Annual AAAS Meeting
Remarks by Bill Gates
Monday, February 17, 1997
Seattle, WA

I want to talk this morning about some exciting possibilities in computer science. The amazing thing about computer science is that the hardware and software combination that's created is a tool, the best tool that's ever been created for leveraging human innovation. As this science advances, it is an aid to all the other sciences. I'm a big believer in research investments. Certainly for a software company research is everything. We have a part that we call manufacturing, but today that's either just putting bits up onto the Internet or pressing CD disks that can be manufactured for less than 40 cents. So it's a business that's totally centered around putting a lot into research and pushing the frontiers.

It's an exciting business because the expectations are always changing. There's probably no business like it in the sense that with Moore’s Law type improvements in chip technology [doubling in power every 18 months], our opportunity to do more in software goes up very, very rapidly. So every company is pushing to get at the forefront.

I think there is systematically an under-investment in research. The benefits of research are very difficult to measure, and I think it's quite unfortunate that economists have latched onto productivity as their measure of economic progress. Even given that you're going to use productivity as a measure, the way it's put together for the service-side economy vastly understates the benefits that are taking place through product innovation all driven by research. If you look at the overall level of investment in research across the board, the fact that that's not increasing as years and years go on, it's rather disappointing. It's rather surprising. As the world gets richer, and we shift resources into new areas, this should be the area that's the greatest beneficiary.

Now, for Microsoft, in a way, the fact that people are under-investing, it does make it easier to hire research scientists. But it would be a lot better in a broad sense if this was the field that more people would go into, and more money was spent there.

At Microsoft, a lot of our work doesn't deserve the pure term "research." It's research and development. And there's a spectrum there. It's very hard to divide, you know, what's pure research and what's not. In the strictest sense, I suppose research has to be something that has a high likelihood of not panning out, and very few of our projects meet that criteria. [Laughter.] For most of the things we're doing it's all a question of when, not if.

I'm showing a picture here of our campus, which is where we do 90 percent of our product development. It's turned out it's very nice to have everybody in one location. It's a little bit easier to get people coordinated and doing common work. And we've intentionally made it a fair bit like the college campus. We give the people who do the development their own offices. We have two and three-story buildings, and lots of meeting areas. And the atmosphere is very helpful, very conducive to the kind of work that we do.

One interesting fact is that a lot of what we do is testing. I mean, the fun part, the glamorous part is coming up with the idea for a new software product, and actually creating that software product, the coding part. But if you just look at the man-hours expended, a little over 50 percent are spent on testing these complex systems. And I think that's typical. Whenever you get complex systems, testing it becomes the most difficult aspect. In fact, some of the most interesting research that we're doing on this campus has to do with, can we make a breakthrough in testing and have large scale systems where we actually use computers themselves to do that work.

A software factory like this -- and this particular factory creates about 400 million units of software a year -- it's not very impressive compared to, say, a chip factory or a car factory, because it is just people sitting around and typing on their keyboards. We actually mix in, in the same buildings and offices, the people doing the most pure research activities with the people doing the most practical product type work. And that's our view of how you can make sure that as good research work is done, that it's going to move into the product and the benefits will be there.

In the field of computer science, the most famous story is the story of Xerox, who about 20 years ago did a lot of the very best work, work in graphics interface and networking, that many companies, certainly including Microsoft, have benefited from immensely. The work done at the Xerox Palo Alto Research Center is the source of the basic approach to personal computing that's popular today. But it's certainly a cautionary tale from a business point of view that all investment and all that work was done, and Xerox was never able to take advantage of that. I don't think it's something that's necessary. If you do the right things, you get the people working together the right way, there's no better investment than great research work.

