>> Xuedong Huang: It is my great pleasure to welcome Professor Raj Reddy from Carnegie Mellon University to visit us. He will talk about Guardian Angels that will be very exciting to many of us. Professor Raj Reddy is a pioneer in artificial intelligence and robotics and speech recognition whose achievement is without the need to have any introduction. And for many people who are familiar with Microsoft internal people we know that the three managing directors from our research lab in Asia started from Kai-Fu Lee, Harry Shum, Hsiao-Wuen Hon were all Raj’s students at Carnegie Mellon. So, without further ado let’s welcome Raj Reddy. [applause] >> Raj Reddy: Well thank you. It’s a pleasure to be back here. I used to come here two, three times a year as a member of the techno… TAB, from ninety-one to two thousand and six. Finally I said, “I served enough,” and stopped coming. So I haven’t been here in the last three, four years. So it’s a pleasure to be back here at MSR. In the digital world of twenty-first century every person on the planet should be able to get timely warning about potential dangers that might befall them, such as earthquakes or flooding or typhoons or tornadoes. So in this talk I want to kind of give you an outline of what a Guardian Angel Technologies might be like that will provide the right information to the right people and eliminate surprise and protect them from unexpected problems. So, with that we’ll go to the first slide. So, twenty years ago my colleague, Professor Carbonell, who happens to be here—both of us visiting Microsoft in another context—proposed a grand challenge for computer science, that society should aspire to get the right information to the right people at the right time in the right language in the right medium of text and multimedia at the right level of detail. This was, you know, something with a clear vision twenty years ago but we didn’t know how to do it then, it was not clear what we mean by all of these. At that time getting it to the right language meant we had to solve the translation problem. We didn’t know how to do it then. At least now it’s slightly better…sig… it could be significantly better. And text to speech technologies and summarization technologies are also making progress, but the right information to the right people at the right time still continue to be difficult ideas to grasp and, kind of, get your hands around. So, before I… you know, so what… one of the interesting things was—that kind of caught my attention when I heard it from Jaime—was he called it the Digital Bill of Rights. Those of you who are aware of the Constitution of the United States know that they are the first ten amendments, are Bill of Rights. So I said, “What relationship is there between this Digital Bill of Rights and the regular Bill of Rights?” This is the thirty human rights from United Nations which includes the ten Digital Bill of Rights that are in the Constitution of the United States. Maybe not all of them, there’s some overlap and most of them are there… but something’s are not… but there is a statement in the original Bill of Rights saying you cannot… a soldier cannot walk into somebody’s house and take it over. I don’t know whether it is there in that form, but the right to own property might be part of such a statement or something. But, so anyway, the interesting thing about the… this particular bill of rights is the original bill of rights came from the seventeenth century philosophical thought of John Locke and adopted by Thomas Jefferson and James Madison that people are naturally free and equal and are endowed with some unalienable rights. It’s a very famous statement in the Constitution… or is it the Declaration of Independence? Anyway this is… this is something I read about. I thought it was a great statement that until that time it was not clear that every person had some rights. If you’re in a kingdom, king can order your de…, you know, your head shall be chopped off and you’re gone. But here it says they’re naturally free and equal and they’re endowed with some inalienable rights. So, from there we went to the human rights in a digital society by United Nations which are kind of interesting in their right. They’re… you kind of find a set of, you know, rights like unrestricted access, a right to be connected all the time, right to freedom of online expression, security and data protection, and so on… And again, these are, you know, well deserving rights but they’re not necessarily implemented or followed by countries in the… around the world. And different… there are censorships and other kinds of things in different countries… when they see something that they don’t like, like Turkey or Egypt, they will immediately turn off the whole internet [laugh] if necessary. And so the digital bill of rights we’re talking about here is much more proactive. It says if you can get the right information to the right people in the right language in the right time frame, then you can actually help the people in ways that would otherwise may not be able to do. So the question is, “Can we do that and what would it take?” And we’ve made lot of progress on some of the things—summarization, translation, and so on—but not much progress on the first three. And so I just want to make sure that we are all on the same wavelength when it comes to what we mean by these terms. The right information is all information impacting life, liberty, and happiness. In particular, information about safety, security, and well-being and that means providing information about natural emergencies and man-made emergencies. It’s also information about disruption of basic necessities of life, such as water, electricity, food and health, transportation. And it’s also about… information