Graduate Workshop in Computational Social Science
Meet Bill Rand our fall 2015 SFI alumni featured profile. Bill is an Assistant Professor of Marketing and Computer Science and the Director of the Center for Complexity in Business at the University of Maryland's Robert H. Smith School of Business. He attended the GWCSS and CSSS courses at SFI.
We asked Bill the following questions about his experience at SFI and his research interests.
1. What do you know now that you wished you knew when you attended the GWCSS and CSSS courses at SFI and/or what did you learn that has helped you most in your career?
One thing that I wish I would have known when I attended both the GWCSS and CSSS courses at SFI is how unique an environment these schools create. In academia, we often talk about how important multidisciplinary work is, and how we wish that our students could have a broader education outside of their own fields. In reality, it is fairly rare for a student to be exposed to so many different fields and methods, and to have the opportunity to collaborate with students in many different disciplines. The SFI courses are unique in that they provide a very open experience, one where students have the opportunity to learn from experts in fields that are very different than their own, and to apply what they have learned to any problem that they want to explore. In the two summers that I was at SFI I tried to drink up as much of this fount of knowledge as I could, but I always had my own dissertation in the back of my mind. I think that if I had realized at the time how unique the experience was, I might have been even more open to exploring new topics.
On a related note, I felt that I learned almost as much from my fellow students during long, rambling conversations on science, academics, and life, as I did from the SFI instructors. Moreover, these conversations did not happen in the classroom, but instead took place during long hikes in the mountains around Santa Fe, over beers at El Farol and the Dragon Room, and while learning how to play Go in the cafeteria. These are some of my most treasured memories from SFI. One thing I learned at SFI, that I treasure to this day, is the ability to be very open to approaches and thinking outside my own. So much of academia is focused on pursuing one particular question and one particular method to answering that question, at SFI it became clear that there was an alternative approach to research that involved taking a problem and applying every method science had to understanding that problem, or taking a method and seeing how many different problems you could solve with it. Moreover, not only is this a viable approach to science, but I find that it is an approach embraced by people in many walks of life.
2. What are your primary research interests and what are the main challenges for you right now?
My main research focuses on diffusion of information. I am interested in how people discover information, how they act on and how they transmit information. There are many practical questions that revolve around the diffusion of information:
- How do buyers decide which products to purchase, when the state of those markets affects their decisions?
- How do citizens choose to form legislative institutions especially when those institutions affect the choices of those individuals?
- How do individuals decide what to tweet about or post on their Facebook page, when often what they are reading and interested in is at least in part determined by other individuals’ tweets and postings?
With respect to that last question, digital social media platforms have revolutionized the way that we communicate, work, and gather information. These systems are powerful enough to at least partially contribute to dramatic political events (the “Arab Spring”, #BlackLivesMatter, Occupy Wall Street, etc.), provide information during natural disasters (the Fukushima tsunami, Hurricane Sandy, flooding and wildfires in the West, etc.), and affect companies’ reputations and financial issues (early rumors about Steve Jobs’ death, the faked AP story about the White House being attacked, etc.). However, social media as a communications channel can also be difficult to understand, predict and use. Since social media data is largely unstructured and the amount of data is vast, it can be hard for any interested individual to grasp all of the information about a particular event that is relevant to them. Moreover, though powerful ideas and concepts can come out of social media, there is a very low signal-to-noise ratio, since much of social media is devoted to personal and narcissistic communication. I am fundamentally interested in how interesting events emerge from the noise on social media, and how those patterns feedback to affect individual decisions and even their beliefs about the world.
It is often difficult to explore the question of information diffusion within large data, because so much of how information diffuses requires modeling at a very fine-grained level of resolution and understanding how individuals are making decisions even in the context where those individuals can be very different from each other. Thus, to explore these questions requires a toolkit that is geared toward understanding individual-level behavioral patterns. As a result, for my studies of information diffusion, I rely upon the methods of complex systems. In particular, there are two methodologies that I routinely employ, agent-based modeling (ABM) and machine learning.
The main challenge I am wrestling with right now is how to build individual-level models from large-scale data. To carry out this work, I am extending two strains of research that I first learned about at the SFI. The first strain is work by John Miller on Active Nonlinear Testing and parameter optimization of computational models using machine learning. I have been looking at ways to automatically calibrate agent-based models to empirical data in an effort to create models that are automatically validated against real-world data. This work started while I was at Northwestern University as a postdoctoral fellow in the Northwestern Institute of Complex Systems (NICO) working with Uri Wilensky and Forrest Stonedahl. I assisted Forrest as he developed BehaviorSearch, a powerful add-on to the NetLogo programming language (a very popular agent-based modeling toolkit). BehaviorSearch, among other things, gives you the ability to specify a fitness function that evaluates the match between empirical data and model data, and then uses machine learning methods to find the best match. We have used this tool and this approach to explore questions such as:
- Who are the optimal people to incentivize in a social network to maximize the spread of a particular product?
