Remarks by Bill Gates, Chairman and Chief Software Architect, Microsoft Corporation
Microsoft Research Faculty Summit 2003
Redmond, Washington
July 28, 2003

BILL GATES: I thought I'd focus my remarks on the areas that Microsoft is investing in, what we think the software frontiers are, what the tough problems are that, between your work and our work, the software business ought to be able to solve.

I was looking through the attendee list and I'm very impressed at the group that's here. It's fantastic for our research people to be able to interact with you and talk about the things they're doing and how we can deepen our collaboration.

You'll get a sense from me I'm very excited about the frontiers of software challenges, and I'm somewhat surprised that if you look at the number of people entering computer science departments or just a general mood in the industry or people at large, people are missing the fact that it's during this decade that the big advances will be taking place.

The excitement of the last decade was great, it got us a lot of connectivity, moved some things forward, but the really tough problems are the ones that we'll solve this decade.

Microsoft's Technology Vision

The role of Microsoft has been the same since it was founded. As we've grown we've been able to take on tougher problems and take our general philosophy of taking a long-term approach and really apply that in a very strong way. In our research group there are things like the mathematical theory group that look way out into the future to things like quantum computing. There are groups inside that are involved with learning theory, that are trying very ambitious things. And it's fantastic that given the success of our products we can do large-scale efforts that take on very tough problems.

We see enough opportunity in software that we haven't had to go into other businesses. You won't see us buying a consulting company or a hardware company, and we're not going to get into the chip business; we see plenty of tough things and lots of business opportunity just staying with what we've done from the very beginning of the company.

And our business model, of course, is a very high-volume, low-priced approach. Whenever somebody comes to me and talks about a market where you sell thousands of units for hundreds of thousands of dollars, my only interest is if you can take that market and offer something a bit more general, easier to access, for a few hundred dollars and get tens of millions of customers.

So as we look at things like workflow, business intelligence, moving the database to be the replacement file system for the file system that we've had to date, we're not looking at that just in the terms of people who will pay lots and lots of money for it; we're looking at it as a standard part of the platform, and it's got to be everywhere.

And there are lots of things that get proven out and developed at those higher, pure price points. It used to be transaction managers or queuing software or pub-sub software was something that you bought as so-called middleware, and most developers weren't sure whether to use it because of the expense, and you had to learn all of the terminology and the debugging and things around that particular module.

The philosophy here is to say everything you need, even for very high-end server development, we should get down into the base operating system and just have it be uniform so ISVs can depend on it.

We are going after both the business and consumer markets. The business market people in offices is where we get most of our business: Microsoft Office, departmental service to deal with those things; that's a fantastic area of strength and then tying that to our development tools and the operating system evolution, but we're also very anxious to make sure that in the home environment, with the variety of activities from Xbox to the Media Center PC, that we are able to transform that environment and get people access to information there as well.

We try and do the things that only Microsoft can do. There's quite a drop-off, if you look at commercial research projects in the computer industry in terms of software. There's what we spend, and then there's a very dramatic drop, over a factor of two, you get to what IBM spends and then you get well more than a factor of two dropdown to whoever would be number three.

Microsoft Research & Development

The R&D this year is up about 8 percent to US$6.9 billion. That may sound like more than an 8 percent increase versus last year, but there's this accounting change that took place where stock options used to be hidden in those expenses and now all the stock compensation has been restated to make it comparable and so it's now in those numbers.

So it's really now that is the real number. When we think of are we going to get back more than $6.9 billion we have to think of that true cost, and the answer is yes, we think we'll get back more than that.

So an 8 percent increase maps basically to 8 percent more people. There's no shift in terms of the cost per head, and all that money is money on people. The amount of it that's hardware or anything like that tends to be a pretty small amount.

The Digital Cycle

We are kind of in the bust phase of the boom-and-bust cycle by some measures. The boom, of course, was the explosion of the Internet, the idea that every company had to have a Web site and that these Internet approaches would overnight replace traditional activities. There was a lot of unusual valuations, a lot of sort of gold-rush-type atmosphere, some very good things that came out of that. I mean, the connectivity, all that fiber that got laid, that's a great thing. The fact that companies thought about digital processes is actually in some timeframe replacing the traditional processes; that's right.

It's a little too bad if they think just because the timeframes were predicted as being too short that they think that means it's never going to happen. Virtually everything that was discussed, even the most hyped thing, will happen. It just takes more time. And something like e-commerce, the idea that the way that the people interact and that the software systems interact go to pure paper-less processes, we have just scratched the surface of that. Today only the simplest part is done digitally and the human-to-human part that's done in parallel was not very structured and, in fact, often gets out of sync with that pure digital piece. And so the efficiency gains that can come from taking market buying and selling and putting it on a digital basis, certainly less than 10 percent of that has actually been realized.

