You going to kick off this weekend with the book the master algorithm where he takes a look at the future. This is booktv. On tonights guest. For the Computer Science and engineering is a fellow Association Association from the advanced Artificial Intelligence board of the learning journal and cofounder of the learning society and the winner of the intervention award. Hes the highest author on Machine Learning data mining and other areas. Tonight joining us for the purchased new book the master algorithm how it will remake our world. Please join me in welcoming. [applause] thank you all for coming. I wanted to you about the case that is happening today. The intranet, the personal computer and in fact it goes to all of them and touches every part of society and affects everybodys life and touches your life right now in ways you are probably not aware of. Fortunes are being made because of it and there are also jobs in many cases unnecessarily. Children have been born because of this change. This change. Machine learning as computers, learning things by themselves. Its the mission of discovery and its computers Getting Better over time the same way that humans do. Its a Scientific Method on steroids making observations and forming hypothesis refining the hypothesis and there is a result in any given period of the time they can discover more knowledge than any scientist could. The knowledges and the relativity so far it is more modernday knowledge that this is what our lives are made up. What do you search for on the web. When you go to amazon what do you buy . If you are a company that helps you. If youre an individual, Machine Learning helps you find books to read and movies to see, a job, a new house and even a date. A third of all relationships that lead to marriage today began on the internet and its Machine Learning algorithms that pose potential dates to use with there are children alive today who wouldnt have been born if not for the Machine Learning so perhaps it isnt so mundane after all. Let me give an example, the smart phone in your pocket right now is chockfull of algorithms. It uses is to correct your spelling errors, to predict what your going to do and make suggestions and used as the gps to figure out what the routines are and can combine that with the calendar whether you are doing it for meetings to act accordingly. All of this is happening today with the hope of Machine Learning and as we progress, things like cell phones are learning more and more about you and as a result hopefully to better serve you as well. With all of this economic value in Machine Learning its no wonder that Tech Companies are all over it. They are investing heavily no customers and no products just because it has been a learning algorithm that if you google and alerts you predict what people will click on 1 better than before, that alone is worth over half a billion dollars every year. As another example, just this month ibm paid a billion dollars for a medical Imaging Company not because of what it does, but because they want to have access to his medical images so they can train learning algorithms to automatically diagnose Breast Cancer. The kind of thing thats today takes highly paid doctors to do and Machine Learning experts are very highly sought after. The character of research at microsoft says that the package would you pay them a deep learning being a hot area in Machine Learning is similar to what you pay for the top quarterback. So theyve won, finally. Now, well and good as all of that is it often has a downside which we need to be aware of. Machine learning is what is behind increasing automation so it could cost your job. Some people say they use Machine Learning to spy on you. We dont know, but it could lobby. Machine learning may lead to the terminator, evil robots taking over the world. So Machine Learning could be your best friend or worst enemy depending what you do with it which is why we are at the point that everybody is to have a basic understanding of Machine Learning is and what it does and thats why i wrote a book about it and im going to try to talk about them tonight. Its not that you need to understand the details of learning algorithms. You need to understand how the engine works, for engineers and mechanics, but you need to understand what to do with the Steering Wheel and pedal. So today you dont even know where the Steering Wheel is aware that such a thing exists. So we as a society have a lot of decisions to make regarding Machine Learning into the decision is to be informed more so than it has been in the past. So how is it that computers do this thing of learning from data . It may come as a surprise to you thats even possible. Computers are supposed to use these machines we just programmed to do the same task over and over again into discovering new knowledge to require a lot of intelligence. Picasso said computers are useless because they only give us answers. Machine learning is what happens when computers start asking questions coming into the questions a computer asks when its learning is how do i turn this input into this output . For example lets say the input is an xray into the computer is trying to learn is to say there is a tumor were no there is no tumor there and it does this by looking at a lot of data. In the old days before there was Machine Learning this is how things worked. The data might be a Breast Cancer example or the output might be the diagnosis. And the algorithm is the sequence of instructions that tells the computer exactly what it needs to do. Its like a recipe from the for meatloaf, except it has to be more detailed than a recipe. Machine learning, Machine Learning turns this around. The output instead of coming out also goes into figures up figures up to 10 p. M. Put into the output is the algorithm that i need to turn the input into the output if this is the image of the breast and this is the diagnosis, how do i get the answer and the amazing thing about Machine Learning is when you program a traditional way it takes human programmers to do it in and