And the software work that we do results in tangible software products. You can go to a store and see a box there, and it's a software product. But when people think of software, they need to think far more broadly than that. There's almost no product that's coming out today that doesn't have a significant element of software. You know, when a car is designed, the testing work, the design tools that are created there, software is very critical. All the products I'm showing here, there's probably as many man-years of software in the development of these products as there is in the hardware element. Now, think of something like the cellular phone network. Yes, it's very impressive that they can make those phones so small, but equally impressive are the software systems that make sure that the billing and the reliability and all those elements are coordinated, even as the systems become very, very large scale. The software industry is a substantial part of new product developments in companies far and wide.

It's interesting that the biggest advance taking place in software is actually the chips we have to run on. The actual techniques for creating software have not changed all that dramatically in the last 20 years. Still the best way to write great software is to start with a very, very smart person. The most efficient software projects, it's easy to say, are software projects that involve a fairly small number of people. And yet, as you build large pieces of software, you have to have different teams. And so, being able to architect one of these large problems so that it breaks down into smaller pieces, that's really the key, not only to getting it done quickly, but also making sure that it's an extremely reliable system.

There has been some progress in this area. Some of it is hidden by the fact that when the idea of object orientation -- it's a phrase to describe the factoring that let's you break things down into these pieces -- when that first came along, it was somewhat over- hyped, which is typical of our industry. And so, in the first few years, when people really didn't understand how to practice it, people were disappointed that it didn't seem to make much difference. Today it is making a difference. People are able to work in smaller teams because of object orientation.

The greatest trick ever in the world of software development is to use a piece of software that's already been written, and that's been a holy grail for quite some time. That there would just be a building block library where, if you're going to write a big piece of software, you go in and grab one of those little pieces of software and just plug that in. And we certainly are moving in that direction. But it turns out to be far harder than people expected, and sort of like saying, OK, if I want to write a romance novel, there must be snippets of paragraphs, you know, a walk in the park, or a first meeting, or something and I can just go out into this library of paragraphs and weld that into the story. And certainly you could do that. But by the time you've edited that text and changed the details to fit the flow of your story, you would have done as much work as just writing the same from scratch and not trying to draw on previous work.

And that's a very similar story to what we have in the world of computer software. The number of times people have written subroutines to look things up in a table, the number of times they've written subroutines to display a chart, it's been done now hundreds of thousands of times. And maybe the person who did it best did it 15 or 20 years ago, but because it's embedded in something with so much specific context, reuse is fairly difficult. And there are some interesting approaches that could start to improve that.

The ultimate reuse, of course, is just buying a package of software and using that without making any changes, and there companies that would have in the past written their own payroll programs or accounting programs are using package software. In cases where you just want to view a lot of information, now the tools, off-the-shelf tools let you do that. You don't go to a data processing department and wait a year or two to get it done. But, as we've tackled those easy problems, people are expecting more and more out of software. In fact, as machines are tied together for communicating in an electronic fashion, the expectation for the ability to analyze data and make sure that data is already delivered, that's going up every bit as fast as our ability to do great new work is. There will be no shortage in the demand for software and software breakthroughs.

Now, we talk about this as the information age. And I think that's very appropriate. The personal computer, particularly as it's more and more connected to the Internet, is becoming a tool for letting people share information. Now, when I say share information, I mean in quite a variety of ways. A simple way is simply sending electronic mail from one person to another. I'm a big believer that any job that involves dealing with information, certainly a research job, needs that criterion in a very strong way. Everyone who has a job like that should be using electronic mail. It's the starting point for getting into the information age, because as people start to use it, although at first it's fairly simplistic, simply sending a message saying, would you like to do this, and short replies, over time they start to realize that you can put information of all types into these messages. And so whether it's sending a spreadsheet with scientific results, or whether it's sending around a business plan that somebody wants to have approved, it's very straightforward with electronic communication.

Analyzing information is much easier on the computer than it is on paper. Just take one of the simplest forms of information in the world, the sales data. It used to be printed out on paper and put into notebooks. But if you really care about sales data, if you want to see what region is doing things well, or what product is doing well, you have to have it in a form where you can dive into the detail, and see it in different ways. And that means that it has to be in an electronic form.

