Bhavana Dalvi is a Lead Research Scientist on AI2’s Aristo team.

I used to be introduced to the fascinating world of machine learning (ML) for the primary time during my master’s studies on the Indian Institute of Technology (IIT), Bombay. After graduating from IITB, I had a wonderful opportunity to work at Google. Here, I witnessed ML’s effectiveness at solving several real-world problems. I felt inspired to learn more and do something recent on this area, so I applied for Ph.D. programs. I used to be fortunate to have the chance to do my graduate studies at Carnegie Mellon University, specializing in language technologies. In the ultimate yr of my Ph.D., I used to be presenting my work at Automatic Knowledge Base Construction (AKBC workshop) at NeurIPS, where I met Peter Clark. He told me in regards to the excellent research opportunities at AI2. I joined the Aristo team in 2015 and have thoroughly enjoyed working on various research projects ever since.

Advances in large language models (LLMs), e.g. ChatGPT, have left everyone in awe to such an extent that individuals have began feeling existential crisis. Nevertheless, recent studies have also shown that there are still many open challenges to enable AI models to take heed to human feedback and produce trustworthy outputs and their justifications. A few our latest projects at Aristo explore: 1) How can one construct a teachable reasoning system where models can improve with human feedback? 2) Do these LLMs have coherent mental models of the world around them? In these and plenty of other projects at AI2, we are attempting to make use of the LLMs as a tool to construct intelligent systems that may take AI to greater heights.

I’m looking forward to advancing machine reasoning systems to enhance over time by producing systematic explanations, incorporating human feedback, and assimilating recent information. I’m super excited in regards to the NL Reasoning and Structured Explanations workshop that I’m co-organizing at ACL 2023. This workshop goals to bring together a various set of perspectives from different research areas: Explainable machine learning (NLP), Classical AI (especially theorem proving), and Cognitive science (how do humans structure explanations?). It might help us establish common ground for a way these various sorts of explanation structures can tackle a broad class of reasoning problems in natural language and beyond.

“You gotta judge a person by his principles.”- Rick Ross.

Once I visited AI2 for an interview, I very much liked talking to phenomenal researchers in the course of the interview process. One other thing that caught my attention was a set of quotes displayed throughout AI2 partitions. During my seven years of research experience at AI2, I can see that the principles mentioned in those quotes are an integral a part of AI2 culture. Considered one of the quotes seems most appropriate here as advice to an aspiring research scientist:

“While working on cutting-edge research, failure is certainly an option, though failure to measure and analyze our progress isn’t.”

Setting a long-term research agenda, complementing it with short-term goals, measuring progress continuously, and doing self-reflection to make adjustments before later may also help us pursue excellence in our work.

There are countless beautiful beaches along the pacific coast which can be inside driving distance from Seattle. Our most favorite is South Beach in Olympic National Park. It’s a comparatively less popular beach and in case you visit during a not-so-busy time, you get the whole beach for yourself.

Our next favorite beach is in Fort Ebey State Park Situated on the western side of the stunning Whidbey Island. This place offers breathtaking views which can be an ideal combination of sun, sand, and stunning scenery.

Pandemic years had unique challenges: suddenly adapting to work-from-home, minimal social interactions and entertaining a toddler all by ourselves. But every challenge brings a growth opportunity. I all the time desired to do yoga/meditation and it was hard to seek out time with a baby and attend events in-person. Through the pandemic, everyone switched to online mode. So I could attend many online workshops and yoga is back in my regular day-to-day life. Toddlers love sensory play, so I learnt baking: breads, cookies, cupcakes, you name it. My daughter likes to assist me, she would mold the cookies in several shapes, add sprinkles, and entertain me with songs. It’s a win-win activity for each of us.