for every application to write this program, one program to do Credit Credit scoring and wanted a Breast Cancer diagnosis etc. But in Machine Learning the same algorithm as all different kinds of knowledge that have been learned by learning algorithms. So the holy grail of the Machine Learning researchers is to discover this master algorithm. The single algorithm by which you can discover all knowledge if you get the appropriate data. If you give it without credit risks if a lender the diagnosis and so on. What could this possibly be . Then i will talk about the future consequences if we succeed in this enterprise is what is going to make it possible . Part of what makes Machine Learning interesting is that all of these algorithms come from interesting origins and for each of these there is a school of thought who pursued at. Theres five of them and you will see what each one does. With ideas is built on. So its going to do a lot of the different fields. So theres the symbolist who in some ways they have their origin was learning as the induction. Then theres the connection us whose big idea is to reverse engineer the brain so what they do is try to implement the brain on the computer and they get that from science into the maps are something very famous and well see what the idea behind it is. The evolutionary say no, no. Of the greatest algorithm the greatest algorithm on earth is not the greatest evolution. The evolution meet of made the brain and the rest of you and all of life on earth and biologists to regard the algorithm and then its called genetic programming. Then theres those that have their statistics and dniester algorithm and well see what that is as well and then finally there are the analyst risers. That have the idea that all learning and intelligence works by an illogical reasons by looking at similarities and so on. And the most famous in this line of work we will see what those do. Lets start with a visit. Here are three of the most prominent in the world. Ross is actually the alumni on the committee by the way. He was the First Teacher of Computer Science that was awarded. So what is the idea of the symbolist . Its this idea that we are going to learn in the same way that scientists discover things by trying to fill the gaps in the knowledge by reasoning and formulating hypothesis into the basic idea is this notion of inverse induction and how do we do things. It can be deduction goes from the general to the specific and induction is from the specific to the general so we can think of it as an inverse of deduction as subtraction is that deduction of addition or the square root is the inverse of the square. For example, addition gives the answer to the question just by asking what do i get. This obstruction is the answer to the question what do i need to add in order to get to four . Deduction is how you go from human and humans are mortal this is going from the general knowledge about humans to the specific knowledge about properties and the inverse sense if i know its an idea that socrates is human and i also know that he is mortal, what am i missing in order to go from one to the other and of course the answer to that is i need to know that humans are mortal. This is written in english and computers dont understand english just so they are written in formal logic but the general idea is the same venue can learn these rules from the different data which is now combining the ways to answer new questions to things youve never seen before. Heres an example of what you can do with symbolic learning today. Guess who the biologist is a picture is. Its not that i in the lab coat, that is a Computer Scientist i was just talking with the other day and the other guy is not a biologist either, its a machine that the robot that knows about microbiology in the dna and protein and is making discoveries about metabolism and so on and so forth and runs the whole process by itself. He uses this method of inverse direction to make hypothesis and then literally carries out the extent with dna sequencing and whatnot to test hypothesis and like the scientists may be maybe they throw them away and keep going. Right now theres only two in the world they call them at him and eve. They discovered a drop that isnt being tested and once you have two computers its like having a Million People doing medical research so this is a very powerful thing to be able to do. Lets see what the connectionist s are doing. Here are some of the most prominent. He is someone that started as a psychologist and became a Computer Scientist. Ever since the 70s his goal in life has been to understand how the brain works and he has persisted through thick and thin and made a lot of progress although as hes had. He tells the story of coming home from work one day very excited and say yes i figured out how the brain works and his daughter says not again. [laughter] another famous one is very much in the news on the front pages of the New York Times because the success that its happening and keep learning. Its a modern name of connecting or networks as they are also known because the idea of a connectionist is to be inspired by the human brain. So how do we build learning based on how the brain works . The brain is made up of neurons and vendor roots are dendrites and it branches off. So its a little bit like a forest with one big difference the roots of one tree connect with the branches of others and then theres this electrical impulse like this big electrical storm happening. Its an electrical storm encloses continually happening. The connections might be stronger or weaker. The stronger the connection between the two is the more likely the first is to help the second fire. When the amount of charge with the amount of electricity coming in and exceeds a certain threshold and then the fires it can make others fire. The brain is a network of 10 million of these all working together. So the way we are going to do this is number one, we are going to do the symbolist mathematical model that we can out of their neuron works and for example this