A lot of work on information involves collaboration where you want to run data by people, get their input into it, have many people contribute to the final result, and doing that on paper is very, very inefficient compared to having that be electronic.

We've finally gotten this idea of electronic communication to critical mass. If you would have asked someone four or five years ago how that was going to happen, I think they would have said that it would have started somehow in the business world, online services or some initiatives that IBM would take, and so it's a surprise to a lot of people that actually it was the research world, in the protocols of the Internet, most of which have been around for over 20 years, that was the foundation for how this all really came together. It's a wonderful thing because the scientists or knowledge workers of all types that are entering the workforce today, they take these electronic tools for granted, which means that they'll push them to a new level never seen before.

Now, the personal computer is improving at a very rapid rate. The machine that four or five years ago would have been the coolest machine to buy, today it's hard to sell that machine even at three or four hundred dollars. The extra speed, the graphics, the storage has ushered in a new level of applications capability that simply couldn't run on that old machine. Fortunately, it's not just the microprocessor, just the chip that executes the instructions that's improving, it's every aspect of the machine. And so, once again, we can expect the machine that is the coolest machine today, five years from now, you'll have a hard time getting anybody interested in that at all.

Some of the specific advances are in the storage areas including using optical storage, and we've been using CDs, compact disks, for mass storage, but it's been read-only. And with this next generation, the so-called "digital video disk," which has eight times the capacity, we'll also have read-write disks, called the DVD-RAM. This means that even information that you only want to make a few copies of, it will be practical to copy off 8 gigabytes and mail it off to someone. So that, in combination with the network, which will be used for smaller amounts of information, will make it a lot easier to collaborate, a lot easier to get data around.

The displays of the computer are also undergoing a radical change. 3-D graphics, motion video, those have not been standard built-in features, but they will be within the next few years. As we have breakthroughs in flat panel display, I think people's willingness to read information off the computer will change pretty dramatically. Most people today do most of their reading off paper. There are a few exceptions to that. If you take the world of encyclopedias, there has been in the last four years a dramatic shift away from paper encyclopedias to electronic encyclopedias. That is because of the richness of information. It's more up-to-date, it has links, it's got video, it's got audio. If you think you understand a subject area, it will even generate a quiz for you and test out whether you're all that knowledgeable. And the latest versions even have links up to the latest articles. Particularly in the parts that cover science, they go out and link and show the new information that's there.

The one part of the personal computer technology that's a little worrisome, that may not advance as rapidly as we're used to, is the speed that the machines are connected at. Although optic fiber technology is improving at exponential rates. Wherever we have fiber connections, we'll have really unlimited bandwidth. Getting that fiber to be installed in all the countries, all the locations you want to communicate, particularly to the home, that's going to take quite some time. We'll see a divergence where, over the next five years, universities and businesses will be connected up at very high speeds, but homes will not be. We will continue to use the existing phone infrastructure for quite some time. And that's limited to 56 kb, which is about twice what people are using today, most modems are 28.8. But at 56 kb we cannot do video transmissions.

Over time, a number of technologies will get us out of this difficulty. There are great breakthroughs in terms of how much data can be pushed across a twisted pair. If you get within two or three miles of the final destination, you can actually send over 10 megabits of information. And so, as people are installing those things, over the next 10 years, homes then will get the same kind of very high speeds that we've had in the offices. About three or four years ago, people were over-optimistic about this. The so-called idea of interactive TV was something every phone company was rushing to beat other phone companies to connect people up, but the cost turned out to be extremely high compared to the revenue potential. And so now we've gone back to a much more evolutionary approach, which is dependent on people buying personal computers, hooking those up to the Internet and slowly but surely using modems and moving up to these higher speed connections.

