Interview with A/Prof Charles Kemp
You recently joined the School after a long period in the USA. What are your impressions of the University and the School after your long absence?
Many things have changed -- for example, buildings like Arts West and the School of Design are still unfamiliar to me. But other things are reassuringly similar, including Redmond Barry and the smell of eucalyptus along Tin Alley.
You did your undergraduate work in psychology and computer science here at the University of Melbourne, so you have had a sort of homecoming. You would have been no stranger to the Redmond Barry Building back then. Do you have any fond or less than fond recollections of the place and the people back then?
I'm grateful to this place for giving me my start in psychology. The class that stands out most in my memory is a class on motor control taught by Jeff Pressing. Taking that class helped me to see that that formal models could lead to real progress in cognitive science. I also came away full of admiration for Jeff, who seemed to excel in at least three different areas -- psychology research, music, and using AI to invest on the stock market.
You spent many years at Carnegie Mellon University, a renowned centre for computer science, cognitive science, psychology, philosophy and economics. That must have been a remarkable intellectual environment for your first academic position.
CMU was a great fit for my interests because it places such a strong emphasis on computational work. At one stage I realized that there were seven different departments with people doing computational work that closely connected to mine. In addition to the five areas that you mentioned, CMU has programs in decision science and human-computer interaction that both have strengths in computational cognitive science.
Your hiring adds real firepower to the already fearsome research strength within the Complex Human Data Hub. You are playing a major role in the terrific new "Complex Human Data Summer School". How else do you anticipate contributing to the work of the hub?
I'm still trying to figure out how this might work, but I'd love to help make the field of Cognitive Science a bit more visible on campus. This might involve strengthening our connections with computer science and with other places on campus where people are doing relevant work.
One thing you must be happy about is reconnecting with Amy Perfors, deputy director of the hub and an old grad school buddy of yours from MIT. Do you expect to collaborate with Amy, and do you have any amusing and shareable anecdotes about her from student days?
I'm very happy to be just down the hall from Amy and definitely hope to collaborate with her and other members of the Complex Human Data Hub. She and I are opposites in some ways -- when we were students I was boringly regular in my habits, but Amy would do things like deliberately wearing mismatching shoes just to keep life interesting. But we agree on all of the important things and I think that we work well together.
Looking through your CV it is clear that your work pays no heed to the traditional boundaries of academic psychology. You have written extensively within the interdisciplinary matrix of cognitive science and recently published an article in the prestigious Annual Review of Linguistics. What allows you work across these fields and what are the benefits and pleasures of doing so?
I'm often motivated by trying to understand whether and how people's behavior is consistent with computational principles like efficient coding and Bayesian inference. Because these principles are relatively general they can be applied to many different domains, including reasoning, language, vision, and even areas like cultural evolution and economic behavior. I don't work in all of these areas but I do enjoy thinking about how far a set of general principles might extend.
I am blown away by the originality of your work on colour concepts across languages. Can you describe what you did in that work for our readers, and explain why it matters?
Noga Zaslavsky really deserves the credit for that project. She developed a model that helps to explain why systems of color terms vary across languages --- for example, some languages have just two basic color terms, but English has eleven including red, blue, green, and so on. The part I like best is that the model explains how color-naming systems could change from simple (e.g. two terms) complex (e.g. 11 or more terms) by following a continuous trajectory. You can see this trajectory in a video she made:
Previous researchers had talked about how color naming systems change over time, but until Noga's work it was hard to understand how these changes could be generated by a smooth, gradual process.
There appears to be a genuine curiosity about social processes in your work, which is arguably unusual for a cognitive scientist. Your work on colour naming and kinship categories addresses the communicative function of language, and you have written a lot about how we reason about the preferences, choices and relationships of others. How do you see this work contributing to research that is usually tackled by people with social science backgrounds?
My hope is that some of the computational tools that I and others use allow existing ideas from the social literature to be formalized and tested in new ways. The best example from my own research comes from a project that I worked on with Alan Jern, a former student. We were working with a standard economic model for inferring people's preferences, and at some stage we realized that this model captured many key aspects of correspondent inference theory, a classic approach from social psychology. As far as I know that connection had not previously been made.
A student came up to me recently after my first year lecture and said she wanted to study both computer science and psychology, but wasn't sure if she could and thought she might have to choose. I told her she should study both fields, mentioned you and your Hub, and said that there was no better place In Australia to study psychology from a computational perspective than here. What else should I have said?
I think that's a great answer. I might expand slightly and say not only that it is possible to study both computer science and psychology, but that it's important for at least some people to develop a deep knowledge of both fields. I already mentioned a couple of examples of how my own psychological work draws on computational ideas, but there are also good examples of how ideas from psychology have moved computer science forward. My favourite recent example is AlphaGo, the system that managed to beat the best Go player in the world. AlphaGo and similar systems draw directly on ideas that were originally developed by people who were interested in classical conditioning and animal learning.
Finally, is there any recent paper you'd like to share with the readers? What makes it special to you?
I'll pick a paper called Semantic typology and efficient communication that I wrote together with Terry Regier and Yang Xu. Working on this paper was fun because we got to lay out a general framework for thinking about color terms, kinship terms, and many other kinds of categories picked out by natural languages.
Thank you, Charles.