could be the pixels on an image of a breast for Breast Cancer diagnosis and then what we do is multiply to represent how strong the synopsis is and then if this average is the input if it exceeds the threshold and the output is gone the deep learning comes from the idea that they have manual input how did you learn those . If you think about it its a tricky problem because here is the image but say and its supposed to be one but its not porn, its pleased to so it needs to go up into the question is theres a certain era and now how much was in each of these responsible which is accordingly so there is the function of the input for the also have to change into the air waiting for it to go down and others to go up and so back as we saw in the call itself in this problem and there is some evidence that some parts of the brain it is innocence of the Machine Learning that credits the science problem which is when something goes wrong who do you blame if this is the right answer nothing needs to change but when its the wrong answer something needs to change and what that does is propagates the network to figure out where the synopsis needs to change. Now deep learning is responsible for the processing of the visions for this type of Machine Learning. And for example this is how google does the image search both image understanding of speech understanding and they all use this type of learning. But but one famous example that was in the first page of the New York Times is what has come to be known as the google cab network which was the Largest Network of her belt and it has on the other end of a billion which is still small and then what they did is trained to network on the videos. Basically just sit there watching videos for a long time so maybe you should call it the couch potato network. Then people to upload into beginning is the Golden Network and also the center recognize dogs and a lot of other things. The evolutionary say where did the training come from . It was evolution that accomplished that and your in your body and Everything Else is low. John holland died recently for the first couple of decades people used to say that evolutionary computing than in the 80s it took off. They proposed genetic component that evolutionary learning. Its to simulate the process of evolution. We understand ever since darwin and so forth so what happens in evolution in a very schematic way each individual has a genome and they go out in the world and get killed or you have many offspring and so the world will assist tremendous of some individuals turned out to maybe the drafts of a longer neck because it can help them reach and so the individual with the highest fitness gets to produce what do crossover sexual reproduction genes of two successful individuals and the more likely you are up to participate. It also happens because the dna that turned out to be good so after that happens you have the process that repeats again and then it takes you to a human being. They implement us on the computer and this is just on a computer so its like having dna. Otherwise the process is very smart and theyve been able to quite Amazing Things with it or between el paso to her and this is often taken in patterns for the circuits instead of the conversation we would already have passed. The more advanced type of genetic algorithm is genetic programming and the idea here is that why you do this with strength this is a low level of representation. And the program is like multiply c. And we can do the very rich programs except now the crossover and you start with one string and then cross over to the other. You do that with the program but for example if you take this point and this point for crossover than one of the offspring will be the tree in white with a tree in black. Its the formulation as a function its a constant sign of the square roots and you can discover this and much more complicated things using genetic programming. These days perhaps the most interesting and in vicious and scary application of genetic algorithms is not to do this in the computer but to do it in the real physical world so there is an area of robotics where you literally have these robots and this is from harvard and cornell performing whatever function you want them to perform, crawling around doing things and then they get to the three d. Printer to create the next image. Hopefully we will keep these robots under control which brings up a whole host of other issues. Now the visions are known as the most fanatical of all. They are fanatical for good reasons. The paradigm comes from statistics that they were persecuted minority and its a good thing they did because they have a lot to contribute and these days they are very much on the right even in the statistics within the Computer Science he won the award a few years ago that had the other famous visions is actually at Microsoft Research and mike jordan at berkeley so what do they do . They take their name from the base of europe. They see them so much that theres this Machine Learning startup in this theorem put outside their offices. This is an actual picture. So what is this doing . But we give you the basic intuition of whats happening. The problem each of the tribes has a problem in Machine Learning that its focused on and its trying to solve so for the connectionist its to create a final problem of learning the composed knowledge. Life is uncertain you never know anything for sure. So choosing one hypothesis they can see their own possible hypothesis from space at the same time and then what they do is they look at the evidence and that changes. The theorem is just a way of doing that updating how likely each hypothesis is given the data that you see so you start out with the prior. Its how much you be leaving each had offices before you see any data. Then theres the likelihood that shares if it is true how likely am i to be seeing this and then the idea the more likely they hypothesis about what you do is multiply by the likelihood. Its how much you believe in your office is after youve seen the data. Then