Let's take a little bit longer time frame here, let's look out about 14 years, and think about what chip technology will provide. Any exponential series generates large numbers pretty quickly, and so that's what you see here. We'll go from having five million transistors on a chip to over a billion. The clock speed will go up by a factor of 50, from 200 megahertz to 10 gigahertz. And the features involved will be almost unbelievably small to do this.

Now, it's interesting the performance improvements that you're seeing here is a factor of 250 from today's processors to this one out in 2011. Of that factor of 250, 50 of it comes from just pure brute force, higher clock speed. It turns out it's very difficult to get more parallelism. This is another holy grail in our business, which is why can't you just have a whole bunch of processors and have them all doing the work and break the problem down into those pieces, and then when each processor is done, you reassemble the parts. And so if you have N processors, you get N times as much done. If that were true, then any linear increase in transistor counts should lead to a linear increase in performance capability because, worst case, you just take those extra transistors and make more processor complexes whose work you'll combine.

Well, for all but very, very few algorithms, that ability to capture the potential of parallelism has so far eluded us. We don't know now how to do that. There are a lot of great people working in this area, but there have been for over 20 years. It certainly is a very tough problem. The best approach is have a single processor that runs at extremely high speed, and so that's what you're seeing here. It's what a single chip will be able to do. We will also be smarter about connecting them together and getting some parallelism, so that means that we'll have effectively even more performance than I'm talking about here.

Now, people often ask, well, what are you going to do with all this performance? And that is a software problem. To date, the software industry has had no problem coming up with programs that fill up the hardware capability. In fact, sometimes we even get ahead of what the hardware can do. And so it kind of justifies the next round of chip improvements. The most interesting thing we'll do with this performance is, we'll use it to make computers that are far more natural to work with, computers that understand speech, that understand language, that understand vision. Those are problems that will use up every bit of that processing power that I'm talking about there.

The number of areas that we need to make advances to use this properly are very broad. We need breakthroughs in testing. We need operating systems that can distribute work around the network easily. We need new visualization techniques, much better graphics than we have. We need to deal with immense amounts of data. As you have people connecting up electronically, everybody is going to want the information presented to them to be customized. When you go to your bank, you'll want to see recommendations about how you should save or invest your money. You don't want that to be the same for everyone. And so the amount of data that has to be stored and processed will be going up at really unbelievable levels. And so we'll have databases far larger than anything that exists today.

Microsoft is in the process of putting up a multi-terabyte database just to show this. It's a database where we've taken a lot of satellite data and we'll put that online, and you'll literally have most of the world that you can go and zoom in and see with two-meter resolution any part -- anyplace that you choose to. And that is a multi-terabyte database. That's actually going to be considered a small database someday, even though today it will be one of the largest. If we simply look at people up on the Internet and all the things they're doing, what they click on, what they like, what they don't like, if we can capture all of that and process it, which is a marketing person's dream, it will be hundreds of terabytes.

Now, the value of all this really comes through the interconnections, the fact that all the information is available to everyone. And it's only in the last two years that this has really reached critical mass, and everything is coming together around this, new video formats, new audio formats. I'd say that a researcher today who is not using electronic mail and is not out on the Internet on a regular basis is really missing an opportunity. I think the progress in every field would be faster because of the information sharing that can take place there.

When people look back on the computers of today, you know, when your grandchildren talk to you about PCs of this era, they'll sort of say, "Now, what did they do? You know, computers couldn't see, they couldn't listen, they couldn't learn." You know, you use a computer for years and it works the same way, the same lack of appreciation of what you're trying to get done and your work style. It doesn't adapt itself in the way that a human assistant would.

And it's certainly -- the only way -- you interact today is by using the keyboard and using a pointing device. This idea of making a computer able to understand speech has been around for a long time. Twenty years ago, people were quite optimistic that they'd solve that problem within four or five years. And as they got into it, though, it turned out it was much tougher than they expected.