Enterprise applications were big investments. People thought, boy, if I put in SAP, I'm going to have full visibility of where we're spending money and how we're off forecast, and every employee will just be engaged in that. The fact is, the complexity of the information has meant that it's still pretty much locked away. It's not shared in easy ways. At least that's a starting point that we have.

So these years, '98 to 2000 were kind of the explosion years. People tended not to focus on some of the negatives.

Since sometime in 2001, in terms of IT spending, valuations and just general views about these systems, there's been more of a focus on the part that isn't yet right. And this is all real stuff; in no way am I suggesting that people are seeing things that aren't there. The complexity really is there. These systems aren't well modeled. The more moving parts you get the less reliable they are, the less predictable they are.

The security crises, and I could have listed a lot more, but I particularly put forward NIMDA, that was one of the first big ones; Slammer, that was sort of the more recent kind of interesting one in terms of how network flooding actually was very difficult for people to get around, and it's easy to think with a few small variations how that could have been even more of a challenge. And it really talks a lot about how some of the base mechanisms in the system, in terms of filtering things, knowing where information comes from, even down at the standards level there's a lot that needs to be done.

E-mail is somehow getting worse all the time with the spam. I get a lot of spam, but I'm sure everybody here gets a lot of spam, too. And that's unfortunate; I mean, not only is there the wasted time there, you often can miss things that are important. You're just less likely to use e-mail. And yet that's one of these things that, over the next several years, a mix of technology and policy work, I think, will dramatically change that.

Crashes, systems with so many different pieces not always working the way they should, we have a much better view of that now than ever before. These monitoring techniques that we're applying are really giving us incredible visibility into what's going on with the systems. And it's really very revolutionary to have all that data and understand where the fragilities are, which driver pieces, which applications, and to be able to address that with updates that are flowing on a regular basis out to that installed base. I can't overstate the impact now that that's had on how our product groups organize and they look at different pieces of software.

People have noticed that the data availability wasn't as great as they wanted, and one thing I was surprised about in getting ready for this speech is that somehow this mood has reflected itself, either that or other factors, in declining computer science enrollment. And it's an interesting thing where we as an industry talked about job shortages and those things, and these jobs are pretty interesting jobs, I think, doing the most interesting work. I mean, what other science is going to change the world anywhere near as much as computer science? It's the tools of computer science that will be fundamental in many of the other sciences that are much more data driven, as Jim Gray and others have talked about. It's the tools of computer science that will take all the new biological information and make sense out of those new systems-type modeling to understand how those things work and how we can affect those things.

It's all the dreams of software, of vision and speech recognition and business intelligence; those are within our grasp. Some people might say it's three years, some people might say it's 10 years to solve those things, but by and large, those very interesting things, put aside machine learning, the very interesting tool-based things, I think it's very clear that we're on a track to make some incredible advances.

In all the discussions about distributed computing, object-oriented databases, a lot has been learned there by the work that many of you have done, and it will be applied in a very important, very practical way in the years ahead. So working on those problems is a great thing.

These are some quotes from articles about the downtrend in terms of undergrads, and a little bit this is something I'm going to study more and try to understand a little bit better. Certainly it's important to Microsoft that we get the very best people going into computer science.

Now, the trends I'm talking about here, I should be clear, are trends in the United States. If you go outside the United States, particularly to, say, India or China or many other places, you don't see the same thing, and that leads to either a situation where the shortage is being filled by the jobs being in the other countries, and there's certainly, particularly in the services industry, some of that going on, or in terms of having the people have to immigrate to the degree that that can be made simple and attractive, which always has particular challenges to it.

So particularly this Taulbee survey, the numbers were very interesting to me and I'm sure there will be some discussion over the next few days about how we all do our part to make sure that people understand the opportunities and we don't have that shortage.

The Digital Decade Isn't Slowing

So with that kind of general view of the industry and IT spending, it's interesting that the fundamental advances that really let us be so ambitious with software have not slowed down. I mean, if there was an economic situation where all the chip companies were saying, "Gee, that's it, we're not building any new fabs, we're tired of providing more for less cost and cutting each other's throats, so we're stopping at 150 nanometers, that's it, nothing else," that would have a huge impact on what we could tackle with software. If the disk people said, "We don't want to give you more bits on that disk, we've reached some limit or it's hard to make money," that would really hold us back.