- How do users surf the web for news?
- How does information spread during a crisis?
However, there are still many open research questions around this method. For instance, it is not clear what the best fitness function to use is for any particular application or what features to compare between models and the real world if we want a model that we can use for forecasting the effect of interventions. The second strain of research builds upon Jim Crutchfield and Cosma Shalizi’s work on computational mechanics and causal state / epsilon machines. Their approach provides a tool that can infer the minimally complex, maximally predictive model from any times series. The output of this process is a causal state machine that can then be embedded directly into an agent-based model and used to make predictions about how the system will evolve over time. Our group was the first to apply these methods to social data, and we have shown that it can make accurate predictions about individuals on Twitter and the optimal time to tweet if you want to maximize retweets. Interestingly enough I am carrying out this research with Michelle Girvan, an SFI alum, and David Darmon, who just attended the SFI CSSS last summer
3. As Director of the Center for Complexity in Business at the University of Maryland what are your most important task?
My most important task is the development of a field of complex systems for business and management science. To support this endeavor, we have enabled a variety of activities. For instance, we have created two different conferences on complex systems in business. The first is an academically-focused conference held every fall in downtown DC that is entering its 7th year. This conference brings together academics from all over the world who are interested in applying complex systems to problems in business and management science. The second conference is a practitioner facing conference, where we discuss the intersection of complex systems and data science. Data science is related in many ways to complex systems. Both can take advantage of large amounts of data, both regularly use computational methods, and both seek to understand patterns of behavior.
In addition, to the conferences we pursue research grants and help other faculty pursue research grants that are focused on the application of complex systems to business and management science. We often put together interdisciplinary teams of students, professors, and researchers to explore unique and different topics as a result of these grants, or in the writing of a grant. In this context, the Center often serves as a home for the researchers.
The Center also reaches out to corporate partners who are interested in applying the methods of complex systems in their own work. Over the years we have worked with established organizations, such as Teradata, Mars, National Geographic, and the American Red Cross, and startup organizations, such as Vibeffect, an innovative company in the DC area attempting to use complex systems methods to help students and parents make more sense out of the educational system.
Finally, as the Director of the CCB, I often am asked to give talks about the role of complex systems research, and to provide workshops on complex systems methods. Most recently, my textbook with Uri Wilensky on Agent-Based Modeling (ABM) was published by MIT Press and I have been providing workshops on ABM and complex systems based on this textbook.
4. What mark do you want to leave on the world?
This is a question that I have been thinking a lot about in the recently. My thesis advisor and long-time personal supporter, John Holland, just recently passed away. Prof. Holland was a major reason I am where I am today, and he left an indelible mark on everyone he touched. As the SFI memoriam about Prof. Holland noted, he was fond of saying: "I have more ideas than I can ever follow up on in a lifetime, so I never worry if someone steals an idea from me.” Prof. Holland left us with more ideas than any of us could ever fully explore in a lifetime, and I think that is a notable goal to strive for.
I certainly hope that I leave behind enough knowledge to help advance scientific progress at least incrementally. However, though I pursue my own research to answer questions that I think are important, I also hope to inspire others to pursue different and new questions. I want to leave this world having answered some important questions, but I also want to leave more questions behind for other people to answer; maybe questions that we haven’t yet thought of, or questions we couldn’t even answer without new methods and tools.
5. What interests do you have that might surprise your colleagues?
I am a home brewer with the hope of someday becoming a professional brewer, or at least founding a brewpub. I am data driven and scientific, and beer brewing has an art to it, and I love the idea of trying to combine that art with science. I already take detailed notes on all my beer recipes and brewing attempts; recording things like the pH levels of my mash (beer ingredients) and water, temperatures all along the way, and the specific gravities of the beer in process. I use this information to attempt to perfect my recipes.
Recently, I have become fascinated by lambic-style beers, which is a sour beer style from Belgium. These beers need to age for years in order to reach the right taste profile. This December I will create my first geuze, which is a blend of different lambics. In my case, I will be using a three-year old, two-year old, and one-year old lambic. In some ways, this is probably the longest scientific experiment I have ever run! I would love to create a brewpub founded around a data driven approach to beer making.