about daily problems: traffic jams, strikes, school closings, and so on… The missing science of what we mean by right information is… for any one person the right information is the relevant information. Most of the information that is there in the news and other sources is not of direct interest. Secondly, there’s the scarcity of human attention. We don’t have enough time to look at everything that is there so we need to somehow personalize the technologies of information. And the right information is also kind of getting that information in a form that’s easily accessible. So if I’m driving or if I’m, you know, in a meeting or something you can’t have a loud ban… noise on the phone saying there’s some e-mail has come or something has come. And so the notification and alerts that you see on cell phones I think are pretty… pretty reasonable as a way of providing this information. And so given a known problem such as an earthquake, the big problem we face is how do we identify who is going to be affected by that event. And then once you identify the sphere of influence, then you say… you need to say, “Who are all the people that are currently there in that area?” That’s not currently known to most people but it’s known to the service providers because every tower knows what are all the phones that are connected to that tower. That is required by law actually, they have to keep that information about location information. And so then once you identify who are the affected people and what information to be sent to them, the next thing you want is tell them that information’s happening and then prepare them for various contingencies. Give them advice on how they might, you know, kind of protect themselves. But the interesting part is if you turn off the computer, if you turn off your cell phone or if you’re sleeping or… under what conditions should this alert be, you know… how will you get the information? What that means is the current concept of turning off a phone has to be modified. There’s only one situation where even when the computer is turned… when the cell phone is turned off you get an alert. That is if you set an alarm and go to sleep, even if the computer is turned off, it wakes up and rings the alarm, right? So mechanisms are already there inside, the only requirement now is an external agent… external software agent must be able to do the same thing. That essentially is a problem because an alarm doesn’t consume too much energy because it’s not sampling the cell tower continuously and that consumes a lot of energy. And so the issue is are you willing to permit that kind of waste of energy? And I think partial solutions are possible. You don’t have to continuously sample. You could do it every five minutes or every two minutes or whatever, so that the amount of energy consumed is like one percent of what it currently consumed and it’s always on ready to receive a phone call anytime that I call. So the… assuming that’s possible, assuming that you… and this external software agent can actually, you know, send an alert and wake up the computer and then provide appropriate information, then I think using push technology for timely consum… communication can be satisfied. Finally, you need a data base of personal profile of interest, you know. Basically, a lot of people in the world don’t speak English, so giving then an English language alert is not going to help. So you need to know what language they’re speaking. In absence of having a data base your best bet is to simply use the language of the location and you already know the location information from the cell tower. So the idea is people will not be very happy about you tracking them any more than the information that is already being tracked by the cell tower. And so that’s the privacy issue. We’ll come it to a little later. And so the issue is they cannot not only understand English language, they might not be able to read any language, so simply sending a text alert may not be the adequate… that’s where the issue of right medium comes in. So whatever message is being sent has to be then converted from text to speech and then spoken out. So those are the kind of… what I call missing science of what we need to do in order to provide the right information to the right people. Now we come to what do we mean by the right people, right? And the right people, again, is a very complex issue. And the right people is anyone and everyone in the world that may be impacted by the daily events—that’s both literate and illiterate. And people in the, you know, industrialized world and also developing economies in the most remote areas. And so I am saying every person has the right to get the timely information and that’s seven billion people. And unfortunately there are some limitations to it. The two biggest limitations are that many of them don’t have a device that can receive an alert. Right now, it’s estimated maybe about two billion people out of the seven have a device that can receive either a smart phone like message or a simple text message on a feature phone. But the remaining five billion… or, you know, maybe three of those are expected to join the digital revolution in the next few years, by twenty-twenty, and maybe two billion will never have it unless there’s a specific attempt made to get all of them access to the technology. And later on I’ll make the case it is a right of every individual to have the information, therefore it’s a right of every individual to have a smart phone, therefore it’s a right that they can demand. And if every… and in a civilized society everybody should have a road and electricity and sewage systems and water supply, they should also have information access. That’s a right of the society and so we’ll come back to