In-Depth

The Power Inside

As head of Microsoft Research, Rick Rashid leads a team that remains largely anonymous, but whose work finds its way into nearly every product Redmond ships.

• By Keith Ward
• 10/01/2005

Goodman is a Microsoft Researcher, one of about 700 worldwide, and his work in the area of e-mail and spam has been part of the effort that became Sender ID, Microsoft's flagship anti-spam technology.

Of course, you've probably never heard of Joshua Goodman or any other Microsoft researcher. But if you use Microsoft Word, Outlook, Windows Media Player or SQL Server, you're using technology from Microsoft Research (MSR).

Goodman works at the Redmond research lab, one of Microsoft's six research centers worldwide. He's been with MSR for about seven years, and his current research interests are in e-mail and spam. "The goal," he says, "is to learn what good e-mail looks like and what spam looks like." Much of his work is involved in developing algorithms to spot the "Amazing New Diet Pill!!!" ads, and to try to keep up with the endless variations spammers use to elude the spam filters he and others write.

Goodman is an MSR old-timer. When he graduated with a Ph.D. in computer science from Harvard, he was working on speech recognition, and had offers from two other labs. "They were great places, but the environment at MSR looked like it was going to be the best. There was much more opportunity here to have an impact."

Nebojsa Jojic has been with Microsoft Research for five years, and works in the Redmond lab.

Sudden Impact
That's an attitude that separates MSR from much of the academic research community.

"The reason they [researchers] come here is that they want to see their stuff in shipping products," says Kevin Schofield, general manager of strategy and communications for MSR, who handles the technology transfer process. "My job is to help them do that. I don't have to go around convincing them to get their stuff in shipping products."

That's certainly true of Goodman. "Our spam filter is used by [more than] 100 million people. It's fantastic to have that kind of impact."

Impact is much further down the road for other projects. Nebojsa Jojic, of the Redmond lab, has been with MSR five years. He researches machine learning, where he's developing algorithms for "data analysis of any type of data—audio signals, text or anything," he says. He worked on data analysis of graphic images for some time, and now he's working on biological data, including genetic sequencing. While he says some of his work has made it into applications, most of his current efforts won't be in a shrink-wrapped box for some time.

Where's the Beef?
Jojic's experience illustrates the quandry in which MSR sometimes finds itself. On the one hand, good research takes years of painstaking effort; but when there aren't a lot of obvious end results, Microsoft gets blasted for not coming out with lots of hot, industry-changing products. Some of the groundbreaking, paradigm-shifting products like the iPod and iTunes, and Google's search capabilities, were produced by much smaller companies, and Microsoft is scrambling to catch up. Think about it: When was the last time you saw an upcoming Microsoft product and said "Wicked cool!"?

That's not a fair comparison, argues Senior Vice President Rick Rashid, who oversees the research division. "Our job is to produce fundamental technologies. Pretty much any product Microsoft produces will have technology that comes from Microsoft Research."

Roy Levin heads up the Microsoft Research Silicon Valley lab.

Rashid gives some examples. "The development of Windows Media Audio; we have better quality than MP3 at half the [file size]; high-definition video technology came out of Microsoft Research. These aren't the products. We do fundamental things, things that move the state of the art forward."

That's a phrase you hear a lot coming out of MSR: "Advancing the state of the art." It's become MSR's motto. But advancing it how? For example, Rashid points to an area he's particularly excited about, something he calls "social computing." "We're looking at all sorts of different aspects of how computers can be used to manage social groups. Something almost like a super-blogging tool that covers the space between e-mail, IM [instant messaging], blogging, sharing photos, all in one interface. It automatically builds a social network; it looks at who you've interacted with and builds a network out of that."

It Just Works Better
Roy Levin, director of MSR's Silicon Valley lab, says "there can be a huge amount of quite significant innovation going to various places in that product line that are not immediately visible at the end as individual things; it's just that it works better."

He cites the example of the Help system in Microsoft Office. "In roughly 1995, you had keyword-based lookup. Pretty straightforward technology. The natural language processing group in MSR [the oldest research group] contributed technology that did semantic analysis on the queries that were then put in the Help system. So the interface looked identical, but the pages you got back were better, because they weren't restricted to things where you had keyword matching. That's a pretty substantial change, going from string matching to semantic analysis, yet the innovation is almost transparent."

Outside the Pressure-Cooker
Although MSR has contributed a lot of the plumbing to products, Levin is quick to point out that the product teams don't pressure MSR to develop products to get to market. "We certainly consult when they have a problem we can help solve. We do a lot of that. And we develop new things on spec, with the expectation that they'll be attractive to the business units. But there are no year-by-year metrics that say ‘you have to transfer this amount of stuff.'"

That's another way of saying that there's a different mindset at MSR. "We're very disconnected from sales. Customers are interested in today's products; researchers are interested in ‘three-years-in-the-future' products," Levin says.

That doesn't mean the researchers are coasting along, playing Doom on their high-speed work connections. The hours for a researcher are long. Jojic estimates that he puts in 50 to 60 hours in an average week.

He describes his typical routine: "I spend a lot of time with products and in front of the computer, adjusting algorithms. I also spend a good amount of time writing papers and traveling, working with interns, coding, answering e-mail, reviewing papers for conferences and journals."

Despite the long hours he and his colleagues put in, there isn't a lot of turnover. "You sometimes get overworked, but the core work is really fun," he says, "So most people just work until they get too tired, [then take a break and] get back to the trenches. We're not waiting for 5 o'clock so we can get home. I haven't met anybody in MSR that's just burned out. They might switch focus, but nobody really gets burned out."