you see more and more hopefully some would become very unlikely into the qb can very likely hopefully theres one hypothesis youre pretty sure. In the meantime you might just consider a whole bunch of hypothesis at the same time. Doing this can turn out to be very difficult because if you consider they can become expensive. Theres a lot of interesting applications that exist. One that you are very likely familiar with the first and probably the still best way to build filters is to use addition learning that they tried to decide whether it is them or not so the hypothesis that pryor is before you see the inbox how much is to be spent, take your pick and defend the evidence if it contains the word viagra it is likely to be spent in if it contains marks it is more likely to be spent. It contains the words euro it probably cant be spent and you might want to look at it just to be sure. So its one application of learning but there are others and another one is the algorithm to choose what ads to show you is a huge network. A Vision Network is one model with millions of nodes to predict whether youre going to click on the add or not. And the more money that google makes the happy or you should be and the others should be. So finally lets talk about the analogy. They are more of a loose grouping of people who all do during based on this idea of reason by analogy which is something that they have evidence people do a lot. Its iv to solve a problem im going to look for a similar problem in my memory and then go from this problem to the problem i look for the most similar symptoms in my files. A very simple but surprisingly effective. Peter hart was one of the pioneers that invented the machines which is probably the leading analogybased algorithm and then this is the most important algorithm in Machine Learning but also the most famous and how much nicer at all. Hes a scientist and into the one who actually claim the term and knowledge eyes and now a size and everything we do not just learning that all intelligent thinking is just an allergy in a different way so he very much believes that reasoning is the master algorithm. So how does this work . When you propose a simple puzzle to illustrate how this works we will look at the simplest Machine Learning are probably is and it was invented years ago but its still pretty good. If we give you a map of two countries ive shown with a and my puzzle to use just from the location can you tell where the border of the two countries is . Thats a hard problem because they determined the border but probably good is the following. Any point on the map is closer is probably part. So the border between the two consists of all of the points that are the same difference from the positive city into the negative city so this is just reasoning by similarity of the points that are more similar end of the and the ones that are more similar to the negative. This could be breastcancer images. If the image is more similar to one that has a tumor than otherwise i say that it doesnt have a tumor and so on. You can use it for recognizing faces and all sorts of different things. Now you look at this and say the border is a little bit and jagged it looks a little artificial. Too many Straight Lines and sharp corners. Another problem is that if you think about it you could actually throw away a lot of these examples. If you throw away this guy explained by the other two parties and now this doesnt matter but if you are learning from this than having to match them is a really heavy expense. They only keep the support vectors that are the examples required to keep it in its place and then it also does the following exercises. You want to walk with the positive on your left and negative on your right but always think as far away as possible from all of those. Think of them as landmines. You want to walk across no mans land while stepping very far. This is what machines do. Colonel is a technical name for how you measure the similarities. The methods come at the most important thing is how you measure a similarity between the two. Let me give you one example of how we advertise our methods and network in the world and heres an example you will be familiar with. They use all sorts of learning algorithms to recommend projects and its still one of the most possible is this type of algaebased. You find one to figure out whether you will watch a certain movie and one of the best things i can do is book for people that have similar tastes to yours. I will get the movie they give a lot of stars two and these are the movies you give a lot of stars two and also that they didnt like are the movies you didnt like them then you seem to have a similar taste to them and if they liked this movie you will probably like it as well. This turns out to work well. In fact ive seen it in several places. Three quarters of the movies and netflix come from the recommended a system. This is the basic idea. So these are the five tribes. Its quite interesting. This is the problem of being able to reason by seeing similarities and theyve been successful at what they do, so heres the big picture. We have each one that has a problem it has worked on for a long time and its hopes very well s solves very well. In each of the tribes there are people that believe they have the master algorithm. Some connectionist, not all say that is all we need. The 38th edition in speeches in the next with language and reasoning and then we will be done. I however dont think that is the case. I think what we need is a single out of recumbents holds that all of these types of problems because all of them are real and if you have an algorithm that solves the problem it doesnt solve the other problem. Where does that come from in your artificial brain. They know how to answer that question so what we need and what a number of us have been working on in the last decade or so is the Machine Learning. The single master algorithm in a similar way for example how the standard model or the central dogma we are looking for this idea of the learning algorithm