Sangeet Khemlani

Navy Center for Applied Research in Artificial Intelligence, Naval Research Lab

http://mentalmodels.princeton.edu/skhemlani

http://www.nrl.navy.mil/aic/iss

What does your research focus on?

My research examines the mental representations and cognitive processes underlying deductive reasoning, creative thinking, and abductive explanations. A major challenge is to explain why people are predictably poor on some tasks, e.g., making certain deductions or estimating probabilities, but extraordinarily skilled at others, e.g., devising explanations. My collaborators and I think that the answer to this question is that mental simulations are the basis of high-level thinking.

What drew you to this line of research and why is it exciting to you?

Studies of logic and computer science: programming a computer is a way of making a hunk of metal do your thinking for you. But there’s a big gap between what current programs can reason about and the common sense thinking of everyday life. So, I investigate how people reason in order to narrow the gap between the two.

Who were/are your mentors or scientific influences?

Nobody’s had a bigger impact on the way I think and do research than Phil Johnson-Laird, who was my graduate advisor and remains a close collaborator and friend. Sam Glucksberg and Danny Oppenheimer were comparably instrumental in my graduate training, in no small part because of their contagious sense of humor. Greg Trafton, my postdoc advisor, maintains a very exciting interdisciplinary research environment. If I know anything about linguistics, it’s from talking to Adele Goldberg and Sarah-Jane Leslie. And I probably wouldn’t have considered research as a career without the encouragement and advice from Kostas Arkoudas, Selmer Bringsjord, Wayne Gray, Hans Neth, and Bram van Heuveln, my mentors as an undergraduate.

What’s your future research agenda?

I’ll continue to build and test theoretical and computational systems of how reasoners construct mental simulations to solve complex reasoning problems. One major project I’ve been working on is the development of mReasoner, a unified computational cognitive model of deductive, explanatory, and probabilistic reasoning. Another project is to study how reasoners collaborate with each other to solve problems. This project makes use of autonomous robots as programmable confederates in psychology studies.

What publication are you most proud of?

My recent paper, “The probabilities of unique events” in PLoS ONE in 2012. No theory at present can explain how people systematically estimate probabilities of unique, real events, such as the probability that the UK will leave the EU by 2020. The idea we proposed is that reasoners build analog magnitude representations to convert evidence — in the form of mental simulations — into probability estimates. The project synthesizes work from disparate fields such as developmental cognitive neuroscience, judgment and decision making, and reasoning, and it uses a novel experimental methodology to test the predictions of a computational model.