Discussion with Craig Mundie, Chief Research and Strategy Officer and Alan Alda, actor and host of “Scientific American Frontiers” on PBS
March 4, 2008
KEVIN SCHOFIELD: Well, thank you, Rick, that was a great overview of Microsoft Research and the work that we're doing here.
It's actually a rare privilege for me to introduce the next two gentlemen who are going to join us on stage. Many of you are probably already familiar with one of them. Craig Mundie, the chief research and strategy officer for Microsoft, has been with the company for quite a while now. We're also very fortunate to have with us this morning Alan Alda, who's obviously a very well-known television and movie actor, and he has also been the host of “Scientific American Frontiers” on PBS. This is actually an area around science and technology and computing that's near and dear to his heart and he has a lot of interesting thoughts. In fact, both Craig and Alan share a conviction that technology and science can help transform society, and they're going to join us this morning on stage and share with all of us some of their thoughts. So please join me in welcoming Craig Mundie and Alan Alda. (Applause.)
CRAIG MUNDIE: When Kevin was doing his introduction he said, “We have two people, I'm sure you know one of them,” I expected that he was going to say Alan and not me.
Alan and I actually met five-plus years ago. We were both at a meeting in Beijing and we were attending an international conference associated with the Museum of Radio and Television. And we were sort of thrust together in an opportunity to talk. And I'm very technology oriented, you could say, and I discovered almost immediately that while many people would know Alan for his work in “M.A.S.H.” and other TV activities, he really was tremendously interested in technology.
And, you know we immediately lapsed into a discussion about what the latest gadgets were and the problems he was having with Windows and other things.
ALAN ALDA: Not that many --
CRAIG MUNDIE: But I promised to help with those. But we became friendly and have continued our interaction on and off over the last five years.
When we were talking about putting together this year's TechFest and trying to help people understand how computer science is essentially moving beyond just helping our engineers write the code for the next product at Microsoft that is really becoming an integral part of how pretty much every facet of daily life happens and, increasingly, how all aspects of science and engineering are done.
We wanted to sort of have a dialogue about that aspect of the thing and not so much talk about the technology, but the impact of the technology. And I said, you know, they said, you know, who could you have this conversation with? And I said, "Alan." And so I called him up and he agreed to fly out today and join us for this discussion and I want to thank you for doing that.
ALAN ALDA: Oh, it's been so terrific. I had a chance to go in this morning with Craig before we -- before this presentation started, and we saw stuff which I guess you'll get to see in the next few minutes. And it was an amazing -- and we only looked at four or five of the exhibits, but each one of them you could have spent a couple of hours at figuring out what they were trying to do and how they had accomplished what they were doing, what the ramifications of those things are.
One of the most amazing things I saw today was that -- I think it was one of the last things you just saw here, that program that enables you to zoom in on points of light in the night sky and suddenly find out that you're looking at nebuli or you're looking at --
CRAIG MUNDIE: The WorldWide Telescope.
ALAN ALDA: Yeah, the WorldWide Telescope. Well, I think that they mentioned that you can create your own tour, an individual user can create his or her own tour of stars and star systems and then provide a commentary and pass that on to a friend. And sometimes the person who does it will be an expert, and sometimes the person who does it will be just a curious person. And the most amazing thing I saw this morning was this tour created by a six-year-old boy. And you hear this six-year-old voice with all this enthusiasm and the wonder of a child leading you through the universe in a knowledgeable way. I mean, he was – “Here we are, this is M3, this is a really interesting place.” You know?
CRAIG MUNDIE: “Not sure what M3 is, he says, but I'm sure—”
ALAN ALDA: Yeah. “I'm not sure what it is, but it's really interesting.”
CRAIG MUNDIE: “And those light years, that's a long bicycle ride.”
ALAN ALDA: It was a great moment because you saw in that brief tour, you had a hint, a smell of what the impact of something like this can be in terms of education and in terms of communication of science.
CRAIG MUNDIE: Yeah, I think when you even take the WorldWide Telescope and look at it, you saw there are a number of interesting things there. Of course there's the technology of how do you capture those images, process the images, register them, overlay them, because to some extent, it's such a huge body of work that if an individual astrophysicist had to sit there and try to do that for their own work, it would be impossible.
ALAN ALDA: It's really interesting. I don't know if it was spoken about on the video, that these images exist on hard drives all over the world, and you're drawing on them and stitching them together on the fly.
CRAIG MUNDIE: We're making a, if you will, a community of computers, you know, that is supported by a community of people who are both contributing to and then drawing knowledge from the ability to do that. And then coupled with the ability to teach through it and create the storytelling that goes on in that I think is an interesting aspect.
Curtis, you know, before he did the telescope, was involved in a lot of the early work in multimedia titles, including some of the ones -- the famous ones that were done around art. And I think it's his ability to blend that storytelling capability with the core science is a key part of that.
A number of things we looked at this morning I think move us in this direction of augmenting how teaching will be done, the ability to bring people together in a collaborative way I think is a key attribute that we see emerging today. People are communicating through the Internet a lot, but they haven't been collaborating through the Internet all that much. I think that that -- that step up from communication to collaboration, whether real-time or otherwise, is going to be one of the transformational things that happens not just in the enterprise for workers, but for everybody no matter what their question.