that we would be able to solve all at one so what is the algorithm will quite . We havent found it yet but we have made a lot of progress. What is one way you might go about creating such an algorithm . They all look very different that they actually all have the same structure. All of them and many others that ive shown here are composed of three things. The first is representation. Its the choice of programming language in which it will express what its learned. Those are too complicated for this and they make things difficult and it might be things like the Vision Networks but one way or another you have to choose a representation and the natural first step is to invent a representation that has a power of logic and the Vision Networks more generally the graphic and that this is something that weve already done successfully that we developed pushes a combination of logic and the model network. In essence what it does is to have the formulas in logic that you can think of in a way the computer understands. If the formula really be beads in it and meal is false then how likely is a function of how many of the formulas. So lets stay we start with the representation then the next part of every learning algorithm has what is called the algorithm function thats a way of answering the following question if you give me the candidate algorithm, you give me an algorithm for diagnosing Breast Cancer im not going to just start using it i have to figure out how good it is. And for example how good they are and how often do they give the right answer but also how small are they and they can be included in the probability we talked about. The likelihood is how well the data but more generally i dont think they should be a part. It should be whatever you care about. So if you are a company that wants to maximize your profits on google and you want to do the click through ads thats what it should be. If its even in the want to be happy and your goal is to be happy using you name it, thats what it should try to learn it out. You should be able to tell that is what you want and then it goes about fulfilling that and finally there is optimization by which the algorithms find the program that has the highest possible score. And now here heres a natural combination of genetic programming. I need to discover the formula. What are the logical formulas clicks i can discover them because it is a kind of programming language and then once i discovered the formula that way this is the current chart that the master algorithm is still far from answer because there are so many things we would like it to do and it doesnt do and my suspicion is that the real master algorithm requires ideas that we havent had yet so there are these ideas i talked about in which the rest position because they wait for need for their own trial, their own paradigm and the hard time seeing outside of them support supported the reason i wrote this post is to let other people find out about the Machine Learning and start thinking about it and maybe have their own ideas so if you crack the master algorithm between now so i can publish it. Let me conclude by talking about with the master algorithm will make possible. Theres a lot of important problems in the world and things we have been working on for a long time and i think it isnt clear that each of these things we cannot sold without Machine Learning but also the compounds are not good enough for and these are problems that involve each one of those issues weve been talking about. The example is that some would like to have robots and why dont they exist yet. Combining the knowledge on the fly and reasoning by analogy and whatnot the robot in your home has to face instances of each one of these. The home is a structured environment. It needs to be able to think on its feet but i think we will be up to have robots that can do all these things. They turn it into a Knowledge Base of the computers can then reason with. If a knowledge map all the others would happen if happen if there separates like that going on. In order to do this you have to but for example you need the kind of symbolic representation but also its very messy and you need a probability that we talked about so there is another thing that will not happen. The companies are pulling resources into this on the progress of the Machine Learning. Heres perhaps the most important one of all. The reason cancer is a hot problem is that it isnt a single disease. Everybodys cancer is different and the same cancer changes and mutates as it grows. Its a learning program that takes in the genome of the tumor , the medical history. One of the combination of drugs will be designed specifically for that problem. Today isnt capable of doing it because it involves model what they are doing and understand the structure and the parameters you need to reason by analogy and so on and so forth but once we have that out of them we will be able to used such a program and its a large effort led by people trying to collect all this patient data in order to then learn and if the patients do share that data than we wouldve some point people to cure most cancers and if they dont than we probably then we probably wont so we need both of the learning algorithms in order to do this. Let me mention one last thing the one that is most likely to affect you in your everyday life which is going back to recommending. Amazon recommends and vendors of those silly a bazillion of these things. Your smart phone is trying to learn about you, who you are, what your tastes are and what you do whether it is just picking a movie or a house to buy this model will be able to propose things for you and hope help you go from a million options. In order to learn that kind of model we need more powerful learning out of those. To figure out what we want to do, your model you press to find me the find me a job in. So its looking at the model of the company. Then they go with a million different people and comes back and says all of this goes on and the world works better for you so theres a lot of impact that