As part of our ongoing AI In your Community series, I sat down with Dr Rada Mihalcea, a computer scientist and Director of the AI lab at the University of Michigan, known for her work in Natural Language Processing and Computational Social Sciences (including computational humor!).
Rada Mihalcea
Tara Chklovski: What problem are you trying to solve?
Rada Mihalcea: I work in Artificial Intelligence, and then within that, Natural Language Processing (NLP) which is about building computer programs that deal with language. A traditional NLP example (that I am not working on), is machine translation — through which text in one language, say English, is automatically translated into another, say French. What I actually work on is computational sociolinguistics, which has to do with understanding people, their emotions or sentiments, from the texts they write. For instance, one of the systems that we are working on could process a document that you write, and it will figure out what your sentiment is — whether it’s positive or negative, or if it is expressing joy or fear or anger. In another project that is being done here at Michigan in collaboration with psychologists at the University of Texas, we are trying to understand people’s values by computationally analyzing their language.
For instance, based on what they write, can we tell what third graders care about, or what their mothers care about. Is family a high value for them, or friends, or achievement, or religion and so forth. We then explore how these values are reflected into behaviors.
So for instance, what does it mean for a person to say they are religious. Do they go to church on Sunday? Or are they good to others? How is this value reflected in daily behavior? In addition we are doing cross-cultural comparisons, trying to understand how different groups of people — such as people in the US or people in India — would have different values, and how those values are reflected in their daily behaviors.
TC: How do you make the connection to the behaviors?
RM: It’s all through language extracted from social media. We’re building tools that automatically extract behaviors like “I’m having lunch with my spouse” or “I’m at the activity with my children”. But language is very rich, and the same behavior can be expressed in many different ways. In another project, we are developing algorithms to identify and group similar behaviors. For instance, “I’m having a meal with my spouse” or “I’m having dinner with my husband” are different ways of saying the same thing. We are trying to understand how these are similar clusters, so that we can better map them to values.
Another project we are working on is in collaboration with the School of Public Health where we are developing an NLP system that can automatically evaluate and identify the characteristics of good and bad counseling. One step in this process (done through a meta-analysis) is to define who a good counselor is. For instance she may ask many questions, help you reflect, and so forth. We do this kind of analysis automatically so we can provide feedback to the counselors, and also to their supervisors.
We then look at actual outcomes to determine what successful counseling is. An example of successful counseling could be when the counselor was successful in persuading a person to quit smoking. There are of course other reasons why a person may quit smoking, but we use this outcome as a proxy.
The real power of this system is that it is augmenting people. For instance, if I were a counselor, my effectiveness would be really augmented if I had a tool that would bring the expertise from 1,000 other counselors and would recommend saying “x,y,z…” when you hear “Okay, but you know…”.
Another collaborative project we are working on is in collaboration with Toyota, and aims to recognize the state the user is in.The programs we are working on will allow an AI agent to be user-aware, by accurately sensing and interpreting the user state based on an analysis of the user’s language, facial expressions, and physiological data such as temperature, heartbeat, respiration rate, perspiration, etc. The goal is to determine whether the user is experiencing positive or negative affect, stress, or discomfort, just by observing the person and not by asking them to fill out a survey!
TC: You’ve shared some very interesting and wide ranging set of problems that you are tackling! What advice would you give children and families as they try to find a good problem in the community to solve, with technology?
RM: It’s not easy! But you can look around and see what problems people are running into and how many people are affected by it. I also think it’s very important for the person who’s finding a solution, to feel connected to that. For instance, politics is for me a field that I just cannot connect to. It would not be a good idea for me to start solving a problem in politics with AI, because I don’t feel passionate about it.
So it is important you find a problem that matters to the community, and also one that speaks to you. Then even if you don’t have the skills, you will have the motivation to build those skills and to learn as you go.
Finally, it’s also about looking under the hood, and that takes a bit of training. For instance, at a first glance, certain things may look like problems, but if you analyze them, you may find that there is another underlying problem that generates them. An example could be if you see running water on the kitchen counter and you move to wipe the water off. But the real problem may be that there is an underlying pipe that has burst that needs to be fixed.
TC: What inspires you?
RM: I am inspired by people. I like observing people in general, and I also do that in my research. A lot of the research that I do, although it is core computer science and language processing, it is really about people. I realized only recently that I can also do in research what I liked all along: people, their behaviors, and observing what they do. I particularly like observing children; my own and others. I like observing their pure behaviors that are unaltered by society constraints out there.
TC: What do you find difficult?
RM: I have a little bit of an obsession with time — both time management and time perception. Time is something that is limited, no matter what we do, how we do it, where we look. It’s still the same. No matter who you are, where we are, it’s still the same 24 hour blocks for everyone. That’s something I find difficult. I like reading a lot about time and thinking about time. Maybe I observe time more than others would! But that is something I would like to have a better handle on. For instance, in my professional life, I could do even more if I were to have more time. Of course I make up for that by working with many other people, but at the end of the day I still have just 24 hours. And in those 24 hours I want to also do things with my kids and family, travel, read, or hang out by myself in a coffee shop. Time is limited and I find that challenging!
TC: You mentioned augmenting our limited grasp of time by working in teams. How do you think AI can help us?
RM: It can help by augmenting people. Many years ago there was a wrong view that AI was going to replace people. Some of that view is still lingering around. And then we ran into the “AI winter”, when people lost trust in AI, and felt disappointment in what AI was able to do. Now we are back on the wave and going forward fast. To me AI is really about augmenting people. It can make a huge difference in how individuals and communities work. For instance, with counseling, it’s not about building a machine that will be your next counselor, but rather, having access to a world-class counselor because he would be augmented with an AI agent that leverages the abilities and knowledge from the world’s best counselors. The same would be for education, health, finance, transportation, legal systems etc. We would be able to look almost anywhere around us and augment people to do what they do even better.
TC: How can children learn more about AI?
RM: By reaching out and talking to people teaching AI, working on AI products in companies, or by talking to the participants from programs like the Curiosity Machine AI Family Challenge. The children and families that participate in the AI Challenge will themselves become part of the group of people who know what AI means and how to use it for good. To me, this kind of networking among people who are actually doing something in AI is the best way of learning, as opposed to reading some news that talk about what AI does.
TC: Thank you Rada for sharing your research, your insights, passion and challenges with us! Maybe your words will inspire a child to tackle the hardest problem of all – time itself!