But quite the opposite is the case. The kind of exponential improvement that we got used to during the '90s is continuing. Whether it's the chips themselves, although there's some interesting computer science problems in terms of how we map the increased transistor counts into increased performance; we're getting more cores, more parallel threads of execution than we can keep fully busy, particularly with the latency challenge that's emerging at these speeds, so I think tools of computer science really need to advance to take all those transistors and do the right things.

Storage, you know, we almost have too much storage and we need to think creatively about how we do aggressive caching, video scenarios to really make that storage come to life.

The advances in the wireless area I think are probably the most interesting because they change the character of computing. If you get the computer to be in your hands, like the tablet or the SPOT watch, just easy accessible, your willingness to do a wide range of things is very different. Reading, shopping, communicating, it is different, and so 802.11 has been a great thing pushing forward, getting that so it can do peer networking. The mesh type concept is something we're working on in collaboration with a number of the universities here. I think the ultra wideband, although it's slower distance, will play a super interesting role, being able to disaggregate, say, storage and the screen in a more extreme way than what we've done today.

We love the advances taking place in flat panel displays, the idea that most knowledge workers will have at least a 20-inch display in the next three to six years really has us thinking about window management and multiple things on the screen and different ways that you lay things out. We've been very constrained there and we've sort of taking 640x480 as a given. Web pages, when you navigate to them with that large screen, they just sit there and only use a small part of the screen.

And, of course, the answer is not to have long lines of text. If anything, long lines of text, text lines are already too long on classic screens. That's something that the reading mode in Office 2003 has really shown us, that it's so much more comfortable to read on the screen when we do the layout in a more columnar fashion.

So the hardware industry, there's very little that we can complain about in terms of what they're providing us. It's true that broadband to the home is not happening overnight, but it is happening. In countries like Korea it's happening a bit more dramatically than in the U.S., but that's certainly something that we can take for granted that over the course of a decade, there will be high broadband penetration.

So what are the software breakthroughs we need? I'm sure you each have your own ideas about this, but these are some of the ones that we're particularly focused on. You'll hear a lot about Web services here. It's the new standard protocols that allow distributed computing to be done in a way that the information is described in a rich fashion, allowing systems to be connected where the developers never had to meet each other; they'd just engage in those standards.

We think this is a key building block, and, in fact, we think a lot of things like systems management that was built historically on a completely different stack, it had its own platform different from the applications, those things need to be redone where they're using the Web services platform and the model of the applications is used to drive the management activities.

Dynamic Data Center is our term for what Sun calls N1, IBM calls autonomous computing, simply the idea that if you describe, that is schematize the hardware and the applications and the performance requirements you have on those, that a massive number of things that were done manually and very slowly can be done by a model-driven approach, and so you can have a whole data center or a whole computing grid where software is allocating those resources.

For us this is super interesting. You just take testing alone here at Microsoft; this will make us more efficient. We need to build systems and test network configurations. We should be able to do that in 10 or 15 minutes, and not a single person having to go and touch a system. And so we will be able to refine this just with ourselves as a customer in a pretty aggressive way.

It's interesting, when people talk about that, they act like the hardware companies sometimes act like it's a hardware-related problem. It's not a hardware problem. In fact, we can afford to use the computing resources a little bit less aggressively than in the past to drive the reliability in these self-management type capabilities.

I talked about monitoring. The wealth of data we're getting, not just crashes and error conditions, but also feature usage frustration, what is slow, what is the performance, that is a huge deal to us.

I talked about business intelligence and allowing rich information to be navigable so people can tell what's going on with quality or customer attitudes or financial type information. We have spreadsheets but they don't understand that type of structure. We have SAP R3 with its 18,000 tables, and somewhere in between that complex view and that pure 2D grid view, creating navigable, schema-driven representations of the information should be doable for all knowledge workers.

Social computing, that's a fairly explosive area I think, because of the connectivity we've gotten, Instant Messaging really just scratching the surface in terms of what can be done.

Trustworthy Computing: That's our moniker for all the things related to security and reliability and, of course, a lot of invention required before we live up to the standards that all these other applications really are going to need before we realize their potential.

Opportunity to Have Huge Impact

So, very fascinating problems in computer science. It was over 25 years ago when I was last in a computer science department as a student, and I remember we were talking about proving programs, trying to find were programs correct and things like that. Well, now there's kind of a renaissance of that, almost out of necessity, because some of these secure computing things require that we understand the effectiveness of the software, and there's real progress both in the academic space and in terms of our practical use of exactly that kind of proving tool.

Rick (Rashid) mentioned that we're working across a lot of devices. He drove the group that created the SPOT concept, including the watches and all the smaller devices that will come out of that. We're investing heavily in the car design. We've got now under contract -- you go three years out, which is the design lag on these things -- about 30 percent of the cars will have a Windows CE display system built in them, more heavily in Japan and Europe than in the United States where putting a display in the car is still a bit more controversial here than it is in the other countries. So in this country a lot of our design is focused on voice-driven type systems.