What they found is that if you simply took speech utterance and put it on a tape and then played it back to somebody without telling them who the speaker was or what the context was, that even humans couldn't recognize a lot of speech. In other words, our ability to recognize speech is based on context. This is why our speech recognition group calls themselves the "Wreck a Nice Beach" group because at the sort of digital wave level, “recognize speech” and “wreck a nice beach,” those are exactly the same thing.

And yet, if you do a demo for somebody and you say "Recognized Speech" and up on the screen comes "Wreck a Nice Beach," you laugh because it's only subconsciously that you're doing that disambiguation. And, of course, it's a very rare conversation where those two things would be equally likely, particularly if you listen ahead for a while and see what the other sentences are, then you can come back.

What that means is you can't just have scientists working on phonemes and wave forms. You also have to have people understand syntax and parsing. The whole idea of conversational dialogue and the rich context that comes out of that, and these people have to collaborate together to create an overall system that does the work.

The amount of processing power and the amount of memory it takes to have all those data structures around and to deal with them in real time goes way beyond what today's personal computers have. To do this will require the advances in chip capability that the wonderful Moore’s Law will be delivering to us. And by doing this, we can make the computer fit in, in a very different way than it does right now.

One of the areas that we just started working on is this idea of computers that see using a cheap little camera that people are talking about for video-conferencing. So I've asked one of the people at Microsoft Research who works in this area, I've asked him to come up and just give us a little glimpse of the work he's doing to make computers no longer be blind.

MR. MATTHEW TURK: Good morning. As Bill said, I'm trying to make computers see, and the long-range goal of that is to make the human/computer interaction more natural and compelling so people can interact with machines more like we interact with people. So we want to answer questions like, "Is someone there? Where are they? How many people are there? What are they doing? What are their gestures? Where are they looking?" Things like that.

There are two major themes to that. One is the theme of control where you want to actually tell a computer something to make something happen, and the other is awareness where you just want -- the computer should just be aware of these things. Who is there, is anyone there, and react differently to different situations.

I'm not sure everyone can see, but there's a little camera here on top of my monitor and it's going to watching me and the signal from the camera is being digitized in the computer. And remember, this is research, so as Bill said, there's a 90 percent chance it's going to fail, right? So the first thing you're seeing here is the screen saver, and I need to get out of the way for a second. The screen saver is just running like a regular screen saver. I can move in front of the camera and nothing particular happens. But when I walk up here in front of the machine, it turns off. I didn't move the mouse or anything, I promise. It recognized that there was someone in front of the machine and said, "Okay, well, I'm going to change my behavior now. I'm going to turn off the screen saver and let this person do work or something." Okay?

Here's what was actually going on. What you see in the top left-hand corner here is a background image, what was behind me when I stepped away from the machine. Here in the second image you can tell this is a live image of me, and I'll tell you what the green thing is. Well, the green thing is my head. The third image is basically a color segmentation. It's saying, "What is different between the background and what it's currently seeing," and let's threshold that and count the different things as being white here.

On the far right is a profile of what it thinks it's seeing. So essentially, it's just a simpler description of what's in the third window. And the details of that take a little longer to explain, but it's basically a physical simulation of a bunch of point masses that are at the top of the frame at the beginning and they come down with gravity and get pushed up by the white stuff. And so, I'll slow it down a little bit to show you what's happening every frame.

MR. GATES: Thanks. [Applause.] These auditorium environments are tough for a lot of these demonstrations. One of the tougher is to do speech recognition because you come up and you practice three or four times and it works perfectly, but when you come up to speak, they turn up the gain on the amplifiers, and so speech recognition will never work in front of large audiences. Video recognition works about 50 percent of the time. So there's room for improvement there.

[Computer makes sound.] It decided he wasn't there. It’s disappointed he's gone away. [Laughter.]

Once we have the computer doing basic things, seeing what's going on, able to listen to you, these very large screens will be pervasive, on desktops, walls. What kind of things will we ask the machine to tackle at that point? Well, certainly there's a lot of people, a lot of entrepreneurs building sites out on the Internet to make information available.

