Gregory Andrews [pictured left] was not one of the handful of pioneers who, trained in other disciplines, gave birth to computer science. Yet, he was on hand soon after to receive the squalling infant and, along with many others of his generation, to coax it toward maturity. Andrews is now a "rotator" at the National Science Foundation (NSF), serving as the director of the Division of Computer and Network Systems. I spoke to him about the current state and the future promise of research careers in computer science.
Andrews was an undergraduate at Stanford when its computer science department was formed. His first computer science course was taught by Stanford's first computer science Ph.D. Andrews received his bachelor's in mathematics from Stanford in 1969--Stanford didn't offer an undergraduate CS major until the early 1970s--and earned his Ph.D. from the University of Washington in 1974. He immediately joined the faculty of Cornell University. In 1979, he moved to the University of Arizona, where he's been on the faculty ever since, serving as department chair from 1986 through 1993.
Andrews has received two distinguished teaching awards from the University of Arizona, including a career award. He has served on editorial boards and on the board of directors of the Computing Research Association, computer science's research-focused professional organization. He is the author or co-author of three books and numerous journal articles. He is an avid golfer, proud of the fact that his handicap is smaller than his shoe size.
Next Wave: Thirty years as a faculty member; that takes you almost back to the beginning of this field, doesn't it?
Gregory Andrews: I was learning from the people at the beginning of the field.
Next Wave: How are career opportunities in computer science today, compared with how they've been in the recent past? Are they better or worse?
G.A. The Computing Research Association has for 30 years had an annual survey of the production and employment of Ph.D.s in computer science, called the Taulbee survey. The last few years' worth are all online and freely available. [Here's a link to the most recent survey.]
Next Wave: Looking at Figure 2, it looks as if there's been a general downward trend over the last 10 years or so in the production of Ph.D.s.
G.A. Yeah, but [it's not a dramatic reduction]. We were flat for the first 25 or 30 years of our existence, at 200, and then it started to increase in the 1980s as we became more of an experimental science and started to grow. The last few years have been down a bit, it isn't bad.
The number of people in graduate school now is up; however, at the other end of the pipeline, undergraduate enrollments are down the last few years, and that's a direct reflection of the economy.
Next Wave: Is a Ph.D. in computer science just for academic careers, or are there industrial positions, too?
G.A. That's the dramatic trend. Look at Figure 4. It shows the change in percentages going to academia vs. industry and then the small percentage that goes academic but doesn't choose a Ph.D.-granting department. For the entire decade of the 1990s it was approximately 50-50, research and academia, and that has historically been the case. What you see 2 years from the end, all the sudden things cross, and that's the closure of industrial research labs.
Lucent/Bell Labs has shut down. AT&T is minimal compared with what it was. IBM still has research capacity--Almaden has done pretty well, Yorktown Heights is still around--but it's not what they used to have. But we've seen a continuing trend; on the communication and networking side of the business, the rapid growth of the 1990s ended, and ended fairly abruptly, and they didn't have the profits to put into the labs. Microsoft is the only growth component of the computer science research community.
Next Wave: Figure 4 shows permanent positions; what about postdocs?
G.A. Something quite unique about computer science: We don't have a postdoc tradition. That's because we have historically had so many positions available in Ph.D.-granting departments and available for fresh Ph.D.s. Traditionally, when you get your Ph.D. [in computer science] you've got some place to go. There was always more demand than there was supply, and industry was soaking up half. So the field was growing throughout the 1980s and 1990s, but we were only producing fewer than 1000 Ph.D.s, and half of those were going into industrial labs. And so we just didn't have an adequate supply of new faculty members, even though until a few years ago we had practically no retirements because the first generation of people, like me, is still in its 50s.
Traditionally, people going into academia did so by choice, not because there were no other jobs for a Ph.D. It was because you were interested in being in the university environment and working with students and so on.
Next Wave: So what happened around 2000 that caused the industrial labs to close or cut back?
G.A. The dot-com bust. Because a lot of the economy turned down and turned down sharply. There was a big shakeout in all the high-tech industries.
Next Wave: So why did this spell the end of the pure corporate research labs? Don't these companies still need their research base? Why don't we have those anymore?
G.A. They don't have the profits or the excess to be able to finance that anymore.
Next Wave: Where's the new technology coming from these days?
G.A. All over the place. More of the basic research is coming from the academic labs, and NSF has come down to being almost the sole supporter of things. DARPA [the Defense Advanced Research Projects Agency] used to play a major role, and they're completely application-driven now, or almost completely.
Next Wave: With the closure of these corporate labs, what has happened to the academic job market? Is it still really strong, or is there more competition for fewer positions than there used to be?
