’s a whole slew of these problems where the target audience are arguably underserved by our current NLP systems, and when we make the information available to people, they will come and look for it.
What advice would you give a new YI?
Tom: Reach out and actively foster collaborations both within and outside your lab, and gain experience in mentorship of students/interns.
Lucy: Most people starting this program probably don’t need this…but one piece of advice I’d give myself if I were to start again is to have a very clear vision of the outcome. It’s good to explore a little bit, but also to have a plan for getting where you want to be, and use the program to maximize that. Regardless, I just want to say that it’s been an amazing experience for me. The people at AI2 are incredible and I’m lucky to have gotten the chance to know and work with many of them, and I would encourage others to do that as well.
Learn more about the Young Investigator Program, and check out the open roles available at AI2.
Follow @allen_ai on Twitter and subscribe to the AI2 Newsletter to stay current on news and research coming out of AI2.