ALAN ALDA: Within a family -- figuring out what to do about Johnny who's liable to get expelled from school, you know, or figuring out a vacation spot that the whole family would be able to enjoy. I think that's an interesting model because the kind of collaboration you're talking about, different from the kind of collaboration that takes place just by virtue of the ability to communicate, is significant, I think, because time, it's happening in real-time. This kind of collaboration—time after time this morning, I saw collaboration that was possible and computer modeling that was possible where things happened in real-time. And that means that when you suggest something to me, I can have an immediate response to it. And I don't think about it, I don't look down in the e-mail to where you said it, and then respond in a more formal way to it. It's much more intuitive.
CRAIG MUNDIE: And that moves us in a direction, again, we showed Alan some things today that are in this vein of what we call the Natural User Interface, where the machinery of interacting with the computer is receding from consciousness. It becomes a more natural experience. You can collaborate without having to think a lot about it. Merrie Morris spoke here, too, and we looked at her work. We also looked at another component of it, where some of the work from MSR Bangalore, which we call MultiPoint, where you can plug multiple mikes, computer mikes, into a single computer, project that on a screen in the classroom, or in front of a small display, and the kids can come together and use it. And each of them has their own distinguished cursor, but we've been able to make interesting ways for the kids who don't have access to individual computers to be able to collaborate in a micro-cosmic way even within that application.
ALAN ALDA: I bet it's going to be very interesting to find out in the future if the kids experience working a computer in collaboration with other kids that this kid knows and is used to working with, if that kid's experience is going to be significantly different from the experience of working on a single computer. Right now, we think the kid is lucky to have his or her own computer. The kid might be luckier to be forced to collaborate with other kids on a computer. That social experience might lead to more learning.
CRAIG MUNDIE: Yes, I think that both are going to come into play. In fact, it's not here at TechFest, but Microsoft Research has supported a piece of work down at the Catholic University in Santiago, Chile, which you didn't see this morning, but I'll explain a little bit because it's in the vein you described. There they gave small pocket computers to each kid in the classroom, so in a sense they each had their own, but the curriculum was developed such that each stage of the problem solving had to have a contribution from each of three kids. They were broken into teams. And each team had to ultimately cross the finish line with the problem to get a grade, but each kid had to solve part of the problem themselves. In fact, they have video studies of these classes, where before this technology was introduced, you'd have the kid who was the smartest in the class, or the one who always wanted to be visible would dominate. But once this happened, it was a great equalizer. Then the kids would actually start helping each other, because they said, hey, I can't get my part if you don't get your part, so I'll go over and help you learn about this.
So I think the collaboration, whether through multiple use on one computer or innovative ways of linking the computers together, whether it's locally or across the Internet, I think this collaboration thing is a big deal.
ALAN ALDA: I think so, too, and it really hit me strongly this morning as we went through that festival, and exhibit after exhibit seemed to indicate that you’re you're putting your finger in something that hasn't really been explored much, it seems to me. The benefit of building a giant brain out of the various brains that exist around the world, or the brains that exist in a classroom, the smartest kid in the class now has to share some of his brain with the other kids, without sitting in the car in the parking lot letting them copy his homework. This is a much better way to do it. But, the idea that the knowledge, and the ability, and the analytic skills, and the modeling skills of scientists around the world get focused onto a screen that I can look at, and I don't have the programming skills to bring that together, and I don't have the knowledge to put those ideas together myself, but to make use of all of these brains, and make them work, and then to take the data and put that all together. For instance, in the model of the cell, or the rainforest, collecting the data from all over, the way you are with those sensors, when you combine those, and you combine the collection of the data with the modeling of it, and you start to put these programs together, I don't think this kind of stuff has been done before.
CRAIG MUNDIE: Not at this scale. I think the last big evolutionary step in this, you could say, was when the PC got introduced with word processing and spreadsheets. Those were tools that moved the abstraction of how you interact and get things done up a level. Before that you had to be a programmer, you know, yourself, and actually write a program that could do something. And very few people really were adept at that, but when you could give them a spreadsheet as a metaphor that they could understand the cells, and you write little formulas, link them all together, people were able to do incredibly interesting models within that environment.
Some of the things we showed you this morning, I think, are the beginning of, again, making a qualitative change in the levels of abstraction of the tools that we're giving people.
ALAN ALDA: Meaning what, what do you mean by that?
CRAIG MUNDIE: Well, when we were looking today at the tools that we're giving the climatologists, and the people studying the rainforest, you would basically just have little blocks that represented whole data sets. So you said, oh, I have that data set over there about the currents in the ocean. And he said, and here's one about the wind. And he would basically just take those things and he would drag and drop lines together just onto the globe, and they would essentially connect him, through the Internet, to those data sources, and then it would draw pictures of them, and map them onto the globe. Much as we saw with the WorldWide Telescope, where each astrophysicist or astronomer, they might have spent their lifetime, as Ray Gould said, I think, in the video, studying one cluster, or they would study the ultraviolet spectrum, and somebody else would do the infrared spectrum. They never actually saw this stuff together. And so whether it was out in space, or here, we were giving people tools now to bring things in where they don't think about writing a program, they don't even think about filling out cells of the spreadsheet, they basically just sort of reach out and say, bring me all the data from that environment.