G.A. There is more competition for fewer positions. If you look at the production, it's gone down a little bit, so it's not bad. There were lots of openings last year, although not nearly as many as there were 2 to 3 years ago. There was a big spurt in the late 1990s because of the dot-com boom and, consequently, new allocations to departments of computer science and computer engineering, and that pressure is off.
And now we've got declining undergraduate enrollments. And that is a worrisome trend for us as a field, that we have fairly precipitous declines in entering enrollment.
Next Wave: So, how many new openings are expected in the coming years?
G.A. Every single time, projected faculty growth is way more optimistic than the actual. People expect to be getting new positions, and they don't materialize. In Table 17, you can see the expected 2-year growth.
Next Wave: It's modest, but it's better than most fields of science, especially for researchers.
Next Wave: Look at that postdoc category--up 25%.
G.A. Yes, but the total number [390 by 2006] is still very small. In some subfields--the theoretical computer science subfield was the first one--there was an oversupply, and so it became more like you would find in the biological and physical sciences where a postdoc was the only alternative for someone with a Ph.D. It wasn't the starvation wages of the life sciences--it was maybe two-thirds of what they could get if they had a tenure-track position--but they had to take what they could get.
Next Wave: Do you see this--more postdocs in computer science--as a thing for the future?
G.A. We've talked about it, and I've now seen it clearly from the funding side. We just don't have the money for it. We are very, very hard strapped just to support the basic research going on with faculty and graduate students. Postdocs are very expensive.
Next Wave: Many administrators in the life and physical sciences see the postdoc as an extra training phase beyond the Ph.D. But if you talk with the postdocs themselves, often they're just biding their time until they can get a better job. Is there a training role for the postdoc in computer science?
G.A. I haven't seen it. Someone who has had a postdoc before they enter an academic position has an advantage, certainly. I have a colleague at Arizona in that boat, and he's ready for tenure sooner. He's got a stronger record. It's also a problem in evaluating CS faculty, compared to faculty in other science departments.
Next Wave: Because you don't have a whole fleet of postdocs in your lab doing your research for you.
G.A. Right. And you don't have a whole armload of publications as a result.
Next Wave: What about the graduate students? I assume that computer science is a field like any other field of science in that, once you get into graduate school, you're going to make your way through it and you're never going to have to pay a cent, and you're probably going to get a decent stipend while you're in it.
G.A. Yes, for Ph.D. students that's the case. But one problem that we've had is that there are lucrative outside opportunities for graduate students, especially when the economy is booming. Anytime there's been a boom in the 30 years I've been a faculty member, there has been, essentially, raiding of the graduate students, real financial incentives for them to leave with a master's degree. In the dot-com boom, it got down to the undergraduates.
Next Wave: A technical question: Our feature is focused on software; in computer science research, is there a clear distinction between software and hardware?
G.A. There still is a pure hardware side. It's a fairly small component, interested in how to build better chips and fabricate new things. But hardware exists either to facilitate the creation of software or to take over software functions. It's so intimately connected with the systems-level software that that distinction is blurred. And the most interesting architecture work is really systems-level architecture. It's not the chips; it's the global functionality of the components.
Next Wave: There's just one more topic on my list: diversity
G.A. It's absolutely woeful.
Next Wave: Six [newly hired] African-American tenure-track faculty members, three Hispanics, and one Native American/Alaskan native [nationwide].
G.A. It's a shrinking pipeline, and it's got problems. There are lots of articles, lots of studies that have been done on causes. We've got some NSF-supported efforts that are trying to solve the problems. But it's a tough one, because there is an image about the field. And it is an image, frankly, that is promoted by industry and advertisers. I can remember one ad--it might have been Microsoft['s]--where all these people were trying to understand something and then the geek ... they had him stereotyped as male, white, and funny-looking. And that's just a stereotype that we've got to overcome. It's humorous--I can see that--but we're shooting ourselves in the foot.
As for gender diversity, there's a real difference in gender balance by subspecialty. Software fields are not good; but when you get closer to hardware it gets even worse, more male-oriented. The closer you get to people--interfaces, robots, artificial intelligence--the better the gender balance. In some subfields, it's close to equitable. Mathematics is pretty close to even, but computer science is heavily skewed. The subfield that is most mathematics-like--theory and algorithms--has a much better gender balance.
Next Wave: It seems that women aren't that interested in hardware.
G.A. Hardware and low-level software. There's something about that level of abstraction. You're dealing with strings of symbols. Not a lot of human interaction. Women move more toward the topics in which there's more contact and more relations with humans.
Next Wave: Why is that?
G.A. I don't know.