Create a future with ai that is positive for humans and machines. We are fortunate these conversations are happening more and two of the president fought leaders driving these conversations our special guests tonight. We will hear from the director of the mit initiative on the Digital Economy, Erik Brynjolfsson, and author of two bestselling books. The second machine age and machine platform crowd harnessing the future. Next we will hear from max tegmark of the Life Institute and author of the new book just coming out, life 3. 0 being human in the age of Artificial Intelligence. And i understand it is in its second week on the New York Times bestseller list. The two gentlemen have a conversation together and time for questions with the audience before we conclude with a reception and a book signing. With further a do, join me in giving a warm welcome to Erik Brynjolfsson and max tegmark. [applause] i am so delighted to be here, a pleasure to have a chance to share the research we have been doing in the work we have been doing on the Digital Economy about Artificial Intelligence, how it is changing society. I had my thanks for supporting this and inviting me and especially thanks to max tegmark. We have known each other three years but he is one of my best friends, on the cover of his book, and absolutely joyful mind and when he said lets do this i jumped at the opportunity mainly because it is fun to talk and you will find out in a little while we will have a fun conversation where we interact and looking forward to hearing your questions and comments and everybody will participate. We live in very unusual times right now. As you may have read, cars are beginning to drive themselves, people are walking down the street talking to their phones, not talking to another person, they expect the phone to understand what they are saying and talk back to them. Still a little bumpy, not that good at it but beginning to talk to us and about a 10 year period we are going from machine mostly not understanding us to machines understanding us pretty well. That is a unique time in Human History and we are lucky to participate, to be part of it. It opens amazing possibilities, these could be the best decade in the history of humanity or could be one of the worst because the power being unleashed by Artificial Intelligence is unlike anything we have seen before. Let me set the stage by defining what we are talking about. Artificial intelligence is a set of techniques that imitate the human mind, the turing test, you may know proposed by the famous Computer Scientist alan turing, could you speak to a machine and not know whether you were interacting with another human aura machine . Machines arent passing the test yet but are getting closer and closer especially in certain narrow areas. With that is a category called Machine Learning and this is driving a lot of things, good oldfashioned ai was an area where we would teach machines what to do, we write down symbols and say this is how you play checkers, this is how you play chess, these are the rules and how you prepare taxes and machines follow those instructions. Now the Machine Learning revolution is taking over. Instead of us humans having to tell machine stepbystep what to do which frankly didnt work that well. It was okay but ran into a lot of barriers. Now machines are learning themselves how to solve problems, figuring it out and the way they do that mostly is different techniques but the main approach is we give them lots of examples, this is a dog, this is a cat, this is a dog, this is a cat and i wondered how many times, they do it 10,000 or 1 million times and the machine goes i see the pattern here. It will start learning and the nice thing is we have so much digital data we show lots of examples of fraud or successful go moves, and eventually start learning these patterns statistically. And a subcategory of that but the biggest part of the breakthrough especially in the past about 5 or eight years, deep learning or deep Neural Networks, loosely based on the human brain, let me show you some examples from a subcategory called reinforcement learning, and new strategies on their own so when example is the group of people at a Company Called the mine, what they did was on the cover of nature, they gave the Machine Learning album the pixels for atari games, how many of you have played Space Invaders . How about breakout . I will show you the game of breakout, they just gave the machine they didnt say this is a panel, this is a block, this is a ball, the machine had to figure that out, learned on its own. They gave it the raw pixels, controller that could move left to right, and a score that said heres the score, look at the score, your job is to move the paddle, move around the controller, they didnt tell it was a battle, to adjust the controller and make the score as high as possible so at first machine wasnt all that good. It would sometimes get lucky and hit the ball. Other times it would completely miss it. It was randomly moving but when it was successful the score went higher, got to do more than that, reinforcement learning and after 300 games it was pretty good, as good as a good teenager, planning successfully, let it running for a wildland they dont play breakout but a strategy here figured out how to send the ball behind, we didnt know you could do that. The machine not only learn how to play but how to play better, a newborn baby being born and in the hospital, you hand it a game and by the end of the day beating the surgeons, the doctors, that is how fast it is learning and that is a cool little example, a scientific progress and the same technique, worked on Space Invaders and other games, didnt work on all of them but had a feedback loop where if you did something the score would change