Tianmin Shu joins the Johns Hopkins University as an assistant professor of computer science. He is jointly appointed in the Krieger School’s Department of Cognitive Science.
Shu received his PhD in statistics from the University of California, Los Angeles. Before joining Hopkins, he was a postdoctoral associate and then a research scientist at the Massachusetts Institute of Technology.
Tell us a little bit about your research.
I work on machine social intelligence. In particular, I focus on engineering fundamental building blocks of social intelligence for assistive AI systems, including the ability to understand and interact with humans in rich and complex ways—just like how humans understand and interact with one another. I am also interested in computational models of social cognition and how those models can help us reverse-engineer human social intelligence for AI systems.
Tell us about a project you are excited about.
One of the basic social skills that people have is called “Theory of Mind,” or the capability to reason about other people’s hidden mental variables—such as goals, beliefs, desires, and emotions—from their observable behaviors. One of my long-term efforts is to develop machine learning models for open-ended machine Theory of Mind that can understand humans’ mental states from multimodal inputs in real-world settings. Such machine Theory of Mind models can be used for not only interpreting human behavior, but also guiding assistive AI agents (such as home robots or a virtual assistants) to better help humans based on their personal preferences and needs.
Why this? What drives your passion for your field?
I’d like to imagine a future where AI systems can help humans in a wide range of scenarios, from daily chores to health care to education. I believe that it is not enough to have AI systems that can perform complex tasks—we also have to make sure that these systems can truly understand what humans need and prefer, and help them in ways in which they want to be assisted. For instance, voice assistants such as Alexa will not help users unless they give explicit, short commands; they also cannot engage in long, meaningful conversations with users to figure out how they want to be helped. Failing to understand humans’ minds and interact with them safely and productively can be counterproductive or even detrimental, particularly with increasingly powerful AI systems. We do not want to have physically capable robots running around in our living environments without knowing how to interact with us. So I believe that human-level machine social intelligence is a fundamental ingredient for engineering socially intelligent AI systems that can offer successful and flexible assistance to humans in the real world.
What classes are you teaching?
I’m teaching Cognitive AI, a new course that connects cognitive science and AI. In this course, I use insights from classic and recent cognitive studies to motivate what kinds of humanlike AI capabilities we should try to build. I then discuss a broad set of computational tools (such as probabilistic inference, neuro-symbolic methods, planning, and reinforcement learning) that can allow us to build humanlike AI systems that can reason about and interact with the world and other agents.
Why are you excited to be joining the Johns Hopkins Department of Computer Science?
I believe that the study of social intelligence can really benefit from interdisciplinary collaboration. JHU—both within CS and across its various other departments and schools—has many intellectual communities that I am very excited about and want to engage with. To me, the Johns Hopkins Department of Computer Science is certainly one of the best places to build an interdisciplinary research program for social intelligence.
Besides your work, what are some of your other hobbies and passions?
In my free time, I enjoy reading—particularly sci-fi—and listening to my growing collection of vinyl records, mostly classic rock and jazz. I like trying out new restaurants or new recipes, though I don’t get as much time to do these as I’d like nowadays. I also like hiking and look forward to exploring the trails around Baltimore.