And one of the big improvements there is the data now is becoming self-describing. You don't have to have somebody who describes in advance to you, what is all that data. You know, what's the format of it, what does it really represent, and what's its scale, et cetera. We can put metadata in these things that allow all these programs that never knew about each other before to essentially bring the data together, composite it in new and interesting ways and, therefore, it supports this collaboration process, or this joint discovery process at a level that really never has been done before.
ALAN ALDA: Am I right about this? I got the impression that a good example of that was the ability to model what was happening in the cell. All the different parts of the cell, you could see them going through a process, and you could put them through a process experimentally and see if you were right about it. Is this an example of what you're saying?
CRAIG MUNDIE: It is another example. Alan is talking about this modeling language that Steven Emmott, I think, actually talked about in the video on Rick's talk, where cell biology is incredibly complex, so the solar system, and space, and that's huge and complex. But even down at the level of cellular biology, these cells are incredibly complex machines. And so, you know, again, we see people spending their careers studying one little pathway in a cell, interaction, one effect of genes, and proteins, and things. The question is, how do you bring all that together? So we're now giving them tools that allow them to describe the cell biology, but not just to describe it, but that description itself becomes a model. It's sort of like the spreadsheet of the cell. Therefore, you can calculate the model.
Even though in the wet lab, it might be only possible to say, “Well, I can see what the cell is doing when it started, and then I could see where it was at the end, but man that chemical cascade is either complex and I don't understand it, or it's actually happening at a speed where I can't really stop it midway and understand. But I know there's something going on.” There hasn't been a great way for them to express it or explore it. So by, again, extracting that into these computer based models, you can take something as complex as cellular biology, and begin to give people tools for this collaboration of their work, and then ultimately this ability to make some type of joint insight.
ALAN ALDA: I love that, especially the spreadsheet of the cell, because that's a very close analogy, because you can tweak the cell as if you were giving it a drug, but you don't have to give a real cell a drug, and you can see the reaction you're getting, you can do experiments on the computer to some extent, can't you?
CRAIG MUNDIE: Yes, and I think one of the things that's going to be important is this ability to transcend physical scale, from literally the atomic to the universe. Of course, without computing we don't have a way to give humans the ability to interact at scale across these things. And I think for the first time we're really doing that, with a level of detail and precision that was never possible. The cellular modeling is interesting, not just because it helps us understand the most basic processes, but it holds the potential, as our computational assets and understanding increases, to move towards things like system biology, which essentially is building bigger and bigger computer models of the entire network of interaction.
Many people now think that biology is a network of networks, sort of like the Internet is. And if we knew how they all worked at each level, and we could compose them into bigger and bigger models, in fact, the things that you talked about of the future of understanding medicine, or pharmacology could be explored without actually doing it initially in living organisms.
ALAN ALDA: Something that I kept asking as we were going through the exhibit was how do I get my hands on this, or who gets their hands on it, because when I looked at the night sky on the computer screen, and saw us zooming down into these star systems I thought, so what, is this connected to some giant computer that a lab has to buy? And it turns out, did you say it was a plug-in on the
CRAIG MUNDIE: Internet Explorer.
ALAN ALDA: That's unbelievable.
CRAIG MUNDIE: Well, that one wasn't that was an application. There are elements of it that you can get in the frame of a browser, and in fact, you can send people links through the Internet or through e-mail that would allow them to go. That particular application that Curtis wrote is a specific piece of software that takes in all of these things, and presents it in that nice environment. But, certainly a number of the things that we did show you today, Merrie’s work in terms of the shared user interface, the work in climatology, those were, in fact, browser plug-ins, where we're essentially taking a piece of software, adding it into the browser as a frame in which it executes, which allows you to have an intelligent way of interacting around this topic, and then the traditional browser presentation, for the rendering, or the integration with other Web data assets.
So it's a new way for people, again, to put things together and make them easily available. Today we refer to a lot of this as sort of click to run, where you didn't have it before, and you found it on the Web through a search, or a link somebody sent you. You click the link, and you're running it, because it just comes down through the Internet, executes in your machine, takes advantage of the local computational and storage facilities that are there, but blends it seamlessly into this Internet experience.
ALAN ALDA: How did you how did Microsoft get started in this, because this seems like I'll just give you the background of my question, what I mean by this question, you're dealing with basic research, and some of the things that I saw today are extremely exciting, because it seems to me to raise the level of education, and it will raise the level of communication of science to ordinary people, and raise the level of communication among scientists, which is extremely important. That doesn't necessarily feed into the coffers of Microsoft, and this is basic research that doesn't necessarily lead to a product. How did Microsoft decide to do that?
CRAIG MUNDIE: I think there are several things that come together here. One, as Rick said, the fundamental raw material of Microsoft is smart people, and the output of those people is intellectual property, intellectual assets that we try to deploy in a wide array of ways. One way is we use it to enhance the products we have. Two, we use it to create new products that we never had before in our traditional field. And three, we use it to either respond to opportunities, or things that are happening in the society at large, or even on a competitive basis.
So the research is the mechanism that allows us to continue to draw that in. One of the things that is different, though, in the last few years, is moving beyond just using this computer science research in the computing product itself. As you and I discussed before, Bill will retire from Microsoft, Bill Gates, and spend most of his time at the Foundation at the end of June, and this year in January at Davos he gave a speech, which I think is I call his transitional speech, because he's now both famous for the work that he's done here, but he and Melinda are now increasingly famous for the work they did in the foundation, too.
