Epidemics are caused by pathogens maintained in animal dna. Theyre difficult to monitor because if you take ebola, for instance, its believed that ebola has a reservoir in egyptian fruit bats. How do you monitor in the environment the population of egyptian fruit bats and whether or not they have ebola . Its a very hard thing to do. So we imagine using a mosquito as a device because they bite all sorts of animals, taking little blood samples. They bite everything from animals from amphibians and reptiles all the way up to humans and chickens. Host so why would microsoft be working on a project like that . Guest well, when we think about Machine Learning, when we think about robotics, one of the things we want to understand is what how is it going to be useful. And disease epidemics are an example where we know getting the right data is incredibly challenging, analyzing the data is incredibly challenging, so we believe there are some important Inflection Points. Host all right. Walk us through your model. Guest right, yeah. So in order to do this, to use a mosquito as a device, we basically need to collect interesting mosquitoes at scale, and it turns out collecting a mosquito is harder than it sounds. We imagine we go outside and we just immediately get bit p by mosquitoes, but thats not how it works. It isnt that high, the density, and they sneak around and creep you and get you when you least expect it. So the First Technology that we needed to build was a robotic field biologist that could sit in the environment. This one stayed in the environment for over 20 hours to understand the insects that are encountering it and selectively captures the ones that are interesting host whats an interesting . Guest one like this, this is called ad. C. Egypt tie, known to carry zika, and it is an example of a mosquito that in houston where we collected it is very interesting, because houston is near places that could have zika, and so you want to understand whats going on with that particular mosquito. When Something Like this flies into this trap, theres a sensor that looks at how its wings are moving, understands the shadow and uses that as a fingerprint to tell it apart from other insects and other types of mosquitoes. At that moment it can make a decision about whether to capture it or not. And the flicking sound youre hearing i dont know if you can hear that is this trap capturing another type of mosquito host so thats the trap inside there. Guest and whats flying around in here is a type of mosquito that carries malaria in india. So these things are buzzing around. The trap is [inaudible] as a counter, the trap is trying to, its analyzing those wing signatures, the shadow that they make, and they can make clues about whether to capture it or not. Takes about seven milliseconds, and one of these little trap doors swing shut. At the moment we capture the sample, we get all this digital data that goes along with it; what its shadow looks like host why dont we just kill them . Guest right. So people try to do that, and thats mosquito control. The Public Health tries to control mosquitoes, but it turns out its very difficult to do because you have to apply things like insecticides, reduce their habitat at the right time. Part of what [inaudible] give you much more insight into what are the right places and what is the right time, because we know exactly when they showed up, and we can even build predictive models of when they will show up in the future. Host so, mr. Jackson, did you import these mosquitoes from india, and are any of them carrying malaria . Guest no, none of them are carrying malaria, these are laboratory mosquitoes. They are very safe. And, actually, if they leave this chamber, because this room does not have the climate of india, they will die quickly which is why theyre in this nice humidified host just to bring it, is there a. I. Attached . Guest absolutely. Theres a. I. In this device in ch in realtime tries to make a decision about whether its something interesting. And as i mentioned before, its looking at the movement of these shadows, it has information about when and where these things showed up in the past. So its really thinking about whether it needs to capture that insect or not, and thats the a. I. Component thats actually on this trap running in realtime. Host this machine here in the middle, this cylinder, whats it communicating with . Guest so this is a selfcontained so to a smartphone right now using bluetooth host okay, its being recorded . Guest its being recorded. When you bring it out of the environment, that data set is all selfcontained. Just to demonstrate a. I. , im going to do something a little crazy. Im going to put my hand in here, and im going to show you its probably going to mix up my hand with mosquitoes host so should we keep an eye on the phone . What should we keep an eye on . Guest keep an eye on the phone. Host okay. Ill keep this up here. Guest im going to shoot for that column labeled four. Host right in the center. Guest im going to poke my hand in some of those cells, and youll see them flash. Host and were trying to figure out if youre a mosquito or not . Guest trying to figure out if im a mosquito. Host now, youre turning up red every time you hit something. Guest the red means that looks nothing like a mosquito. You can see some of these other cells that are closed, things very much that look like mosquitoes showed up. Host whats it going to do with those mosquitoes when it captures them . Guest we want to understand what theyve encountered in the environment. And we dont want the presume we know. Zikas an example we didnt know we needed to worry about this. We want to turn this into data. We do that by gene sequencing. All of the genetic material colocated with that mosquito and turns it to digital data. To put that in perspective, from a set of mosquitoes, we get several hundred gigabytes of Genetic Information that tells us about the mosquitoes themselves, blood theyve consumed, bacteria, a big mixture of all sorts of organisms, and now we want to computationally reconstruct that. We want to take that several hundred gigabytes of data, compare it against all known [inaudible] life forms which is over one trillion pieces of Genetic Information and give you back a recipe that says you have, you know, 90 mosquito, 5 human dna in that mosquito, lets say 1 malaria parasite in that mosquito. If youre very unlucky, some presents zika in that mosquito. We want to, in an unbiased way, scan for everything. And thats what we do. If you actually just take a look close there, this is an example of an ad. C. Egypt tie mosquito from houston, texas, and we see we reconstruct a ratio of components. 