that. And… but the right people then, is everybody on the planet. The right time is very interesting also because the right time is just in time. As Jaime used to say, “If it’s too soon, we forget.” If I send you a reminder of this meeting a week ago, you may or may not remember it, right? And… or if you send it one minute, you know, before the meeting, it’s too late. By that time you may be far away and can’t come and attend the meeting. So it has to be at the right time. But more importantly it’s not something… it’s not always something that you can plan for. There are extraneous events in life that happen at the most inopportune time. It’s not only the natural emergencies and man-made emergencies, it’s also personal emergencies. You know, suddenly your father is seriously ill and you have to drop everything you’re doing and fly halfway around the world, and that’s what you may have to do, right? And so the issue of right time is very important to, you know, kind of understand and deal with. It also it turns out time is money and certain things… there’s a new book by Michael Lewis called, Flash Boys, I think it’s called, which is about the fact that some of the hedge fund traders built a separate fiber line between Newark and… New Jersey and Chicago, so that they would get five millisecond advantage in trading and make millions of dollars. And it’s all in a… if you haven’t read this book, it’s fun to read [laugh]. It reads like a detection novel. So finally, right time, it turns out, in emergencies, like traffic emergencies and so on… It is known in medical literature that if you can get an ambulance to the accident site in less than ten minutes, you will save eighty percent of the people that would have otherwise died. And I know this first hand. I was, kind of, on the board of a… there was no nine-one-one in India for a long time—in two thousand and five it was set up by a thing called EMRI—Emergency, um, Emergency Research Institute. And it turns out they had to get a special number called one-o-eight, it was not nine-one-one. And… but they were able to use the latest available technology of location based services and what they knew exactly where the ambulance was and where the accident was and they could find the right one and route it. Unfortunately, in the United States we’re still on, you know, kind of living with ancient nine-one-one because the nine-one-one’s in the United States are run by counties. And if you’re happen to be in the neighboring county and there’s an ambulance right there that can get to you within two minutes, you don’t get that. You have to [laugh] get that ambulance, you know, an hour away in the county. And so that kind of location information is not there in the US systems right now. And they might change, but right now it’s not. So, coming… going to the next topic: the right language. Right language is a prefer…is the language of preferred mode of communication. So that means we’re assuming that we have a database of… for everybody that we know what their language of preferred communication is. But because of privacy considerations, this information may not be available. However, it is possible for us to create a database of location-based language, you know. So if you know where the person is supposed to be at that particular time from the service provider logs, then you can send them a message in the local language. It could be, you know, Marathi or Swahili or something else, and you would do that because you know where they are. And that… I’m not sure that such a database exists. If not, we can, I’m sure, create it and that should not be a big problem. Secondly, once you have the information about the preferred language of communication, then we need to, kind of, take the message from English or whatever… however we generated it, and translate it into the local language. And then once we translate it, it may be… it may have to be converted by using text to speech into local speech and then played out. And so, the translation systems are not that perfect. They’re still… but I don’t know about Microsoft. Google claims they have eighty plus pairs of eighty squared language translation that they can do at this point. And speech to speech translation is a little bit worse, but I think people are trying to get… make progress on all of them. The right medium, you know, basically we have two problems in the world. There is the problem of language divide and literacy divide. The language divide is, you know, from translating from one language to the other and if most people… four billion people don’t read English, or Chinese, or European languages, then text message in English is not adequate. The literacy divide is even more serious problem. There are about a billion people that can’t read anything. They’re illiterate, yeah? Right? So…so to reach every person on the planet then, you know, messages may need to be sent in all different media to all different devices. And now you’re kind of swamping the system with all kinds of data glut. So…so the issue may be if there’s a database which tells you what to… what kind of information to send where, but the… it turns out all of us live on multiple devices. Sometimes we use the laptop, sometimes we use a cell phone, sometimes we use even smaller devices perhaps. And if you knew which is the current unit that I’m using then it’s easy to send. If you didn’t have it, then you may have to send it to all of them. So one of the things I complain about is, if you’re using cell phones there are notifications you get, right? And… but I said, “Why can’t I get the same thing in my e-mail because I may not be looking at my e-mail?” And I think there’s no reason today why I can’t get the same notification, you know, that, you know, you’re just