 I think that we're still at the very beginning of a decades-long trend toward increasingly evidence-based data-intensive decision-making across really all aspects of life, including medical care but through to education. And machine learning applied to big data as well as driving this trend. One of the interesting developments in machine learning is a new class of software agents that learns continuously. For example, at Carnegie Mellon, we've developed our never-ending language learner, NEL, which is learning to read the web. And NEL has been learning now continuously for five years, improving its reading skills weekly, and now building up a collection of 100 million beliefs that it has read from the web. Beyond reading, NEL is also now learning to reason, to infer new beliefs that it hasn't read before. Those beliefs are inferred by rules that it has learned, hundreds of thousands of them, including one that says, for example, if some athlete plays on Team T and Team T plays Sport S, that athlete probably plays Sport S. So NEL is one example of this new kind of continuous learning system that requires only limited human input. A second example, also from Carnegie Mellon, is our never-ending image learner, Miele, which is continuously learning to analyze images that it finds on the web. Now, recently, NEL and NEL have begun to routinely communicate so that they now form a team of learning agents with complementary expertise. So how will we apply machine learning to health data? Some suggest that we should form a global database with all the world's medical data and then analyze it. But I think not. I think privacy issues and competition among organizations make a single global database impractical. And instead, I think we may see the development of many never-ending learning agents specializing on different patients, diseases, hospitals. Like NEL and NEL, these agents will learn continuously and collaborate. One class of agents will focus on patients. Each of us will have our own personal assistant that tracks our medical history throughout our lifetime. It'll know about the diseases we've had, our treatments, our genetic makeup, allergies. And it will use this data to act as a kind of patient representative, an advocate for us, providing our data to doctors when needed, but also sharing that data with personal agents of other people in order to find the most similar other patients and to use their medical history to suggest treatments for us. So as a second kind of agent, we will also have agents that focus on diseases. For example, if I come down with diabetes, my personal agent will contact the diabetes agent, providing my clinical details and asking for advice. The diabetes agent will then assist and gain some more training experience about its disease. A third class of agents will focus on individual hospital rooms. It will use sensors like cameras, microphones, patient monitoring equipment to build up a model of exactly what are the routine things that go on in that room. Then it will use that model to monitor the room 24 hours a day to assure that standard medical practices are being correctly followed, to sound an alarm if the patient is not responding as desired, to sound an alarm if a hospital orderly accidentally brings in the wrong medications. So just think of the collection of experience that these different agents will collect over the 24-hour-a-day operation. They will actually collect more experience than a human nurse and doctor could ever in a 100-year career, and they will be the most experienced agents in our healthcare system. So the challenge for us is to find a way to convert this growing volume of medical data into improved medical care. Some suggest a global medical database, but privacy, trust, and national boundaries raise roadblocks. Instead, I propose to you that we may have a better solution with a collection of learning agents. They can deal with a privacy issue in a way that we already deal with it in the medical system. Each agency is only the data that it needs to do its job. And as these agents learn from that experience, they can provide doctors and nurses with more informed advice which allows them to provide us with more effective medical care. Thank you.