We think in the living room the fact that there's just very limited set-top boxes, that's finally going to change. It will change because the idea of what's a set-top box and what's a videogame, those are really coming together. You shouldn't have separate electronics, separate remote controls and things for those things. And the high end of that for us, of course, is Media Center, which gives you the greatest flexibility and centralizes all the information and yet distributes it out to inexpensive screens throughout the house, using a terminal services protocol that's been enhanced for rich media experiences.

So we see all the different form factors, the pocket form factor, the Tablet; these all have to work together. It's still kind of ridiculous that, as you move from machine to machine, the amount of automatic roaming really isn't there for even simple things like schedules and favorites and files, and that in the next generation of Windows that roaming will just be something that's built in, a key core concept of the storage system.

I really don't need to tell you there's a lot of great research directions, large-scale computing broken down and with different approaches. The mesh network thing, it takes what's the most expensive thing about consumer computing, what's the hardest thing. Well, by far it's the cost of communication. As my Foundation is going around the world trying to put computer systems in all the libraries so that people can come out and get to those, the actual PC and the software training and that part we've got down. In fact, (in) five countries now, including the U.S., Canada, Chile, Mexico, that's gone super well, but the thing that holds it back in every case is that the connectivity is just super, super expensive, and the rural areas are often having to resort to satellite-type approaches that are way out of the reach of the library budget.

And so if some advances in wireless and mesh computing can bring the cost of connectivity down, that idea of being serious about the digital divide, that is what would be the huge breakthrough. It's not the $400 PC, it's not the software, because Microsoft in all those Digital Divide cases is willing to come in and provide software donations for those things; what holds it back is that just having a standalone machine, as nice as that is, really isn't the vision of all the information that you want to get at.

The new form factors, people like these tiny, little PCs, the pocket-sized devices. That means it's finally time to get these natural-input techniques into the mainstream.

Moving the database into the center of the system, there are some very interesting problems in terms of versioning, security of how we're doing the schemas for all the classic information types that we're interested in.

The efficiency of software development, that is not something that's changed. People have talked about model-based approaches but we're putting an immense amount into that.

So no shortage of things that would have a huge impact, that whoever does the advance should feel like they really changed how things are done.

We have a lot of programs, as Rick mentioned, that try to get you software and any of the tools to work in these areas. We have an advance in this, which is getting some of the low-cost hardware for embedded devices out to you. We have some partners that are helping you with that. We'll help with that. In the DemoFest there will be an in-depth discussion for anybody who's interested in that.

There are a lot of skills that go beyond just classic programming skills that are very important to us as we bring people into our product activities. Security: That makes the top of the list no doubt, understanding what is required to build secure systems, what kind of modularity, what kind of techniques, what kind of modeling, what kind of proof tools are going to be out there.

Programming in large environments where there's millions of lines of code, that's the problem we're dealing with and very interesting to us to get people who think about that and think about better approaches to it.

Of course, Web services, most of the code written going forward this decade will be in some way connected to Web services, either the platform for executing it or applications built that way. Networking, of course, is part of that, quality issues and then all the different interface things.

We look forward to having students who have got some exposure and some new thinking in every one of these areas coming in and helping us move forward.

One thing we've done to help this is we worked with MIT on having a course that's about Internet applications, and this is a course that talks about UNIX platforms and Windows platforms, that shows what the .NET runtime looks like, it shows what other Web- services-type platforms look like, and it's been a great collaboration where we've learned a lot, been able to contribute. Already 50 students have been through this and this is a course being given again this fall. We really appreciate the collaboration we've had here with MIT and we think there's a lot that's come out of this that can be applied in many different universities. It was a pioneering effort really with the view that many other people could drive some of the same things.

Why We're Optimistic About Software

So let me just finish by saying that you're in a place where we're very optimistic about software and what it can do. I've got to prove that to our shareholders by showing that we will get more than $6.9 billion back for the $6.9 billion that we're spending.

Just in terms of the kind of excitement and neat things that are done, this has got to drive the work between all of you and Microsoft to see what more we can do together, and we're very committed to doing this, to taking the long-term approach.

Sometimes financial people give us a hard time for keeping so much cash in the bank and having such a strong balance sheet, and the reason we do that -- and we may have to moderate it somewhat -- but the reason we do that is we really want to be able to keep a long-term program going. We want to be able to look out literally at a 10-year timeframe and know that we're going to be growing research, growing the product groups, really solving these interesting problems and getting those into the high volume platforms. So I'm excited about where we can work on those things together.

Thank you.