received an e-mail or from somebody or notification about some news item, but it could come as a subject line or short message subject line in the e-mail also. But that… people have not gotten to that stage of saying it should be device independent communication. And so, I think we’ll get there. And finally, the right level of detail. Summarization is an active area of research around the world. I’m sure there’s a lot of work here at Microsoft, but what is not yet clear to me is there are other kinds of summarization that we may need to do: summarization of speech, summarization of music, summarization of images, and video and art and movies, including, you know, three and a half hour football games where… which can be summarized to one hour or half hour or fifteen minutes. I believe this is being done. It is being done semi-manually. You know, basically what you do is all the key events you flag manually and then the system looks at on the video and then clips appropriate parts. But I think ultimately we need to be… we… there’s no reason why we shouldn’t be able to do all of that fully automatically. So… so the other difficult problem is context aware summarization, you know, basically different summarizations for experts and novices and summarization of movies for young children where… which might leave out some things that is for adults, you know. So basically there’s lots of interesting issues. So, all of that leads to: what is the nature… what is the solution for this problem of getting the right information to the right people at the right time frame in the right language in the right level of detail and granularity? So, what I am propo… we are proposing to get the right information to the right people, that every person on the planet will have access to a smart phone with an embedded personal Guardian Angel. If it doesn’t have an embedded software that is aware of Guardian Angel requirements, then it’s much harder to wake up this system and then provide the right information. So a definition of a Guardian Angel is a Guardian Angel is a virtual avatar that provides future-aware computation. It knows about things that you have no reason to know and it whispers in your ear that a tornado is heading your way. So what we’re envisioning is not just one Guardian Angel in your phone, but a Facebook of Guardian Angels, a society or social network of Guardian Angels, all of whom are talking to each other and know what each other knows. That means they not only know what I should know but they also… this is very important because if they can talk to each other and share anonymized information about what kinds of things have affected other people and how they were able to resolve or solve them, then that becomes input to this Guardian Angel. And so this component is part of what is sometimes called transferred learning where if somebody knows the answer, I know the answer. And I know the answer because the person that knows the answer has similar properties. He’s in the same location, has the same language or same age group, or same, you know, expectations, or whatever. And based on the similarity of interests and personalities and properties, now these Guardian Angels can share information. This makes it extremely powerful. Not only do I know what I know, not only do I know what my own agent can know, now I’m saying I know what you know and what your agents know. >>: That can be too much information. >> Raj Reddy: Huh? >>: It’s too much information. >> Raj Reddy: Too much information. >>: Yeah. It’s too high. >> Raj Reddy: No, no. But that’s where… that’s where the whole issue… I’ll come to that in a minute, but, you know, basically, you know, H.G. Wells, you know, imagined such a world of interconnected global brain and I’ll tell you about what he said in a minute. So, and these Guardian Angels, their social network is sharing anonymized knowledge and we understand how to do that with publish/subscribe mechanisms of social networking. So that… the issue then is that you have too much knowledge. I, as a person don’t have to look at seven billion pieces of information. All I’m looking for is anything happening in my neighborhood or anything happening to other people like me, and so on. That’s where we already understand the solution. Search… called search technology. You know, I have seven billion posts of something and I need to only know the few that really impact me. And we understand how to do search of very large databases of that kind, right? >>: Called a super [indiscernible] >> Raj Reddy: [laugh] Exactly. So I think this should be done, and so you’re sharing this anonymized knowledge. So, getting the right information is a big idea if… and if everybody has to get this information there are lots of issues. Just the question that was just asked about how do we filter out all the data glut and only get the right information? How do we determine what right information might be for each person on the planet? And I am assuming if I have a Guardian Angel, it’s taking care of my information, not yours and not anybody else’s. But taken together, I’m assuming there’s a huge amount of computation being expended across all the planet, in some sense. And so how do we know what is of interest to each of them? How do we communicate with the, you know, each of these Guardian Angels, and so on. So, let… there’s a whole set of design issues of that kind that come up. So what we imagine the Guardian Angels to have—they’re like an app except they’re on steroids if you want to think of it. And they’re also not like an app in that they’re always on—they’re residing on the cloud. They’re always present and always working. And they’re life-long. They, you know, you don’t kind of turn on and turn off and tap like an app. They’re