In this first generation, those sites will just be a lot like paper was, like a brochure. You want to see about this hotel, we'll show you about that. You want to see this stock price, we'll show you that. But already people are talking about a second generation which takes the idea of customization to a very deep level, and, in fact, Microsoft is one of the companies investing in this.

We are one of many companies that want to build a site that you would go to, to decide what do you want to do for a weekend, what do you want to do at night, and so, it will know where you live. It will know your preferences. If you're getting together with friends, it will take their preferences, their location, take that into account, and it will make recommendations based on knowing all the things that are going on.

One of the technologies here is called collaborative filtering. That's the idea that if you like the same movies as someone else, then when a new movie comes along, if that person goes and they tell the system they liked it, the chance that you'll like it is very, very high. And likewise, for all sorts of activities.

This will be a site that not only will it be recommending things to you, it will say, "Well, did you go?" And then you'll say, "Yes." And it will say, "Well, how did you like it?" And you'll want to participate in that because the more you tell it about those things, the better job it will do for you next time in terms of recommending things that might be interesting.

In terms of buying products, this will get very interesting. Already there's a lot of information up on the Web about buying products. For example, before you go to buy a car, it certainly makes sense to look at a Web site because they'll tell you exactly what the dealer paid for that car. You can actually make your deal on the Web.

The toughest problem of all in computer science, much tougher than computers that hear or see or any of those things, is to make a computer that learns. And here, we don't really even have a basic approach yet. When you hear about advances in computers that play chess, that's brute force. I mean, it's very impressive that they can do it, but it's purely brute force. It's not based on the same kind of pattern recognition that humans have that is very general in nature.

In fact, there's so little progress in terms of this basic learning, there's even a question of will the scientists who are sequencing DNA and learning about how human systems work, will they understand human learning before we’ve learned how to copy it. Eventually what we'll do is simply take those algorithms and transfer them from carbon into silicone. You know, I feel we have at least 20 years to still do it -- from scratch -- which would be more elegant if we could do that.

But there's no extrapolation you can do from present progress that would make you feel like it's even in the next 20 years there will be much progress there. And that's good. I mean, it's nice to have a problem that's considered very, very tough. Once we solve that one, then who knows what comes next.

But there's no doubt that if you take all these pieces and put them together, the high-speed connections, the faster machines, the natural interface, the amount of information that will be on-line and be analyzed in a way that will be customized for you, this is going to have a huge impact. The first impact will be in the business sector, helping knowledge workers who manage projects and make decisions, helping them do things better, and that will come quite rapidly, over the next two or three years, the rise of E-mail and intranets. That will be the dominant theme of what's going on.

But if you look at a ten-year timeframe, the idea of doing it out of the home, getting news this way, collaborating with people on political issues this way, and perhaps most excitingly, even applying this in a pretty deep way as a tool in education, has a lot of potential. I think if you just took the benefit to the world of research, allowing researchers to work together more effectively, that alone would justify the investment that's put in here.

And so, it's an incredible opportunity and a very exciting field to be working in. Thank you. [Applause.]

JANE LUBCHENCO, AAAS Conference President: Mr. Gates has agreed to answer a few questions.

ROBIN WILLIAMSON, ABC in Sydney, Australia: My question is about collaborative selection to do with forecasting because you mention it in your book, "The Road Ahead," when your house might have a selection of radio and television programs for you so you don't have to tune into a network. When will that be available, not just in your house, which I assume that it already is, but in my house.

MR. GATES: The data that requires the most bits, the most difficult data to send is high quality video, and as soon as you can have high quality video sent point-to-point to tens of millions of homes, then you'll get this ability to choose any program you want. It's important to distinguish between broadcast and point-to-point.