And he introduced a phrase that he called “creative capitalism” in that talk. And he said, look, today we have a billion and a half people who do pretty well on the planet, we've got another two billion today that are sort of coming online, and then there's at least another two billion people that really live a subsistence existence. And if we're ultimately going to improve the lot in life of all of those people, in his mind, it's pretty clear that neither government nor philanthropy will be able to do it by itself.
But, one of the things that Bill has encouraged, and which we've done, I'd say, in some natural way with this MSR work, is to take some part of the company's assets and, in service of helping solve these societal problems. Whether they're in education, or health, or farming productivity, or the water table, many of these challenges that the global society is going to have cannot really be addressed at scale without the use of computing and software, and notably, at least in Bill's mind, they ultimately can't be addressed at scale unless the big multinational companies begin to figure out how to at least make a business, or a break-even activity out of helping at some of these other levels.
So partly that's been the motivation for us to go beyond just using computer science to improve our computer products, but to actually take people into the company and to create more relationships around the world with people who aren't the computer scientists, but really are the scientists and engineers who are trying to solve some of society's toughest problems. So whether it's working to help with the genetic analysis of the AIDS virus to look for a vaccine, or modeling the water table at Berkeley.
We showed you, for example, the sensors, and one place where we're doing a project is in the Alps, where they're basically measuring the retreat of the glaciers in the Alps.
ALAN ALDA: So you drop these sensors all around in the Alps and they communicate with each other?
CRAIG MUNDIE: Right, they collect data about wind and temperature, reflectivity, and pollutants. And all of these things, essentially, are a way of trying to understand how global warming, or the actions of man, or something are going to affect the glaciers, which turn out to be the source of all the fresh water supply in that part of central Europe. So we look at each of these problems, the cell biology problem, and its future impact on medicine, all these areas, they all have applicability not just in the traditional places, as you say, where we make money, but it's important that the company contributes to the resolution of these other problems.
ALAN ALDA: And how long will you stick with an exploration, if it seems there's no hope of it producing a product?
CRAIG MUNDIE: Well, I don't know that we've been at it long enough to fail a lot and be able to say there's a general way to answer that question, though one thing that has been an attribute of the company commercially forever is we're very persistent. I came here 15 years ago to start non-PC computing, and one of the first things we started on was television, interactive television. And today it's called IPTV, and it's only now becoming sort of mass deployed on a global basis.
So the cycle time of these things that are changing society's infrastructure is pretty long, 10-15 years or more. And I think that Microsoft has been willing to make long-term investments, not just in the research itself, but in its applications, and be patient about how long it takes to make these things emerge.
ALAN ALDA: And if one of the programs that's on exhibit in the other room, if that continues to have beneficial effects in terms of education or science or both, but it looks like it will never turn into a product that Microsoft can sell, will you stick with that?
CRAIG MUNDIE: Again, I don't know that there is a universal answer. But I think we will continue to look for ways to place these things where the benefit can continue. I don't know that we would always just keep doing it ourselves, and that's why I think, as Bill said, this concept of creative capitalism requires some intersection between the role of government, the role of philanthropy, and the role of business. We might say, hey, look, this thing is never going to be a business, but if it turns out it's the key to improving agriculture for poor people in rural Indonesia, we might find a way to transfer that technology to the academics in Indonesia. I think we will look not just to develop these things, but to find ways to let their benefit be sustained over long periods of time.
ALAN ALDA: It seems to me there's also a model for doing this work in the first place which involves handing it off at a certain point, otherwise the body of work just keeps growing and chokes on its own size.
CRAIG MUNDIE: That's right. Even within our own research efforts, not everything we explore materializes something that we know exactly what to do with it. And so we have a variety of ways we deal with the ultimate disposition of the research activities. Some of them just get transferred as technology into the product groups, and those 6,000 people that will wonder around the exhibits here from Microsoft this week, that's a key part of how that technology transfer happens.
The second thing we do is, sometimes we need to take some of these concepts or a set of the research assets, and bring them together to create a conceptually new thing. What Curtis did with the WorldWide Telescope, I think, is an example of that, where there wasn't an individual product group you could go to at Microsoft and say, oh, you should take all of these particular technologies and just use them, because you had to show how they could be composited in a way that did something that hadn't been done before. And so that may become either a service or a product. In this case, we decided to make the WorldWide Telescope available for the world for free. And that's just a contribution of Microsoft to the community, and as such it becomes a repository for all of the work that that science community will do.
The third thing that we do is, sometimes we bring these things forward in ways that allow us to create fundamentally new businesses. One of the areas we started three years ago was a business in health. One of the things that certainly motivated me to think we could make a contribution there was looking at how health itself, healthcare, is going to evolve. On one hand, medicine is going to become a more data-driven business. When we were down there looking at the cell biology and the computer models of these things, I mean, at some very, very low level, computing is going to be important there. But genomics, and proteomics, the use of these things to really get benefit has to be made personalized, and to some extent the amount of data that has to be collected, analyzed, maintained, integrated, and studied in order to produce a personal prescription, or a personal pharmaceutical almost will dwarf all the data that has been used in medicine up to this point.