always on there; they’re working all the time. And they’re autonomic and nonintrusive. They’re autonomic in the sense, you know, you know, just like your heart is beating, they’re always kind of doing the thing they’re supposed to be doing. And more importantly, they’re nonintrusive. We don’t want them to interrupt you all the time and saying this thing is happening, that thing is happening. For example, I get alerts all the time—like stupid things, like there’s a meeting reminder. The meeting reminder should not ring a bell or anything. It should appear on my screen perhaps, and other than that it… right now they go out of the way to make a nuisance of themselves [laugh], for meeting remind. If there’s an emergency, by all means, I need that. So there must be distinctions between the nature of the interruptions. The nonintrusive part is very central to the success of these things because people will turn them off completely or not even carry them if they become a nuisance. And they must always learn… continuously learning what things are similar, what things are not. And in so far as we were just talking about, when some error is made, how does a person let the Guardian Angel know, “you know, you didn’t do the right thing here?” And then how does it figure out and learn from the errors, you know? That turns out to be a very central part, you know, of these kinds of systems—always learning. And they must be device independent. They should not assume, you know, the information is being sent to one kind of device. If I’m wearing… have a cell phone or a watch phone or laptop, I should get the information on all of them. And the only issue is, ideally if it knows exactly which one I’m using, it should only send it to that. If it doesn’t know that information, it can send it to all of them. But as soon as it sees that I’ve used one of them it should wipe out the other two. So the idea is to protect me from this data glut all the time. It’s a problem that… So the question is, these Guardian Angels, they monitor the activities all the time, whatever’s happening in the world, and then they analyze all that data by searching for things that are relevant to this location, this person, or this kind of activity. And then learn from that experience, or learn from own experience and experience of others and share the knowledge with the community of Guardian Angels. So automated discovery of data and information sources becomes one of the requirements of these kinds of things. Everything I’m saying here is a nontrivial problem. It hasn’t been done, but they’re doable. They’re not something that we should shy away from. So the next task is functionality, right, of the Guardian Angels. The personal Guardian Angel must continuously derive future courses of actions from local observations. The Facebook of Guardian Angels will be able to know all the disruptive events that might affect your life. And the personal Guardian Angel can also assist you in day-to-day problem solving. It doesn’t have to be a hurricane. It can be: what is an alternative route in case of an accident and so on? So, that’s the kind of functionality we’re imagining these Guardian Angels might have. As I said earlier, every Guardian Angel is always on, continuously monitoring and accumulating large amount of episodic knowledge, right? This data suitably anonymized can be used to learn appropriate responses for each possible situation. So that means the learning technology that you have to build easier to learn preferences by observing devices, learn task similarity and user similarity, learn error correction, learn by error correction and simply learning through clarification dialogue. And all of them are different kinds of learning component that we need to build into this systems. And not everything is there now. What do we mean by task similarity and user similarity learning? How do you learn from clarification dialogue when the system makes an error or something? So there’ll be enough data in this case so that it’s probably will be one of the big… large, big data problems that the world will face. So, the next set of issues that implementation infrastructure issues, one is: I’m assuming by twentytwenty everyone on the planet has an access to a smartphone and global connectivity. This may be a tall order. It’s only seven… six years from now. It may not happen but I think it can be made to happen and I’ll come back to that issue of how… what… how can you force this to happen? But by what I mean by have access to a smart phone; if you’re in a family, not everybody has to have one. Initially, even if every family has one, that may be enough. But later on, you know, if you’re kind of rich and in a city, maybe everybody will have one. But in, you know, remote areas, even having one… the harder problem is connectivity, you know. Right now, the largest parts of the globe are not connected. You know, three quarters of the globe, which is the oceans, are definitely not connected easily, and even the rest of the globe—they’re talking about flying balloons and drones and all kinds of things to get connectivity. I think there is no simple solution and it’s… but it’s… the cost is limited, you know, bounded. The cost of getting everybody connected is about two hundred billion dollars. And it’s a onetime cost. And if you think about that and, you know, the gross national product of U. S.