quickly, able to learn on its own pretty quickly with no control programming. You can use the same techniques for other things, not just games but a data center where google has their Computers Running as a big videogame, all this data coming in and temperatures, the score is try to make it is deficient as possible, lower the cooling bill is much as possible in your controller, just little valves, left and right. They had a bunch of smart phds working on optimizing this so they thought they had it running as efficiently as possible but once they got the Machine Learning algorithm against it, it got dramatically better so you see before the Machine Learning algorithm, turning machine on, it is 30 more efficient. It was back before. The machine figuring out how to run their david Center Better than all these geniuses at google were running it and doesnt take much imagination for that one data center, now we do it for all the data centers, all kinds of factors, steel finishing lines, and makers of any kind of object. There is room to apply these things and improve all sorts of categories. Three big breakthroughs where Machine Learning has made a big difference that werent important as recently as 10 or 15 years ago, in our book to some extent, vision and language, the physical world and problem solving so for instance you may want to get some snacks afterwards, be careful what you are reaching for. There are muffins here, not all of them are muffins. Sometimes we can make mistakes. At stanford, a very large database with 14 million images and each was painstakingly labeled by humans as to what they are and back in 2010 when they did that, the machines, they were wrong about 30 of the time. Today they are wrong about 2. 6 of the time. The steep curve was when they started using the neural net Machine Learning algorithms. As a Reference Point humans are 5 . They havent improved a lot in that time period. We saw the same hardware and software. It is important Inflection Point because what that means is many tasks that used to be better to have humans do them, better to have a machine do them or more accurate to have a machine do them and that shows up in a number of areas, you can use that to diagnose the same algorithm and show examples of patients the dont have cancer and patients the do have cancer and machines figure out as well or better than human pathologists, paper looking at skin cancer and it did better so this is the past few months or you few years. I mentioned voice recognition. 8. 5 error rate, that is in the past year, not like over the past we 10 years, that is the past year since july 15, 2016. Humans are 5 error rate too so it is in that ballpark. Not quite better than humans and that is opening a lot of economic possibilities, tracking the physical world, when you see and recognize things, pedestrian or bicyclist starts becoming feasible to give control of the car to a machine when they start doing these they made errors one for 30 frames, once per second, now it is once per 30 million frames. You can go years without making a mistake, better than human so very soon we will see more of these. I have written in a bunch of them and feel comfortable driving down the road, being driven down the road making a left turn in traffic waiting is ultimately much safer, 30,000 deaths by humans dropping by 90 or 99 when machines, probably not 100 so we have to face some ethical issues when machine drivers still make mistakes but it will be dramatic and safe and beginning to work in factories, at the Computer Science at ai lab at mit has a company in boston called reef robotics, baxter works for 4 an hour doing simple tasks. You dont have to do Computer Programming, you show baxter what you want to do and put it in a box and after a couple examples, i like what you do and it does that path and 7 days a week, 24 hours a day. I was just on thursday watching another robot a little bit like this, sorting things, soft object like clothing, faster than humans did and that will replace a great deal of work in those areas. Last but not least, medical diagnoses, the legal area talking to guys at jpmorgan and they see a lot of relatively routine legal work, 30,000, 360,000 hours of legal work, what does this mean for the economy . Let me touch on that. First off, there is good news but also big challenges. Makes the pie bigger but there is no economic guarantee anyone will benefit. It is possible for some people, to be made worse off than they were before. That is part of what is happening the past decade and it could get worse if we are not careful, productivity has continued to grow and gdp is at a new alltime high but the median Family Income is lower now than it was in 1990. They just had reports in 2016, there was an uptick over 3 last year, depending how you do the adjustment it may have matched the previous high although if you normalize it it is still lower than the previous high in 1997. We are roughly flat in that period. How could median be so much lower than per person . Median, the museum of science, the 50th , not the average but a person in the middle, half of people are higher and half are lower so the median can stay flat if you have a bunch of wealth going to the top 1 or top 1 10 of 1 and that is what is happening as computers kicked in, there has been biased technical change and you heard about the 1 , the one present have their own 1 up here, the share of income, a new record high, only time it was close was before the Great Depression if there is any consolation. We are having an economic challenge of the pie getting bigger, when happening, the distribution is becoming more and more skewed and many reasons for that, part of it has to do with tax policy, international trade, most economists including me see the Way Technology