its but its already had but using both the data that is being produced we will actually get to a lot of these things. Thank you. [applause] there are microphones on either side of the stage so please go to the microphone if you have a question. Please keep your questions in the form of a question and we will get as many as we can. What are your thoughts on the composition and the master algorithm . Meaning using firstorder logic they believe that the composition is essential. You need to be able to decompose your problem is and then combine them together. We just have a big network that solves the problem and that is one example. The other ones dont so much. Language is very compositional. I can combine the words. You can do that with symbolic learning but not connection learning. Theyve warned about progress yet other candidates are waiting for it. Wondered if you have an opinion on what we need to do in terms of if it could be regulation or you fall into one of those camps. This is a prominent question over the media and there is i think that neither side is right so theyve been worrying that they are going to take over and i dont know too many experts that take. The reason i think people have those thoughts they confuse and think if we make a supreme intelligence that will have the same desires that we do but the truth is that they are different from human intelligence. They could actually be infinitely intelligent and we have no danger provided that they only get to solve the problems we get them. I dont think that is the biggest problem that on the other hand there are some bigger problems and one of the reasons why there is a rough understanding of what Machine Learning is it has the potential for good and for bad and the people who master it will be the ones the technology service. You want to take control and make sure your part of the conversation. Now another example is jobs. There are a lot of jobs because of Machine Learning. How does my job relates to machinery and is it something they can do easily maybe i need to get on top of this and start figuring out that machines dont have common sense. They are requiring integrating a lot of information from a lot of different sources and understanding the context will. To leave behind the ones maybe in the long run i dont think its going to happen if we get the point that computers and robots are doing everything better than people than they will get more jobs there will be a good thing because people just wont have to work. Then maybe everybody will get a check. On the other hand human nature being what it is some people want to make more money than the other and some will be necessary. It will be a bit of a turbulent ride but in the end everyone will be much better off. Finally the single biggest danger of Machine Learning is the sorcerers apprentice problem. You tell the computer what you want it to do and the computer does it but the computer is dumb so it does the wrong thing, it says what you literally said instead of what you really wanted to do or something in the world that is the wrong thing because it doesnt understand the world very well or it just makes mistakes. A lot of credit scoring, the computer might decide wont give you a credit card or might crash at times on the server. The queue for that is to make computers more intelligent. People worry computers will get too smart and take over the world but the root problem is they are stupid and they are already taking over the world. Thanks for your talk. I am curious, i have a two part question. What do you think are the toughest problems yet to be solved either in the master algorithm or Machine Learning and secondly how long do you anticipate the master algorithm, all long will it take. A lot of progress has been made but just your thoughts. Theres a barrier we already see which is the combining disciplines are difficult. Combining logic and probability is something people have worked on for number hundred 200 years and think it is impossible. Theres a problem in which we have made a lot of progress and we are on the way to solving it but it took a long time and effort to get there. Now people are doing it by combining these things we to edit time or maybe three, things get harder as you go along so it remains to be seen whether we will be able to do that, or maybe someone will have a completely different idea to solve the problems in a way people havent thought of yet. That is one thing. Then there is the question of once we have this on a rhythm it can learn commonsense knowledge, how much do you have to live like human life to acquire common sense. Theres a school of thought that says if you read the web and understand and read a lot of novels and textbooks there will be all the knowledge you need to have. Things like the Google Knowledge graph, theyre trying to solve the problem at the end of the day but as much as i think you can get a lot of mileage that way in the end if you cannot only acquire common sense by living life, the way you are going to do that is moving around the physical world, breaking legs and getting up again and interacting with people. Will bring up problems we dont even see yet but it is already starting to happen. There are self writing cards, sold driving carelf driving cary get better. When will we discover the master albritton . Progress in science is not linear. You have computers that make no progress and then suddenly a breakthrough. This happened in vision and speech, 10 or 16 years nothing was Getting Better and then came deep learning and you have vision and speech systems that are borderline practical. This depends on peoples creativity. It could happen tomorrow. Or it could take many decades. One of the reasons i wrote this book is to make it possible for other people to come up with ease ideas. If some kid somewhere read the book and has an idea and solve the problem, that is the outcome i could talk about. Thank you. The approach to solving problems changes so much, what do you think should change in the education system, kids and learning in