… United States is like about twenty trillion dollars. The gross world product is about a hundred trillion dollars. So we’re talking about connecting the whole world, two hundred billion dollars compared to hundred trillion dollars is point two percent and it’s a onetime cost and it’ll be there for the next, you know, hundred years if you, you know… The two… two hundred billion dollars is if you want to connect to the last mile with Wi-Fi, Wi MAX, White Fi, you know, one that was done here at Microsoft research. All of that, using all the technology, you can get connected. My vision is, everybody should be connected to gigabit. There’s absolutely no reason not to do that. Even wireless technologies will be there, at gigabit, if, you know, and we need to work towards it. But most of the people in cities and towns and even some villages nearby the… on the main route can get gigabit connection directly through fiber. And the reason it’s two hundred billion dollars is it’s the last mile connectivity that’s the expense. Otherwise you can do the whole global connectivity for about twenty billion dollars, one-tenth of the cost. Every country and every major city will be connected, but that still leaves out a lot of people that are not connected. Once you get the device and connectivity out of the way, which are big problems, then you also need your own personal computational resources. And I’ am assuming, given the way things are going with, you know, cloud computing and free resources—Microsoft is giving everybody seven gigabytes for free on the OneDrive today, so I immediately signed up for [laugh]… I have free memory from Google, free memory from Microsoft, free memory from Dropbox, everybody. And so I end up having like twenty, thirty GB of online resources for free. And so I’m saying, unlimited computation, unlimited memory, and unlimited bandwidth must be the goal and I believe it’s doable. If you… because every cell phone, if you look at it, the computation involved in the four NVIDIA chips that are in the cell phone is close to a super computer power of ten, fifteen years ago. Everyth… and it already has that. So if we assume everybody has a cell phone, then maybe what I’m saying it is not so unreasonable. And so the issue of, you know, cloud computing for prediction and notification, most of the work is done on the cloud, because they’re always on, always working—car… cognitive agents. All that they… you’re doing on the platform—mobile platform—is the rendering of the result. So the programming on the mobile platform becomes much more simpler. You’re not… and you program it once and then only rendering is if you have ten different devices and iPhones and android phones and Windows mobile phones. It’s not a big deal having all of them work smoothly. So the next two topics is: one is cost and the other one is privacy. The cost if, you know, if you assume that seven billion people and the cost of a device now is like two hundred billion, we’re talking about— two hundred dollars—we’re talking about a hundred and forty billion dollars. I think we can reasonably assume that it’ll be down to about fifty dollars. May… and people like Eric Schmitt talk about twenty dollar smart phones. It’ll happen, but not immediately but it may be somewhat brain dead, namely, it will not have sixteen GB of memory or not, whatever. But in… if it is fifty dollars, now we’re talking about three hundred and fifty billion dollars of cost. Now if you take my number of hundred trillion dollar gross world product, the cost of three hundred and fifty is only like point three percent onetime cost. And if you, you know, assume that you’ll use the phone for four or five years, it is less than point o one percent, you know, point one percent. So we should not scared… be scared away. There’s another reason for it, not to be scared away. The estimate is: the growth in economic activity by the networking effect of seven billion people interconnected, if one billion people are connected and then go to seven, you’re talking about seven squared impact, you know, fifty times more activity. And the predictions are, I don’t have exact numbers, there’ll be economic growth as a result of this connected network effect of between ten to thirty percent of gross world product. That means by investing three hundred and fifty billion dollars plus another, let’s say three hundred billion dollars, two hundred billion dollars in connectivity, you may be creating economic activity of thirty trillion dollars. So it will kind of, more-orless pay for itself in one month. If not in one month, six month or a year. It’s still worth it, and it’s worth it even if it costs more time than that. The second one, the second big elephant is the privacy. Namely, we’re assuming these Guardian Angels can share information with each other. The question is: what information are they sharing? Do you want everybody to know where you are or where you are sleeping that night? Maybe not. And so the question is, you know, what you’d assume, you and your Guardian Angel are one and the same. Anything you know, they… it knows. But just because it knows—it’s like your consciousness, knowing what you know—doesn’t mean your privacy is violated. The question is: what information can it genuinely share with other soc… in the social network of Guardian Angels so that others can also benefit from it, and so on. That anonymization research has not been done fully… we do anonymization now, but… and what I’m saying is at the worst case, I already know where every cell phone is through-out because that information is known to the cell providers, service providers. It is logged. It can be found out by… subpoenaed by the governments if you want. In that sense there are certain things that are beyond your control. But it may not know precise GPS location of where you’re sleeping that night, but do you know… it knows within one square miles… kilometer, or a few square kilometers where you are. That may be enough. As for most of the events like tornados and earthquakes and typhoons, that may be enough. But it would be better if we can get more crisper information. And the other