And so when I look at the research assets, whether they're in image processing, for example, one of the biggest breakthroughs in medicine in the last decades was all the imaging, started with X-Rays, and now we get CAT scans, PET scans, and MRI scans, and all that imaging has been a huge and valuable diagnostic tool, but much as what's happened with the space guys, no one today registers all those images. Every image that you've ever had of an X-Ray and say, “Hey, can you look at them all at once, can you actually take volumetric scans, like a whole body CAT scan, and then correlate it with individual PET scans, or MRI scans, would that be useful or interesting?” Would it be like the scientists who are looking at the WorldWide Telescope and say, “Hey, I never, ever saw my data in context.” I think in medical imaging, we're likely to see that. And some of the things that we do there may be able to be brought into the medical sphere.
And then the other thing I think that's going to happen globally in healthcare is that it's going to shift to a focus on prevention, personal involvement. And so many of these things where communities, or families, or individuals are brought together, because they share an interest in a disease, some member of the family had it, or you may want to talk to other people who have a chronic disease that you have, all of these mechanisms in the Internet, I think, are going to be very powerful in bringing together not just the medical and research community, but the actual individuals who either care about wellness, or have a particular interest in a specific health-related matter. And then allowing that energy to be harnessed in some way is another thing I think that we would like to do.
So there are many ways that we take individual components of the research, none of which were started for the specific purpose of serving a need of the company in health, some of the work that's done here in the machine learning area, Eric Horvitz's group, David Heckerman, these guys have been they started out doing machine vision, they got the machine learning, and a few years ago they got asked to look at the AIDS virus, the people who were searching the genome sequences of the AIDS virus, and they were able to create a new way of using machine learning to study this that all of the people in the medical community had never really thought about, and to recreate the experience and understanding in a short amount of time that the world had spent five years looking for up to that time.
ALAN ALDA: I was wondering as you were saying that if eventually you would be able to use this model of the sensors in the same way that you planted the sensors on the glacier, and are studying how the glacier actually deteriorates, under what forces, and through what dynamics it goes. I wonder if you could use a similar technique and apply it to epidemiological problems, like the spread of bird flu, and see where things start to crop up and find out what the pathways are, and see if you can choke it off at an early time, and human diseases.
CRAIG MUNDIE: Actually, Stephen Emmott, we didn't have time to show that this morning, but one of the other projects that group is doing is with one of the universities in the UK is actually having a model of large scale, almost global epidemiology, and being able to do that kind of thing. What's interesting is that you don't just have to use sensors in some discrete way to do that, some of the work that we've been helping with in India, for example, is where people use cell phones to report anomalous observation.
ALAN ALDA: Yes? Like what?
CRAIG MUNDIE: Like “I saw a dead bird, birds don't usually drop out of the sky here, and this one did.” And they sensitized the population to say, “Hey, you know, bird flu, and yet birds do weird things, they die in ways that you don't normally observe. If you see that, send a little SMS message to this number.” And so all these messages come in with some triangulation from the cell as they put their location in, and you start to build maps of where these things are happening. And then you use the Internet to collect these things and scale them up. And, in fact, the population at large is becoming the sensor network to basically make judgments, or provide raw data, which you can then mine, and do noise reduction on, and determine, is there a pattern here that's anomalous.
One of the things we learned, you'd say, in dealing with the health of computers as opposed to health of people is, years ago, using some techniques that came from research, and the Windows group, we built this error reporting mechanism. Almost everybody who uses our software today at some point, some bad thing happens, your app crashes, the machine has a problem and says, hey, can I send some data to Microsoft? And so we get millions and millions of what you could have said were random almost pieces of data from all those kinds of failures out there.
ALAN ALDA: These are failures on individual computers.
CRAIG MUNDIE: On a computer, individual computers, one at a time. But when we look at them on a worldwide basis, we discovered we could learn things that we never anticipated by doing pattern matching, and machine learning on these kinds of things. So today, we not only are able to find an individual literally, almost an individual machine that breaks for some anomalous reason and say, no, your machine just broke, and distinguish that from the introduction of a virus, for example. So today we can detect the release of computer viruses around the world because some of the machine that get infected start to have failures, and the failures get reported, and that machine wasn't failing yesterday, in fact the whole group of machines here weren't failing yesterday, and today they're all failing. And when they all send the reports in, little sensors go off at Microsoft that say, something is happening in Jakarta this morning, and we should go figure out what it is. And you look at it and say, oh, look, somebody released a virus, and it's spreading out of Jakarta today, and we'd better figure out how to stop it. And so essentially we're dealing with epidemiology in the computer network.
ALAN ALDA: Yes.
CRAIG MUNDIE: And the question is, how do you take that learning and bring it over to epidemiology in the physical.
ALAN ALDA: That's a theme that goes throughout that exhibit. One is collecting and amalgamating data that hadn't been collected before and massaged, and collecting and amalgamating brains, and putting those two things together, I wonder if that isn't going to have an effect that we can't imagine now, just in the same way I have so many friends, because I became interested in computers, and trying to learn simple computing languages 15 or 20, 20-25 years ago, and I had so many friends who said, “What would I do with a computer? I'd like to get one, but I don't know what I'd do with it.” And as soon as they got one they found out what they could do with it. I'm wondering if some of the things in the other room there are going to have the same reaction on people.