Follow the Money
Part of that is due to the fact that unlike most other divisions within Microsoft, MSR doesn't have to make a profit. "Our focus is in creating new technology. It's not really a corporate-focused goal," says Rashid. Explains Schofield: "Rick [Rashid] doesn't run a profit/loss center," he says. The overall R&D budget for the coming year is between $6 billion to $7 billion, but of that R&D, according to Levin, "D gets most of it."

Precise figures were not forthcoming, but Rashid confirmed Levin's estimate of where the lion's share of money is spent. "The vast majority [of spending] goes to building our products—Windows and Office and Xbox. Basic research is not an incredibly expensive endeavor."

But R&D isn't incredibly cheap, either, given that most of those 700 researchers have Ph.D.s. "We try to make their lives easy so they can concentrate on research," says Scho-field. "The rules are simple: If people ask for resources they need, they get them. If they ask for resources they don't need, they don't get them. We don't ask principal researchers to manage their own budgets; we don't want them focused on budgets. We want them to focus on doing great research."

To keep things simple, MSR is organized like an academic computer science department, with a flat structure. Also similar to academia, researchers are given a lot of autonomy, but Microsoft researchers might have it even better. "I have a huge amount of freedom," says Goodman. "It's an interesting thing about MSR vs. academia: [in academia], you typically have grant money you need to apply for and then have somebody approve your time [to meet grant requirements]. Here we have more freedom than we might in an academic institution."

That's by design, says Levin. "We hire researchers for their creativity, and the job of management is to provide an environment where they can exercise that creativity.

Management doesn't say "Gee, I think you should go work on this," because then it would be limited by our ideas, and that's not good."

That management style has led to a work atmosphere that Goodman describes as "laid-back intensity. It's laid-back in that we can do whatever we want. For a lot of what we're doing we don't have deadlines, but what we do, we do in an intense way."

Running Out of Researchers Bill Gates must have said it four times during a recent press conference describing the state of technology research in the United States: "We're keeping smart people out of the country." He was referring to the H-1B visa program, which limits to 65,000 the number of aliens who can work in the United States. Gates is concerned because Microsoft relies heavily on computer scientists from foreign countries, especially in Microsoft Research (MSR). Gates paints a bleak picture. "At Microsoft, we're having a tougher time hiring. If you look out at the future, that's just going to get tougher. The jobs are there, they're high-paying jobs, but we're just not seeing the pipeline where it needs to be." The H-1B visa issue is critical because there are fewer U.S. undergraduates going after computer science degrees. That will have a "trickle-up" effect in the years to come, according to Senior Vice President of Research Rick Rashid. "This year, hiring has been spectacular. These are the students that were in school when the boom started. The problem is that you look out in two years, and the number [of computer science Ph.D.s] drops by 30 percent, the year after that maybe 60 percent, and that's when you start panicking. Some of the people we've hired were in school before there were Web browsers. That's a long pipeline. In a few years, we could be in a world of trouble." General Manager of Strategy and Communications for Microsoft Research Kevin Schofield is concerned about the narrowing pipeline of researchers. Gates concurs. "We're quite concerned that the United States will lose its relative position in something [research and development] that's very critical to the economy. Our elite position, where we develop the best people in this country and many of the best people from other countries come here, that's certainly eroding." The effect, according to Rashid, is that Microsoft Research has "grown mostly outside of the country. We have about 200 [researchers] in Beijing, about 100 in Cambridge [England], and we started a new research center in Bangalore [India]. Expansion outside of the United States is necessary to continue growth. We'll do a significant amount of growth in Redmond," Rashid says, but non-U.S. hiring is still seeing greater increases. Gates says it's by necessity, though, not choice. "We're going to the top universities, but there just aren't as many graduates with that specific type of background. We have those open positions, and it creates a dilemma for us in terms of how we get our work done." What this means for you is that a lucrative, satisfying career in computer research awaits if you go back to school and get your Ph.D. "We're always looking for new, smart people," says Kevin Schofield, general manager of strategy and communications for MSR. Without new blood coming in, he says, "Two to three to four years from now, we're going to be in serious trouble." — K.W.

Getting Ideas into Products
There are some deadlines, though, and most have to do with the technology transfer process. Goodman says that "when I work with product groups, I need to do what I promised to do and get them stuff on time."

There is no formalized "technology transfer" process, though. It's mostly about relationships, says Jojic. "In the end, it's about having a relationship with a product group, and it happens a different way in each instance." There's a lot of give and take, he continues. "The product teams come and give talks to MSR to say what they're doing next. Technology transfer is about working with someone in a product group."

Schofield echoes that sentiment. "People think that technology transfer is a Rube Goldberg contraption. It's about people, relationships, communications and trust."

One way that's facilitated is a yearly Microsoft-only trade show, where groups present their projects. Following this, Rashid says, "The product group takes the research and integrates it, moving forward." Levin says that "Rick Rashid is very well plugged in to what the product groups are doing, but the real choice of what to work on lies in the hands of the researchers."

In the end, that kind of freedom may be the key to happiness as a Microsoft researcher. As Jojic says, "It's hard to imagine I'll do anything else. It is a really good job, a really good thing to do, at Microsoft or anywhere else." Says Goodman: "It's not the easiest or most lucrative way to spend your life, but we love what we do."