thing I’m saying is, there is no breach of privacy since each Guardian Angel only shares information already known to service providers and other nearby agents. You know, whatever is happening to you is also happening to your neighbor and so it’s not as though the information that is being broadcast is personal information, you know, saying who else is in bed with you, kind of thing. That’s not the issue here unless, of course, there are two cell phones in the same room and they’re sharing their information. Even then it’s not clear that anybody else needs to be, you know, told that information. So, so there’s not much… I believe it’s an acceptable level of loss of privacy that’s already… we’re not doing anything new, it’s already gone because the government says everybody must… the location information must be kept and locked. That’s part of the current system. So this brings me to the end of the talk. Basically, there was a very, you know, kind of visionary book by H. G. Wells about a hundred years ago—maybe eighty years ago now… eight, in eight… nineteen-thirty, I think it was, sometime around—where he imagined a global brain. This global brain… he was a synthetic super-mind within the species of Homo sapiens. And, you know, he envisioned this super human intelligences to represent the ultimate unification of all knowledge, of all the memory… and memory of all the people. I think the way we… and I… the instantiation I think of is, I think now we can accomplish that by connecting our Guardian Angels. We don’t have to connect… put, you know, electrodes in our brain. We just connect all the Guardian Angels into your social network and they will be able to monitor, analyze and learn from their own experience and experience of the others and share anonymized knowledge through publish/subscribe mechanisms. Finally, so the grand challenge for the twenty-first century as I see it is the creation and deployment of Guardian Angel technology for providing the right information to every man, woman, and child on the planet. And this will be a disruptive technology if we succeed in the IT eco system because right now at Microsoft, everybody’s thinking about the billion people and what they can do for the billion people. Nobody is seriously saying, “If I have seven billion people I am serving and they all have different needs”—they’re not the same needs as the most professional billion people that use the technology today. It’s the same thing that happened in the sixties where you were having main frames. In the seventies we had minis, in the eighties we had PCs, we… in the nineties we had network, and twenty to twenty-ten we are having mobile systems. Now the issue is, as you go to this newer system the requirements of computation become completely different and we need a different mindset to think about, “what will those people need and what can we work on now that will actually be truly revolutionary at that time?” So, I… my prediction is, given… we are willing to spend point one percent of the global world product, the… we will save fifty to eighty percent of deaths every year. The number of deaths, accidental deaths globally are about two to three million every year. A large number of them are automotive accidents. I think the self-driving cars will eliminate a lot of them already. But, you know, the… I think even things like earthquakes and other natural disasters, annual number of deaths are like ten, twenty thousand, which, you know, by the time you add up—that’s in the United States—if you add up the whole world where they don’t have as good warning systems, the… it kind of accumulates. So if we can eliminate fifty to eighty percent of those deaths, accidental deaths, that would be fantastic and I think it is worth doing. And it’s … my view is, Microsoft should adopt this project and do it. [laugh] Okay, thank you very much. [applause] >>: [indiscernible] Do you have the [indiscernible] of government and other companies? >> Raj Reddy: Yeah, basically. Let me go back there for a minute. So basically, what I said here was there’s a way of making it happen. And as I—you know—there are four stake holders. The government is one of them, the manufacturers of the phones are the second one—T-Mobile, Nokia and all of them— but telecom service providers—Verizon and AT&T and China Mobile and Airtel and so on—are the… and finally Microsofts and Googles of the world that would benefit from having all this infrastructure. I’m saying each of them should kick in one-quarter of the cost to the cell… the smart phone, which is nothing and… because you’re saying, “I’m going to give it away.” My proposal is: don’t give it away, rent it away. Or simply say, “We’ll give you a free phone, a starter kit. After a year, you’re now more sophisticated; you want some better one, we’ll sell it to you for a small increment in cost.” Now you’d recycle the old one to somebody else who doesn’t have one yet. And so the… you’re creating a market space where this whole thing can be done. Connect to it is a different matter. That requires all the governments to come together. There is… was a thing called Intelsat that was created in the sixties—I don’t know if you’d remember it. >>: I thought that this… Facebook is doing this to get this from [indiscernible] >> Raj Reddy: They’re buying the drones. They bought a drone company or something. >>: They might be already shared [indiscernible] >> Raj Reddy: No. No, no. Basically Google and Facebook and everybody’s struggling to find a solution to the global connectivity problem. They all see this is coming. They all see that the smart phones are going to go from two hundred, three hundred dollars to fifty dollars to twenty dollars. And they all see that whole market will open up, instead of just one