We are undergoing a transformation right now from analog broadcast to digital broadcast. Already the Hughes Direct TV system is digital, and the cable companies will go from analog to digital over the next five years. There is even now a concept of over-the-air TV broadcasts being done in a digital way. That's why these HDTV frequencies are being allocated to the current TV licensees.

When you go digital broadcast, you can compress things so you can get more choices, and you get higher quality and we'll even build in a little bit of ability to interact and see more detail. But the real big win is when it's not just channels you select, but it's just what you want to watch so that everybody in the country could, say, watch a different movie and that would all work. The aggregate bandwidth there is equivalent to millions of channels, not just hundreds of channels, millions. And so it really requires running new connections out to lots and lots of homes and, in terms of a big enough market to make that substantial, that's certainly more than a decade away, probably 15 years away before we get to that.

BOB BOYD, Knight-Ridder Newspapers: Mr. Gates, there's a lot of concern about the decline of basic research in private industry in America. What is Microsoft's current level of basic research and what are your projections for the next two years?

MR. GATES: We're expanding our research and development substantially every year. In fact, the only factor that constrains us there is just how easy it is to hire great people and assimilate great people. Microsoft Research will more than double in size over the next couple of years, and part of that is that we'll increase the number of groups, we'll increase the size of the groups. So for us, it's very straight-forward. It's the best investment that we make.

Separating out our pure research from our broad R&D budget is pretty tough. We spend over $2 billion a year on R&D because we don't have any other uniqueness except the results of R&D. Every product we make will be obsolete in a few years’ time. It's just a question of do we make it obsolete or does someone else. And there are plenty of people out there. The rewards are there if they can do it.

We're adding to our research because we think it's a smart business decision, and we're sort of surprised that other companies aren't doing the same. We don't think it necessarily requires more incentive to go and see the potential payoff that's there.

MARIE GRANDMAR, Swedish Broadcasting Corporation: Do you think we should invent another prize in computer science and why?

MR. GATES: The whole area of prizes, I'm not really an expert on that. I'm sure in this whole area of information science, people will come up with awards of various types. I think that's a great thing to do that. Like all fields of science, deciding who did something first and really who deserves the credit, you know, picking one person is immensely difficult. The more you know about a field of science, the more difficult it is to think, one person made the break-through. And yet, prizes do create motivation.

This is also a field where you get to see your products out there being used and that's very motivational to people. People are more excited about talking to people using their software and finding out how they can make it better. Now with the Internet, we get so much more feedback than we ever got before about what's wrong with it, how can we make it faster. It's great.

You know, what you want in this world is a feedback loop that guides you to go in the right direction and the Internet can be used for that kind of facility. So I'm sure there will be some prizes of various types that people put together. I wouldn't say it's a critical issue for our field.

QUESTION FROM FLOOR: Mr. Gates, enjoyed your talk. I'm a medical doctor and a philosopher. I'd like to elicit a human reaction, a human question, a philosophical issue. It's really a philosophical question. Just fantasize with me for a second. Here we have a very, very wealthy man. Let's say his wealth is $28 billion. That needn't be you, but it could be somebody. And a genie appears in his life and says, "Hey, I'm going to take your eyesight away right now and then give you an opportunity to buy it back forever." And you're a little bit stunned. You haven't had breakfast yet and, you know, this guy comes into your life. Okay?

So he takes your sight away and then says, "I will give you your sight back for the rest of your life, but I need all your money." What I want to know, in your opinion, is what is real wealth? Is it economic or is it natural? Can we say to the poor people of the world, and even the very poor people, "Hey, the blind man wants nothing but his sight," and if the genie said to you, "I'll give you ten seconds to decide this issue, I get your money or you remain blind," what would you think this mythical person, if you want to speak personally or if you want to just generalize, what do you think he would be apt to say? Would he say, "Hey, like with Jack Benny, let me think about it?" But you can't. You've got ten seconds to think about it.