CRAIG MUNDIE: I think they will. In fact, I think that today many people in the science and engineering community have a bit of a way of thinking about computers, roughly the same way they think about pencils and graph paper and things. It's a tool, I wouldn't want to do my science without having these tools, but many of them haven't yet made the leap to understand, this may be able to effect my science in a more profound way than I thought, either because it brings me access to data that I didn't know exists, or didn't have any reasonable way of integrating into my own work, or, in fact, it couples me into this giant brain pool, in ways that I would not have historically done.
I do think that these are going to produce not, I think qualitative changes in the result. I actually think it's going to result in an acceleration of science. We've already seen knowledge growing at exponential rates, but I think that supplemented by this level of computational horsepower, and connectivity, and collaboration, as promoted through the connectedness here, and the tools to support discovery and sharing, I think is going to produce another significant acceleration.
ALAN ALDA: How would you that seems like such an important idea. You're distinguishing between analytic tools, the computer up until now has largely been an analytic tool, like a good slide rule, or a good piece of graph paper.
CRAIG MUNDIE: A big calculator.
ALAN ALDA: Yes. And this is something else, but I can't put a name on it.
CRAIG MUNDIE: I think I don't know how to name it, but I think this idea that you're able to more or less directly manipulate very large datasets, or interact at large with big groups of people without having to think a lot about it makes this collaboration a lot more possible. The ability to just use data you didn't know existed.
I think one of the things that people are fascinated by when you give them a broadband connection, a personal computer, a Web browser, and a search engine, you don't have to teach them almost anything, and they discover that there's a lot of stuff out there, and that just type anything in, something comes back. That becomes a training mechanism for you poke it, it comes back with something, you follow that link, you go again.
This, I think, recursion is a very natural thing, in terms of people's curiosity and interest. What we're doing now is allowing simple mechanisms like that to be applied to things that are wildly more complex, and to be integrated in some way that makes them useful, without having to stop and learn a huge amount about all the work that everybody else did. I think one of the things that we've struggled with in computing, and software in particular in the past, was that our tools didn't really lend themselves to composition. People tended to say, I'm going to write an application, a piece of software, you sit there and you write the whole thing, you may use a little library, but at the end of the day you architected and wrote this application, big or small.
Once you had that application, unless you had specifically thought about how it was going to relate to other applications, they frequently couldn't talk. So in the last few years, the introduction of meta-data, and the formal ways of describing it like XML has had a huge effect in allowing people to contemplate the use of data that they didn't know about it advance.
I think as we now move to these higher level abstractions for how people will "program" a computer, where they're not sitting and casting programs sort of one comparative statement at a time, but are linking together the works of other people, much the way that people who design buildings or bridges, and other things, they take the work of lots of other people, some guys who are specialists in concrete, foundations, and ceilings, you bring them together through software that aids that.
We've never really empowered people at any scale to build bigger and bigger edifices that are these computer programs, all working together. But, I think what you're observing as you walk around today is that we're moving beyond this traditional model of writing programs, or designing them for a priori interaction with another set of programs, to a world that's a lot more free form, in your ability to discover, link it in, and then utilize it.
ALAN ALDA: It's almost like the difference between when we had to walk around and look for a bush to pick berries from, that was like a first step, and then we figured out that we could grow farms of berry bushes. That's, in a way, the way we've been using computers up until now. And now you've got farms of berry bushes that are producing bowls of chicken gumbo, or something. I mean, it's like it goes through a whole other level that's not that's much more complex, and yet gives it feeds it to us at a much simpler, easier to understand level.
CRAIG MUNDIE: These ideas that the man-machine interface is getting easier, whether it's moving beyond typing and pointing with a mouse, we looked at that sort of research on this touch interface, where people say, look, it's really natural to look at a picture and touch it, but if it's an electronically generated image, and you're interaction is through your fingers, you don't have the kind of resolution you want, you have the occlusion of the picture while you're trying to interact with it.
ALAN ALDA: You have to learn which part of your finger is going to eventually touch it.
CRAIG MUNDIE: Either that, or genetically we're going to evolve to have very pointy fingers, so you can really point with precision. But, I think we also see the desire for mobility, the ability to use computing not just at my desk, or when I can move my desk someplace else with a laptop, but today everybody use of cell, the idea that you have this thing in your pocket, it's always on your person, you can get at a lot of information, is creating more and more desire to have access to these things in a portable environment.
As we move to introduce you could say there's we showed you both ends of the scale today in some of the research. The WorldWide Telescope, which is at a dimensionality that's hard to fathom in big scale, and then we're down there looking at some of the cell biology, and very, very tiny things.
I think one of the things that's going to be interesting is the work of blending together the real world, and some synthetic worlds. Today everybody thinks about that as mapping. You can look at maps, Virtual Earth, or Google Maps, and others, are a way for everybody to say, I didn't if I wanted to have a map before I'd have to get the atlas out, or I have a little local map, but the ability to go and zoom into any place on the planet, and see the roads, the trees, and the aerial photography, the was a big step.