billion or one point five billion people now connected, it will go very quickly by twenty-twenty to four, five billion. By twenty-thirty, fifty, a hundred, you know… all of them, whether we do anything or not, whether we do Guardian Angel technologies or not. And they’re kind of saying, “What is that market? What do we need to do? How do we connect up all of them?” And that’s why, you know, Facebook is willing to pay, you know, seventeen billion dollars for WhatsApp because it gets them connected to a lot of people that don’t have smart phones. They just use messaging. >>: Well, it was just to leverage their investment, and then on top of that, we can do all this stuff without having to put that cost into [indiscernible] or putting here. That’s a huge amount of cost. >> Raj Reddy: Right. >>: Even if you cannot reach seven billion people—just one billion people have connectivity—you can really actually make that Guardian Angel useful, right? >> Raj Reddy: Yes. That… even if you connect to a hundred million it’ll be useful, so you can begin de… you know, prototyping and testing the technologies right away. All I’m saying is if everybody’s connected, then there is this networking effort you… basically, even if I don’t have information about what—you know—your own situation, if I know everything happening in the neighbors, then I can assume you also have… going to be affected by the same thing and notify you too. And so there’s a whole set of, you know, related issues of how to scale up the thing, but… >>: Yeah. And then another important challenge that you mentioned earlier, you know, is other people knowledge become part of your knowledge too. >> Raj Reddy: Right, right. >>: That’s not obvious how to do that. It requires kind of intelligent to figure out what part of knowledge is useful for you. That’s not really research issue; it’s not really… it’s not… >> Raj Reddy: It doesn’t have to… no, it doesn’t have to be useful. All I am saying is if you know a hot stock market, you know, [laugh] you want to invest in, tell me about it too, I might invest. But the issue is, you may not want me to know. >>: I see. >> Raj Reddy: You know, but in a… some… because of insider trading laws, some peoples trading, you know, things have to be public, but the rest of us don’t have to be told, you know, but... And so there are things that you may or may not want to know, but what I’m amazed at—surprised at—is what’s the so-called Facebook effect, right? People seem to want to share even the darkest secrets though [laugh] with everybody else. You know, I don’t understand it, but that seems to be the new ethos in the system, you know, and I’m surprised… >>: [indiscernible] >> Raj Reddy: Eh? >>: People tend to be more liberal in terms of sharing >> Raj Reddy: I’m not sure liberal. They want, you know, to shock—shock value, maybe, but I have [laughing] no interest at all in sharing what I’m doing and not doing. Not bec… and I… Not because I… there’s some secrets in there. I don’t think it’s interesting to anybody else. Why would they care? [laugh] >>: Now you don’t know until you share. [laughter] >> Raj Reddy: That is true, that is true. Interesting… I’m told the reason why so many people spend so much time on the Facebook is this voyeurism effect—namely, they want to know what is everybody else is doing, you know. It’s just the curiosity of what everybody else… it seems… an interesting. >>: They want to be popular too. >> Raj Reddy: Perhaps. >>: They post those things about themselves, so hopefully they can get popular. >>: So, this entire personal Guardian Angel thing is really nice, but I have one slight doubt in my head about it when I think, and it’s that it’s very good to address… so if you just take the needs of every person on the planet… >> Raj Reddy: Right. >>: It’s going to be very heavy data distribution. And for me, when I think about it, it’s a very good instructor to actually address just the commonality part, but not the more personalization or the more heavy data part. And I think getting to that is where actually becomes really infeasible and becomes really, really more challenging. >> Raj Reddy: Right. >>: Just getting to this particular part—just alerting about disasters, of these sort of things—it’s fine. But there are just certain things which very, very few set of people just get around… >> Raj Reddy: Right. >>: …and then addressing that becomes a huge issue later on. >> Raj Reddy: Basically, what you’re saying is very true. That’s how the current IT market is behaving. Basically, you look at what Google now does. All that they do is they will tell me about some restaurant or something, you know, and the things that they think everybody wants to know. My view is: the long term eff… you know, benefit is heavy personalization, mass customization of everything so that every person is able to get personalized knowledge and solutions to their problems. And I believe we are at a stage we can begin to think about it. Now what you’re saying is the tail may be, you know… but it’s only if a human being has to program it, it becomes a big deal. If I can invent autonomic… in a system that’ll learn with experience and share that experience, then it may not be so bad. Even if only the… it’s like the Usenet, you know, people in the communities that… fifty people with a common interest get together from all over the world in the Usenet because that… they had interest in that small topic, and in that sense I think we—you know—it’s both an opportunity and a need to kind of serve the tail. And if we can figure that out to serve the tail without requiring lots of people sitting and coding this stuff, then I think we’ll get there. Okay. >> Xuedong Huang: Okay, Thank you so much. [Applause] >> Raj Reddy: Thank you very much. Yep.