is being used as the number one driver of that. That is not inevitable. Ultimately we have an opportunity to rethink how to organize the economy. Of the pies getting bigger creating more wealth, for everyone to get richer at the same time we can make the rich richer, the middle class richer, the poor richer, we could all be better off at the same time, that mathematically adds up but there are some choices we have to make as a society what we want to do in terms of taking advantage of this. Business as usual wont falls the problem. We are doing a number of things to address it, trying to understand the drivers of this, do some research, we organized the inclusive innovation challenge, and to another event on october 12th at Boston City Hall plasma, the governor, eric schmidt and a bunch of other people will be coming to talk about how to use technology to create shared prosperity for the many and not just the few. With that, let me leave you with the closing thoughts and these technologies are certainly wondrous give us all sorts of opportunities, they can be used for good or to create vast wealth but dont automatically lead to distribution makes everyone better off so it is important, glad youre here, for all of us to about what we can do to change the society we have towards a better one end how to use technologies for broadly shared prosperity. Let me turn it over. Thank you for inviting me here. And thank you for your friendship and kind introduction, setting me up. So wonderfully here. Lets see if this will cooperate. I will continue further forward in time and talk about what will happen if ai gets even smarter. Will it be like the human in the age of ai and what it should be like. Lets go far back and look at the big picture. 13. 8 billion years ago i have to start here. The universe was very boring, almost uniform plasma everywhere and nobody there to witness it or enjoy it. Gradually the laws of physics clumped this into galaxies, stars, 4 billion years ago the first life appeared on earth, that life was pretty good. I call it life 1. 0 because it couldnt learn anything in its lifetime like bacteria here. Life 2. 0, we can learn and if we bring a metaphor of thinking of the computer of sorts, learning corresponds to uploading new software into our minds. If i want to learn spanish i can spend a bunch of time uploading the spanish module and have new skills. Bacteria cant do that and precisely this ability to learn, change our own software enabling cultural evolution, made us the most powerful species on this planet. 3. 0 would be if you can also design your own hardware. We humans are trying to head in that direction, we are at life 2. 1 right now, artificial needs, artificial pacemakers, but true life doesnt exist, Artificial Intelligence might help us get there. We heard from eric that ai is getting smaller and we heard from eric, traditionally Artificial Intelligence used to work, the world chess champion 20 years ago, people taking their own intelligence and encoding it into a Computer Program with them, think faster and remember more where the recent progress as we heard from eric has been driven by Machine Learning where you have simple machines with simulated Neural Networks inspired by human brains and train them with massive amounts of data and pixels and outcomes, young people playing a game of frisbee, the same thing in this image, it heard of elephants walking across the glass seal. Even more striking if you look at playing Computer Games as we saw in this example eric showed. Once the computer can learn to play party games, that tells you there is a lot of room for growth. Thinking of life itself as a game, when you get rewarded for certain things and try to learn, the same company that did this just came out with the following result where they trained toy robots to see if they could learn to walk, points when they were able to move forward but the software, had never seen a video, nothing about the concept of walking, did this in simulation because it was cheaper than doing it with metal objects the kept falling down but the idea is the same. Nobody taught them to do this. It learned by itself. They tried a variety of different body types including animals who learned to run or jump and so on. Anything where you have a goal can be thought of as a game, whether it is the stock market or playing a sport. Where is this taking us . I like to think of all intellectual tasks as forming the landscape like this where the height is how hard it is for machines to do. The ocean level being how good machines are at doing it at present. Our chess playing skills have long since been submerged by machines to multiply numbers even longer ago with calculators and that kind of thing. The Worst Career Advice to give our kids is it encourages them to do jobs on the boundary about to get submerged. Sea level is rising in this metaphor of a landscape, machines keep Getting Better and fascinating to wonder what will happen. Something machines will never do the things we do but a lot of leading Computer Scientists machines will eventually do everything we can do in 30 years or so and then what . Very interesting choices to make here and i feel these are choices we shouldnt make after thinking about them. In that spirit i found a bunch of colleagues, do you want to stand up . Two of the cofounders here, and your son is also someone we are happy to have with us here, the idea of the Organization Helps create the best possible future with technology seeking in advance and what we do to get things right. Im optimistic that we can create a wonderful future with technology as long as we win this race with the growing power of technology and the wisdom we manage the technology but in order to win this race we need to change strategies, and we sta