December 3, 2009
Kevin S. McCurley is a research scientist at Google Research, where he has worked since 2005. He previously held positions at IBM Almaden Research Center, Sandia National Laboratories, and University of Southern California. He has published in the areas of information retrieval, algorithms, parallel computing, cryptography, and number theory.
Ivars Peterson: What do you do at Google Research?
Kevin McCurley: Research there is a little different than other places where I've worked. It's a lot closer to product. This doesn't mean that my freedom is restricted in what I do. It's just that the ability to transfer things to product is so ridiculously easy. I previously had worked in industry at Sandia National Laboratories and IBM research. Those places felt very different.
Basically, I would say that my job is to invent the future. So I spend time trying to work with groups where I think I have expertise and can help advance things. I've worked on search and ads and some stuff for Android. A lot of it involves software.
I love writing software. When I was a child I built model planes and cars, and I loved to craft things with my hands. I decided at a certain point [later in my career] that the reason I continued to write software is because it's an extension of my feeling of pride in craftsmanship in building something. So writing software is very much a part of my day-to-day work. I think part of it is that I can write software that actually gets used by humans. That gives you a nice feeling.
IP: What are you areas of expertise?
KM: I have described myself in the past as intellectually promiscuous, which is perhaps not the right phrase. I go looking for interesting problems to solve. Mathematicians often characterize themselves as either theory builders or problem solvers. I think [Paul] Erdös was a prime example of a problem solver. I definitely fall into that camp. So I go look for problems that are interesting for the company to solve and at how I can bring mathematics to bear on them.
What I have observed about my work in the past is that I am fearless about reading mathematics in a field that I have never looked at before and have no basis for understanding. A few years ago, for example, I became interested in mathematical economics. I found that there are some very big issues in economics that concern our work at Google. We are directing the flow of money from advertisers to content owners and from advertisers to ourselves—the matching of advertisers to humans. A lot of this is, at its core, economics and mathematical modeling.
I've also been spending time recently reading about mathematical psychology. A lot of research at Google is about how the mind works, and many people work on machine learning, which grew out of artificial intelligence. The primary reason is that we deal with human interaction with information, so we have to understand how humans cope with information, how they consume it, how they produce it, how they transmit it.
We have a chief economist who recently discovered that he could predict unemployment claims before the federal government could. It turns out that [the rate of new unemployment claims] is the best leading economic indicator for the end of the recession.
A person usually loses their job on a Friday. The first thing they do is go home and tell their spouse. The second thing they do is probably get drunk. And the third thing they do is go to Google and type in "unemployment insurance." They don't get to the state unemployment office until Monday, and the state unemployment office doesn't report it to the state or the federal government until some weeks later. So we have a very big lead time in discovering the likely rate of new unemployment claims. It's quite odd what you can learn from the query stream.
IP: The tendency to go first to a search engine tells you something about how human behavior has changed in recent years.
KM: Librarians traditionally served the role of helping people navigate through a collection and find answers to their questions. Frankly, when I was an undergraduate student, it was pretty painful to find answers to questions.
It is still very painful. A couple of years ago I had a question about diffusion in metric spaces: When two points are close in a metric space and you map them to another metric space, when do they become far apart? And I was thinking, "How do I search the literature for this?" I realized we still don't understand a lot about search. You have to have a mental model of what the answer will look like in order to phrase your question about it. I had a pretty precise idea of what I was looking for—certain kinds of information about certain functions—but I did not know where to begin.
The usual model is you go to an oracle. You ask a mathematician, someone who has knowledge in the field, maybe published a paper about metric spaces.
IP: You started out as a mathematician.
KM: My Ph.D. was in analytic number theory, which I did at the University of Illinois in Urbana. I also did a master's in statistics there.
IP: When did your interest in mathematics arise?
KM: I think that, for many people, their first realization of their interest in mathematics is when they discover in school that they're good at it. It was a subject that I found easy to grasp. I went to a public high school in San Jose, California. I had a teacher, Judy Jones, who was influential by showing me that mathematics really could be fun and interesting. She encouraged me to go to high school mathematics contests. When I went to college, I decided I wanted to be a math major.
In my first year at Santa Clara University, I had a professor in calculus, Jerry Alexanderson, a former president of the MAA, who was also a very inspiring teacher. He told interesting stories; he could relate the mathematics to history and to activities that were meaningful. That fueled my interest. Thereafter, I consumed everything I could in mathematics.
IP: At some point, you made a transition to computer science.
KM: It was partly fueled by the fact that I went to high school in Silicon Valley. As an undergraduate, I took classes in computer science in the computer engineering department, and I really enjoyed them. That was when I first noticed my interest in craftsmanship of computer programs. But it was a side activity. When I did my Ph.D., however, part of my dissertation [in number theory] was on calculations involving zeros of L functions.
I still have a love for number theory, though the natural transition for me was to start working on cryptography. After I had a post-doc at Michigan State, I took a job at USC in Los Angeles, in part because I wanted to work with Len Adleman, who was the "A" of RSA. It worked out quite well because he wanted to learn more number theory. So we wrote some papers together.
We worked on algorithms and complexity in number theory. One of the great problems in theoretical computer science is the P = NP problem, a question about what problems are hard to compute. If you make that specific to an area like number theory, you can tap into a rich history of trying to figure out what problems are easy and what problems are hard. Gauss worked, for example, on trying to determine which numbers are prime and on factoring numbers, and he wrote quite a bit about algorithms. That was an easy field for me to move into.