90 , some percent homo sapien, a virus in the same family as rabies but very different, so it means its most likely a novel virus in the same family. Nobody really understands yet what this virus is, whether thats a threat or not. As we do these experiments more and more, we start to correlate known and unknown viruses with particular mosquitoes and particular hosts trying to triangulate what that virus is. Host another look at how microsoft is using technology. Mr. Jackson, thank you. Guest thank you so much. Thank you. Host and now on the communicators we want to introduce you to dr. Eric horvitz of microsoft. Dr. Horvitz, whats your role here . Whats your title . Guest my title is managing director of Microsoft Research lab. Host and what does that mean . Guest it means that in my purview are several laboratories across the world. Its the core of what we call Microsoft Research. The fundamentals are research and Development Technology labs for microsoft corporation. Host so when we laymen hear about r d, are you the r and the d . The r part of that . Guest were the big r in r d. We do both and work closely with the robotic teams as well as [inaudible] and innovate over the horizon. Host what kind of projects are the labs working on right now . Guest boy, we span the gamut of what are called the computing sciences. Even getting into, like, biology and affiliated fields that rely on computation. These days were focusing spencively with extensively with a hard push on Artificial Intelligence research. Theres been an Inflection Point in this technology. Its going to be central in our mission at microsoft, which i love, by the way, to say and i love in particular empowering people and organizations to achieve more throughout the world. Its a very inspirational statement, and you look at a. I. , Machine Intelligence more broadly, Machine Learning and automation and planning, language technologies, speaker recognition, perception and so on as really being harnessed to empower people and organizations in new ways to to get more done, to inspire, to visualize. Several of the technologies were showing today here in washington are at that notion of how can we empower people in new ways. How can we change the way we do agriculture, for example. How can we communicate, how can we visualize complex data sets that are [inaudible] throughout the United States in cities, for example . Host whats the definition of a. I. . Guest a. I. , i particularly like one which says its a science of intelligence. Were looking to understand how it is that people think and get things done, how can we understand and is harness those principles of intelligence and computational systems. Some people say that a. I. Is doing things that we often in the past weve required people to do k and that was one of the early definitions in the 1950s when the phrase a. I. Was first used. Host how long have you been working in a. I. . Guest since 1984, believe it or not. Thats why my hair is gray, and im so impassioned to get things done host in 1984, what was that project you were working on . Guest our team was one of the early teams at Stanford University looking at moving from core, logical methods, from a to b to uncertainty using probability. Precision science to grapple with the complexities of the open world. That paradigm which back in those days was small bunch of rebellious grad students became the dominant power rah dime in a. I. Now paradigm in a. I. Now. How to take data, process it, predictions, make decisions and advise in the broad, open world. Now, the open world is typically far more complex than the computations in the machines that we build, so machines have to be humble and understand what it is they dont know as well as what they do know to make progress in the world. Host now, you mentioned earlier, dr. Horvitz, theres been an Inflection Point in Artificial Intelligence. What is that or what was that . Guest well, one thing i want to make clear to your viewers is that the field of a. I. Has been making advances continually for almost 45, 50 years now. However, the public tends to see these kind of salient demonstrations like jeopardy being won by a computer or a chess game, the chess leader of the world being beaten by a machine. The advances have continued. Now, i say Inflection Point, think whats happened is in the last id say 15 years the combination of incredible data resources becoming available now, and they werent. People dont remember that in the late 80s and early 90s, a hundred megabyte drive was hard to come by. All of a sudden we can store data almost for free, and all of a sudden almost all of human discourse has been digitized. That together has given us incredible quantities of data. That data has become the fuel of many algorithms including Machine Learning procedures that learn to predict and learn to act based on that data. So secondly, computational resources at the same time have really increased. Its pretty clear that the smartphones in our pocket today are along several dimensions as powerful or more powerful than those Gray Computers you saw in the 1980s with the blue cushion and the cooling towers. And now theyre in our pockets. So you take the computation, plus the