MR. GATES: Well, it's very simple. Everybody would say yes. You'd certainly spend all your wealth in that case. You'd wonder what the genie was going to do with the wealth. [Laughter, applause.] There's a question, will it be recycled into funding research or empowering kids by letting them get access to personal computers and libraries and schools, the kinds of things that I think can have a positive impact.

Computers can't solve the problems like that, but there's some very exciting work that's gone on with both blind people and people with other disabilities, to take the computer and make it a tool that let's them get out and get at information. It's one of the more interesting groups at Microsoft, it's what we call the Accessibility Group, and how they're working with different people. There are lots of jobs that are being done by blind people using computers that they wouldn't be able to do otherwise.

QUESTION FROM FLOOR: The response to the question on research funding was disingenuous because all of Microsoft's research, as you yourself said, will really fit into the category of purpose, direct the research towards making better software. [Inaudible.] … If you've seen the brilliant book by Keely [sp] in Britain, the way to go for really undirected research is privatize the whole thing and let us have the corporate world do what it is doing today, whether Microsoft likes it or not. Everybody else has said, "We will do purposeful research as Bell Labs did always."

MR. GATES: I'm not sure, which category are you putting Bell Labs in?

QUESTIONER: I'm putting Bell Labs in what is mistakenly assumed that they did undirected research. The transistor was developed very specifically to get rid of the filament current and it was directed into their business. So it was part of their directed research. They had a little bit of undirected research.

MR. GATES: OK. And what's an example of undirected research?

QUESTIONER: What people are doing at universities in what they think is basic research.

MR. GATES: I don't know. I guess I don't have enough expertise. I've never met anybody who said, "I'm doing undirected research." [Applause.] I think the question is not whether people lack a direction, but rather, is there a market failure here. There's a lot economists don't understand, but they certainly do understand that the benefits to society as a whole of research exceeds the incentive to private companies to fund that research.

Reasonable men disagree on this, but a lot of people, including myself, would say there is a role for government in encouraging certain types of research. Now, it's very difficult though, how do you get the government out of picking winners and losers? I think the U.S. university system has a system that works unbelievably well because you don't have mega-projects that, you know, represent somebody in the government just picking, OK, this is the way it's going to go, at least rarely do you have those. So you have lots and lots of people pursuing different approaches, and that approach has worked very well.

But I also would say what Bell Labs did is a very positive contribution, and having more places like Bell Labs, I think, would be a very, very good thing. And at best, what Microsoft fits into is that Bell Labs-type category. I agree we don't fit into the most risk-oriented, unknown outcome-type research.

[Inaudible] Funkner, Funkner Archives. The question I have for you, Mr. Gates, revolves around medical research. So much of the medical research you see, for instance today, there's about 55 papers represented here and if you look at those papers, they're very … disease oriented. I'm very concerned about the future of not only our country, but the world in a direction that is holistic and preventive, and I'd like to know how you perceive Microsoft fitting into that.

MR. GATES: OK. I'm a big believer in research on specific diseases. That's a wonderful thing and we should be doing more of that. I'm very optimistic about the results that will come out in those areas over the next several decades. I think at the same time, if you give people -- if somebody is comfortable with browsing the Internet and looking around, they're very empowered.

One of the things they're empowered to do is to learn for themselves what's healthy and not healthy and what other people, what approaches they've taken to diets and lifestyles or things that they're interested in. Once you get that empowerment, people can make those choices. You have cases like a friend of mine, Andy Grove, who is faced with a disease situation, prostate cancer, and was actually able to go out and using these tools to find things that made him believe he should go about dealing with the disease a little bit differently than some of the recommendations that he got.

Certainly that idea that he felt he could find out about it, he could talk to his friends, and now he's available. He gets a lot of e-mail on this topic now and is pleased to help out. I think that's an example of what will go on. So when you have a revolution in communication, it's not just for solving particular diseases, but it's about people communicating about the things that they care about, and you'll be able to publish information very easily. You won't have to own a TV channel or a newspaper to get your thoughts out there.