schools, universities seems to me the program has to change. Very much so. One of the biggest impact of Machine Learning has been education. I sent this as a professor, education is terrible. People learn very slowly. It is very difficult and very painful. There has to be a better way. The better way on the one hand, people watching videos, learning on line. A lot of the critics of this Online Learning what they say is that Online Learning doesnt have the efforts of personal interaction you have in the classroom so the interactive component is very important but the truth is when you are in a classroom with other students and one professor is not that interactive and the other thing, this is where Machine Learning comes in, companies are registering to do this because they were started by Machine Learning people, every one of the interactions you have with a computer weather is an anti debt wrong or something you ask about or something you try is teaching the computer the difficulties that you have and the computer, meaning the Machine Learning algorithm to learn not just from you but from everybody else asking the same question so that the end of the day, you are going to have a computer tutor that teaches you much much better than any Computer Division issue and. People have been working on this for a long time but the paradigm used to be to program and that is too high. With Machine Learning we will see where difficulties arise. People will learn better and faster and what they need to do. In the sense of how we used to learn reading groups and things like that, year the approach is you learn how to learn. It is not that straight forward. Do you think this kind of approach will also come in . I only talked about the main type of Machine Learning, the input and output and learn to transform one to the other but theres also unsupervised learning which is learning without in the end in the long when it will be much more important than supervised learning. Children learn a lot from their parents but they learn a lot more by Walking Around the world and putting things in their mouth and what not and there are arguments to do this type of thing and at the end of the day those will be more important. I wanted your thoughts and opinions on the recently announced darpa project is it feasible, epic quote . It was recently announced in the last couple days, darpa wants to use Machine Learning to create a balance between basically the algorithm to essentially choose privacy preferences and yet allow our data, the rough idea. This sounds like a very good idea. More generally on the solution of privacy. Here is what happens today. Your privacy is being eroded. We all know about that because you dont really have the choice. In fear you have a choice that will be all those arguments and look for all the buttons and what not. There are things like privacy preserving algorithms, the necessary medical knowledge to be extracted without being able to go deeper. There are some problems in that but that is not poll solution. Another part of the solution is we have to get smart about who owns the data and how the data gets shared. You, one the date of to be owned by you in view look at what it can be used for, who you are willing to share with and so on and so forth in a way the is natural and easytouse and part of that can be done using Machine Learning like one thing about Machine Learning is computer systems, theres a new different thing you can do with it but finding the one thing you want to do is often very hard. What the Machine Learning does is automatically adapts, exactly the same approach for privacy. That is what this project is trying to do, i am all for it. What you said about 360 recommendations, the concern people have is where consent concern, is concerned coming from people who use the data or users and individuals . Think about it this way. Whenever theres a transaction between two people, person and computer, both of those dont the data. If i have a conversation with you i have the right to record in my brain, and remember it and you remember it in yours so to some degree there is no getting around it. This 360 degree model is from your whole experience, you are the only one who has your whole experience and you wanted to stay that way. Companies like google and face book would like to be your holy spirit, youre smart phone records of everything you do but that is a little dangerous because of the conflict of interest. Watch it happen is Something Like a data bank or union, something where a bunch of likeminded people get together and say this is our data, we are capturing our data, we own it, we decide what to do with it. Than you might join the union of people who want to do what you want to do. In the end of the day, you want that 360 degree model. Life will seem unbearable if you dont have it but you also want that model to be owned by you. Google wants to be the third half of your brain. That is what they want to do but the thing is the first two has of your brain are in your skull and the third half is on a server in oregon. They have a conflict of interest. You dont want the third half of your brand have a conflict of interest. I ask my girlfriend about the case for organs, basically charge computer is. And recognize the program. How do you keep that up . Exactly. This is a perfect example. You want to have what Computer Scientists call a proxy server where your interactions go through this computer and orbits doesnt know if his you. It is an intermediary. All that they know is it is coming from this x but this x is you and they play those games. I think that is all the time we have. Pedro will be over here. Thank you. Quote and braque and applause] [inaudible conversations] [inaudible conversations] you are watching booktv, television for serious readers. Watch any program you see here online at booktv. Org. The name of the book, american christianity, the continuing revolution. The author, professor stephen cox when we sing about the oldtime religion what was thinking about each nothing. Christian andsomething