At the end of a few years, it became clear to me that my future happiness was probably going to be in computer science more than mathematics. I went off the deep end into cryptography and pursued that for a while.
IP: Did you do cryptography at Sandia?
KM: That's the reason I went to Sandia. They had a good cryptography group there. I applied for a job by sending a resume to Gus Simmons, and he forwarded it to a colleague, Ernie Brickell, who was starting a new group there. So I got hired into that group. I went there primarily to work on the application of number theory to cryptography. At one point I worked on a better algorithm for exponentiation because that was a core operation in both the Diffie-Hellman key exchange protocol and in the RSA algorithm, and I found some interesting mathematics to work on. We built a discrete mathematics and cryptography group, which thrived for a few years.
I then decided to work on parallel computing, in part because there were interesting questions about the fundamentals of computation. I worked on that for a while, but I finally decided in 1996 that the Internet was really changing things at that point. That's when I moved to IBM Research in California. I worked on digital rights management. It was the time when the music industry was coming unglued over the problem of intellectual property. I looked at these problems and tried to work on them, wrote a few patents, worked with a product group there, and came up with some interesting ideas.
What I find in scientific research is that the people you work with are as important as or even more important than the problems you work on. I think there are lots of interesting problems in the world, and I think of science as primarily a social activity. I really enjoy the interaction with my colleagues. That directs me a lot of times to the problems that I find interesting. So [at IBM] I became interested in problems of Web structure and information retrieval.
IP: How did you end up at Google?
KM: I was president of IACR, the International Association for Cryptologic Research, and while I was president, I started paying attention to what was happening to our publications. I became very interested in the fact that our publications were kind of scattered and not well archived. So I decided I was going to build an electronic repository of our publications—a searchable repository. So I did a partnership with Springer-Verlag, and they had people do some optical character recognition on our publications and produce files representing what was in the mathematics papers.
You can imagine the process of trying to do optical character recognition on mathematics. It's highly error-prone because there's not enough redundancy, and there are very strange alphabet changes and symbols mixed in.
We ended up with something that you could never show to the user, but you could use it to construct a search engine, which I did, knowing essentially nothing about search. But that was the action that got me from cryptography to working on search. I was interested in preserving the publications of the cryptography society, and as I thought about how to preserve the value of these for the long term, I thought students should be able to search these. That's what led me to do a couple of interesting Web and search projects at IBM.
Of course, since I live in Silicon Valley, it became pretty clear after a while that it would make more sense to move to Google if I wanted to continue working on search.
Also, there's this term "seven-year itch," which people usually apply to relationships. But I believe it is actually a good description for how you should think about a job. After seven years, you're likely to stagnate a little bit. If you're going to keep advancing and learning new things, professors get a sabbatical after a while. They go someplace else to have a new set of colleagues, a new set of problems thrown at them, a new set of solutions thrown at them. I felt it was probably time for me to do something different. So, four years ago, I moved to Google after being at IBM almost eight years.
IP: So you have three years to go.
KM: It's actually hard to imagine leaving Google. It's such a wonderful place to work.
IP: You've been working on a paper about "attention."
KM: I wanted to understand how the attention of people gets distributed across the different available sources. This is of interest to Google's business because we need to understand where the eyeballs are and what people are searching for. But it's also interesting to understand the nature of how people produce and consume information. It turns out to be related to this question that economists are interested in: How can income or wealth be distributed to a population? The correlation is that attention corresponds to value for display advertising. If attention is present, that means that display advertising can be seen by a consumer. It makes sense to use economic terms to value attention. Several people have studied the concept of the attention economy.
We're looking at this again to try to explain why things come and go so quickly. We have these memes in our society that emerge so quickly sometimes, and then fade from existence. There’ll be an event, people talk about it, and then it's gone.
IP: Other projects?
KM: My other main project is highly software-related and involves something that I noticed a couple of years ago. The Web changed people's relationship to information by making it universally accessible. All of a sudden you could find information at your fingertips about health, about who you're dating tonight, or about politics, business.
Now that phones are becoming so smart and taking over the roles of computers, that will change people's relationship to information again. When you carry information with you, it changes your relationship to it. So that has become an interest of mine lately. How can you make information readily available on phones? That's why I've been working on Android, an operating system for phones.
I've also recently been having discussions with people at work about ad quality, improving the relevance of ads, which is not the sort of thing that mathematicians would think they are qualified to work on. But mathematicians are good at quantifying things—quantifying value, attention, content of information.
We have some interesting problems in this area. So if I can make a 1 percent improvement in the quality of our ads, that can translate into a huge value for my company. So if mathematics can do that, that's great.
IP: Where do you see yourself 10 years from now?
KM: I wish I could answer that question. It's hard to say. The one thing that I'm quite sure of is that mathematics will be a part of it. If I look back on how I have progressed over 10-year periods, it seems to be quite random. It's a random walk through different intellectual pursuits. If I had to predict, it would probably be on some aspect of social sciences and mathematics, given the direction I'm heading in right now. There are great opportunities for us to improve the human condition through modeling of social interactions and human interactions.
I don't tie my future expectations to who or what my employer will be or what kind of work I will be doing, but I do think that mathematics will be a part of it because that's what I enjoy—trying to find value through mathematics.
Read about Kevin McCurley's Distinguished Lecture: "Modeling Similarity in the Age of Data"
Listen to the full lecture.
Papers by Kevin McCurley
MAA Distinguished Lecture Series