Now we can think about using that to synthesize a model of the world, and then to be able to populate that model with more and more precision, and then to study things at that scale, or allow models of interaction that are, again, very natural. I did some demos yesterday for some folks that showed what it might be like to take pictures of the real world, and from that create a 3-D model of the real world. And then to do that, not just for the exterior, and facades of buildings, but to be able to go into a building, or a store, and have a physical model of the interior, too.
ALAN ALDA: Does that involve somebody going in and shooting the picture, and putting it somewhere, or is that use of pictures you've got?
CRAIG MUNDIE: Either one, you can take a bunch of individual photos, you could put little Web cams in the corner of your store, once an hour it could take another set of pictures, and from that
ALAN ALDA: So then, sitting at home, I can go shopping in that store in 3-D?
CRAIG MUNDIE: Right, correct. And so this will be another model of natural user interface, because you know what it's like to be in the store physically, you know it turns our people have their brain really likes it relates to the visual presentation of a real 3D environment. We showed you that soft shadow stuff today, and it was important to see how the visual cues that you get from the shadows were important in your own ability or comfort in looking at these images.
ALAN ALDA: A couple of computerized figures, computer generated figures, but they looked so much more real, simply because they had little soft light shadows on the ground under them, and some of those shadows in their features were it gave it a more believable modeling, because you knew intuitively you knew there was a real light source, and that they were existing in some world like yours, that's how I would describe it.
CRAIG MUNDIE: That's exactly right. And so all of these are computationally intense problems. A few years ago, before Microsoft, 15 years ago, I was in the super-computing business, and the machines that we would build and sell for $1 million a copy couldn't come close to computing those shadow images that you saw on those figures today.
ALAN ALDA: What's the difference, better machines, or better algorithms?
CRAIG MUNDIE: A little bit of both. As he pointed out this morning, in the shadow work, that even the best machines we have, these incredibly powerful graphic computers, are not really capable of doing all of the individual ray tracing to give you the shadow calculations in real time. So in animated movies that's fine, you don't have to do it in real time. If you're trying to make this into a 3-D model, people can walk around and go shopping it has to have some kind of real-time nature.
So there the breakthrough was algorithms on top of the hardware advances. And I think that's always going to be the case, that our aspirations, in terms of the visual quality, or the degree of resolution that you have, will continue to always outstrip our current ability to compute in, but we are moving ahead to a world where there will be more and more parallel computation.
I think one of the biggest changes that's going to come in computing in the next few years, probably the biggest change in the last 30 or 40, is that computing itself is going to have to evolve to be more based on parallelism to get performance.
ALAN ALDA: What's going to make that happen? What has to be done, what has to be worked on?
CRAIG MUNDIE: Well, almost everything. The hardware is going to have to change, architecturally. Human nature is parallel, and to some extent nobody goes around and figures out how gravity here relates to anything else, and you don't design it, but it all runs in parallel all the time. Yet, the way we build computers, they're all running in serial, and everything has to be carefully constructed. And we've made them very, very fast, and so they do a lot of very interesting stuff. But, to some extent, you still do a lot better.
If you were to get assigned sort of a clock speed to a person, a biological system, its system clock would probably be something like 30 hertz. And the current computers are three gigahertz. So there's 10 to the 8th difference in clock speed. So when you think, you reason, you navigate, and figure out how to get out of here, and walk up the stairs, in computers we're still struggling to figure out how to do that. What's the difference? You’re massively parallel, and the computer isn't. Your algorithms are intrinsically operating all simultaneously, and those are not.
The stuff we showed you in the cell biology work was showing that the only way we can model these things now is by assuming that everything was happening in parallel, that there was a stochastic, or probabilistic element to it, and then trying to determine the way nature does, what's the right answer.
ALAN ALDA: When you're working in the cell, you're working with the probabilities of one particle of the cell interacting with another?
CRAIG MUNDIE: Right. But, in the future I think more and more of those attributes will be how we build these computers. The very next step for us is to not have millions and millions of computers on a die, but to have dozens or hundreds of them, and then try to link them together so that they collaborate more on an individual problem.
ALAN ALDA: As long as we're talking about the murky future, what's going to happen, as you look forward, say, to better parallel computing, how is that going to be affected by, let's say, 10 years from now somebody making a really functional quantum computer, where you're working with multiple states, and not just ones and zeros? What's that going to do to you? I mean, you're going to have to rethink everything.
CRAIG MUNDIE: That one will require a rethinking, I think, but I don't think that's going to happen in the next 10 years. We have a project that's working in quantum computation, but my expectation is that's not going to materialize in any kind of general sense in that short a time cycle. I think that's 20 years and beyond. And there are a certain class of problems that that will be good for.
Michael Freedman, who is the guy that started that work, when we talked about starting this project said, well, let's go covert this from a math theory into an experiment. And we actually set out to do that with a bunch of physicists down in California, we moved the lab there. And as soon I said to him one day, I said, okay, everybody talks about quantum computers, and what would they be good for, and it can factor big numbers, and therefore it would break up graphic codes.
He said, yes, yes, you could do that, he said. I said, well, what do you think if we could make one of these things, the first one, what would you do with it? What would you do with it? He said, oh, the first thing I'd do is I would have it compute the materials for its successor, which was a fascinating statement, because it turns out the key to quantum computation is operating down in these quantum physics regimes, which is at milliKelvin, just thousands of a degree above absolute zero. And these states only exist in current materials at that level. And so that's really hard to make, and keep at that temperature.