data, plus the incredible applications and a competitive terrain, you have what i would call an Inflection Point. Host last question. What are we, on a daily basis how are we using a. I. . Whats an example . Guest oh, in many places. For example, i drive to work maybe half the way under complete automation in my car now. Other places that we see its under the hood. So many people dont know that, for example, in windows 10 operating system theres Machine Learning going on in the os always guessing what youre going to do next and precomputing the jump ahead to make the computer smarter and faster. Those under the hood applications are everywhere right now. You see in places like hospitals theres incredible opportunities to predict outcomes and to guide patient therapy in a Cost Effective way to where its really needed. Weve got an incredible project here with medstar hospital in d. C. Our team for several years now looking at how you can take the large quantities of data theyve collected and work with them to build models that can predict whos going to to be readmitted after 30 days, and if you knew that in advance whos going to come back to the hospital in 30 days, you could do special things for those patients that would be too costly for all patients. Host eric horvitz, thanks for your time today. Guest thank you. Host and now on the communicators we want to introduce you to William Mendoza of microsoft. Mr. Mendoza, here at the tech fair what are you demonstrating . Guest yeah. Im just demonstrating potentially how microsoft views Artificial Intelligence and cognitive capabilities can be used to make Customer Experiences better whether its for a company or whether its for an agency or some sort of organization. Host all right. So what is your definition of a. I. And cognitive and then your definition of cognitive ability . Guest Artificial Intelligence is a really complex term, but specifically you think about what humans do which is, you know, vision, being able to recognize something or hearing manager and understanding what it means. So i would host learning from it . Guest and learning from it and being able to make intelligent actions based on past experiences. Host so has microsoft developed software that is starting to do that . Guest exactly. So weve got Machine Learning capabilities. We offer Cognitive Services capabilities available on the microsoft target services web site, and we also offer frameworks in order to help companies and organizations build out chat bots that can help facilitate and make customers experiences better. Host all right. So what have you got here . Guest in this example i am visiting a Car Insurance web site. Im a parent, and i basically want to insure my second car. Its the company i already have my first car insured under. Guest i say, hey, i need insurance. So itll pick up on insurance as an intent. And so itll offer me the various options that this company offers. Host so, in other words, youre not talking to a human being at this point. Guest thats correct. Host this is the software thats responding to you. Guest correct. Host okay. Guest and insurance is the key to that. It says are you a sitting customer, and in this example i am. It asks me for my full name, so ill pretend that im lance for the moment, and itll ask me for my pin number. This is the traditional way of logging in. There is capability too, as well, to train, to train it to recognize your voice. And so you, youre given a few prompts, and so you say this with your voice, and itll learn what you sound like, and so you can log in like that. If youre anything like me, i always forget my passwords and logins. In this case it says, okay, now, this might be alarming, but its actually the case that the customer has already provided this information, and in fact, im logged in at the moment because i gave it the information it needed to pull me up in the system. So i do want to insure because this car is actually for my daughter. And despite my best lets see. Here we go. So now we will, lets say we go with the ford. I went to michigan undergrand, so ive got to stick to it [laughter] is so now i say its a ford taurus. And then finally, this is gathering information is so that it can present to me a quote. Ill say a 2015 because i dont want to fully hamper my child. Do you have a photo of the car, and ill say yes. In this ill you load a photo, and just to keep things interesting, this is demonstrating Computer Vision capabilities. Now its having me to upload an image, and ill to so, but i will throw in a noncar to see how it reacts. Glx okay. Guest ing so i threw in harry potter, and not only did it recognize it, but it also recognized the celebrity, daniel radcliffe. Its not perfect, hes not holding a sign, hes holding a wand, but the computer hasnt been taught to know the difference between a wand or a sign. So i actually, you know, im a good customer and i want to be honest, i can upload the picture of a ford. And as i do that, youll see, itll say, okay, ive got all the information that i need to provide you this quote. So now itll take that back, and itll start thinking. Just governor to give it a just got to give it a moment. As its thinking, its pulling together information about me to figure out what it would charge me. Now, lets say i get this 150 quote or 146. 50. I say, no way. That is way too expensive. Now, if you think about that, like, im not just saying no, im really, like, offended that thats so expensive. So now this is kind of where the bot has done all it can, and its going to bring in a human being host so jane is the human being guest now jane will have this where she can have this dashboard of dynamics, and so she can see this live chat and continue to work that lead but also give recommendations saying, hey, look, you know, this person was at a racing of 75 insurance, heres how we can get them back in our good graces. Host okay. Mr. Mendoza, were here in washington. How do you see Gove