So he said, so what you really want to do is you want to compute the properties of a synthetic material that would exhibit these properties at a slightly higher temperature, and then you could make a bigger one.
ALAN ALDA: And you'd probably need a quantum computer to calculate it.
CRAIG MUNDIE: No, that's exactly right. Today it's completely intractable with computers as we know them to compute those materials.
ALAN ALDA: You know what worries me about this, because of my history in entertainment, I think about the first film was nitrate. And it disintegrated, and sometimes went up in flames, but then when they changed the technology, they copied a lot of the old films and brought them forward, and we have them today, because they did that. And then as we got digital they did the same thing, but every time the technology changes they had to make a decision, they culled, it's like culling a herd. And the ones that they transfer, that decision is made on the basis of the culture at the time, what seems important, what do we want to send into the future.
They're constantly making these capsules for the future every time the technology changes. And when we go from ones and zeros to a completely different way of computing, there's going to be a gigantic culling of all knowledge, not just of movies. What's going to happen?
CRAIG MUNDIE: I don't actually think about it that way. In part, because I tend to think of the information, and how it's represented as something slightly different than the model of computation that we sort of apply to information.
ALAN ALDA: But, will this be so smart, this new one excuse me. Will this new way of computing, which is maybe far less serial, is that going to be so smart that it can take everything that already exists on this one and learn how to translate it?
CRAIG MUNDIE: Yes. That I'm actually quite comfortable. But there is another fascinating thing that is happening now that wasn't happening that many years ago, and it's another area where some of Tony Hey’s projects at Microsoft Research are involved today, and it is with this question of digital libraries, and the whole question of archiving the digital history. And so, you know, when people were making these decisions step-by-step as a business decision, you know, they were guided by one thing. But, of course, society is always depended on archivists whose mission wasn't to preserve everything, but to at least preserve a representative sample.
One of the things that I think is really interesting today is that our ability to store things has grown in capacity to such a degree that we're really getting close to approaching a time where almost nothing needs to be forgotten.
ALAN ALDA: Then you have to worry how to search through it.
CRAIG MUNDIE: Well, and to some extent, you know, we're also getting to the point where we can find anything we remember.
ALAN ALDA: Okay.
CRAIG MUNDIE: So at least at the scale of the Web with billions and billions of pages of information, you know, we not only remember all those things, but we can find every one of them. The next thing that's happening now is that we actually remember all of the interactions. So, for example, the advertising businesses that we're in, and other people are in, are increasingly saying, oh, I want to remember the things you like, and you've done, because then if I have to give you an ad, I can figure out what it is. And so, on one hand, it's scary to some people, but on the other hand the ability for the machine to be an assistant in helping you to recall and discover goes all the way back to the beginning of this conversation, and says, if I'm a scientist and I'm trying to solve a problem, I really would love to find anybody who has anything that contributes to knowledge in this space, or any person who can connect this with this, and somehow bring them into my environment. And I think that the scale at which we're doing that today mankind has never done before.
And so as you talked about, we're sort of building this ability for mankind to work as a big brain, you know, supplemented by literally billions of artificial brains in the sense of the computer algorithms, and the machines that run them, that are really helping us to do these things. It's clearly creating a projection of knowledge and capability in the future that mankind has never known before. And I'm personally super optimistic about that. I think that behind that does lie ultimately solutions, scale solutions to global health, and education, and ultimately helping every person to move themselves to an improved quality of life.
ALAN ALDA: Are there efforts among the people in Microsoft research to, and I imagine there are, to do what you were describing before, to bridge that gap between how the computer can approach the ability of a human to do simple things? I can recognize a face, I think, more easily than a computer can. I can recognize a voice more easily. I can do some spell checking a lot better, because I can tell which word just plain doesn't go in that paragraph, even though it's spelled correctly. Although it amazes me, sometimes, when I'm working in Word and it picks me up on a change of tense or something like that, and I'm grateful for it. It noticed it; I wouldn't have noticed it reading through it 50 times.
CRAIG MUNDIE: Right. It understands grammar, and it will look at every word you type every time you type it.
ALAN ALDA: But these other things, we're going to need computers to do, especially as they sense more and more, to be able to put two and two together, something like the way we do it, which is so efficient, it's been developed over millennia, so we have a head start on computers. But is there much work being done on that?
CRAIG MUNDIE: Well, there is. Almost every human sense is being analyzed relative to how does the computer emulate that capability, sight, sound, even smell, tactics, and other things, touch, there's work going on here and elsewhere in the world in every one of these things. The question is not how do you emulate those things, but taken one at a time how do you integrate them into some system that would help people solve a problem or do something?
You see powerful examples of these things today where in medicine you're starting to see artificial limbs. I've seen people now who have lost their legs, and you give them an artificial leg, and it's not just a fixed thing that they stick on you, it's a hydraulic, motor-operated, battery-powered, computerized thing. And you've got these little intersections to the nerve, and you really operate that machine. So I think there are many, many things that are going to happen that will qualitatively change people's lives for the better.
It's been great talking to you this morning. Thanks again for coming.
ALAN ALDA: I had a good time. Thank you.
CRAIG MUNDIE: Wonderful. Take care.