Destiny. Try ares to assess everything that could go wrong is not being alarmist from the contrary deep thinking, analysis, and advance planning will allow us to think about what kind of future do we want, and ultimately enable us to create a future a. I. That is positive for u humans and machines. Were fortunate that these conversations are happening more now and that two of the press thought leaders driving these kftions are our special guests tonight. First, we will hear from eric, director of the m. I. T. Initiative on dimmingal economy and author with Andrew Mcafee of two best selling books second machine age and machine platform crowd harnessing our dingal pooch. Next well hear from max cofounder of the future of Life Institute, and author of the of the new book just coming out life 3. 0 being human in the age of Artificial Intelligence, and i understand its in its second week with on the New York Times best seller list. Congratulations. Then the two gentleman will have a conversation together and finally there will be time for questions with the audience before we conclude with a reception and a book signing. So without further adieu please join me in giving a warm welcome to eric nelson and max. [applause] thank you. Thank you. Thank you im so delighted to be here. A pleasure to be able to have a chance to Share Research weve been doing and work on the Digital Economy about Artificial Intelligence, and how it has changing society. I want to add my thank you, jack and suzie rebbe know for supporting this and thank you for inviting me and max we know each other about three years but hes become one of my best friends. He is as i say in the back cover of his book, has an absolutely joyful mind when he said lets do this, i jumped at the opportunity. Mainly because it is just fun to talk to him youll found out in a little while where we interact and im also looking forward to hearing all of your questions and comments because well open up and everybody will participate many it. You know we live in very unusual times right now. As you may have read and seen even, cars are beginning to drive themselves. People are walking down the street and theyre talking to their phones and theyre not talking to another person but expecting fen to understand what theyre saying and talk bag to them. You know it is bumpy theyre really are not that good at it but beginning to talk to us and were many the i would say a ten year period where were going from machine mostly not understanding us to machines understanding us pretty are well and thats kind of a unique time in Human History and were all very lucky to be able to parent many that to be part of it. It opens some amazing possibilities these could be the next, next ten could be best decade in the history of humanity or it could be one of the worst because the power being unleashed by Artificial Intelligence is unlike anything weve seen before. So let me first set this stage a little bit by defining a little bit what were talking about, so you can think of Artificial Intelligence is this set of techniques to intimidate the human mind the classic the test, many of you know some would propose bit Computer Science alan could you speak a machine and not know whether or not you were interacting with a human or machine. Machines arent really passing that test yet but theyre getting closer and closer especially in narrow areas within that is a category called machine learn egg and this is what is driving a lot of excitement recently is that good Old Fashioned a. I. Was an area where we would teach machines what to do and write down symbols an say this is how you play checkers this is how you play chess these are the rules and this is how you prepare taxes and machines would follow those instructions so instead of us humans telling machines step by step what to do frankly Department Work that well it was okay but it ran into a lot of barriers now machines are learning themselves how to solve problems. Theyre figuring it out and way they do that mostly theres different techniques but main approach is that we give them lots of examples say this is a dog, this is a cat. This is a dog, this is a cat. And i wont do it many times as they do to the machine 10,000 or o a million times and eventually machine says i think i see pattern here and it will start learning and the nice thing is we have so much digital data now that we can show them lots and lots of examples of fraud or successful go moves or of faces that eventually they started learning patterns. Statistically, and a particular sub category of that where really biggest part of the break through especially in the past five to eight years is in deep learning or deep neuronetwork loosely based on human brain and deep learning sub category called reenforcement learning, these machines can learn new strategies on their own so one example is group of people at a Company Called deep minds google bought and what they did was on the covers of nature the science magazine, about they gave this Machine Learning algorithm pixels how many have played Space Invaders how about breakout you know the game of breakout ill show you the game of breakout they just gave the machine the raw pixel they didnt say this is a paddle, block, a ball the machine had to figure that out. Learn it on its own. They gave it the raw pixels. They tbaif it a controller to move it left to right and they gave it a score and said heres the score look at the score. Your job is to move around panel to get or move around the controller and try to make a score as high as possible. And so at first the machine wasnt all that good. It would sometimes get lucky and hit the ball other times it would completely miss it and it was basically randomly moving arranged but whef it was successful and hit got the score went higher, machine was like i have to do more of that. Reenforcement learning feedback on what tods more and more and after 300 game it was really pretty good never missing good teenager, and playing along there pretty successfully. They decide to let it run for a little while and guys at google deep dont play breakout and didnt know there was a strategy here. Look it seppedz the ball around behind and theyre like oh, we didnt know you could do that. So [laughter] machine had not only learned how to play but learn how to play better than designers imagine, a newborn baby being born and in the hospital and you handed a game and by the end of the day it is feeding surgeons and doctors at the ghaims kind of how fast it is learn now thats sort of a cool little example. You know, a scientific progress. And by the way, this same technique works for a whole set of target it did work on Space Invaders eventually on pacman and other games not on all of them but a quick peedback loop if you did something the score would change quickly. It was able to learn those on its own quickly with no control programming. Now you can use these same techniques for other things, though, not just games you could think of a data center where google has all of their computers. Running as a big video game, these all of this data coming in and temperatures and the score is try to make it a deficient as possible lets lower cooling bill as much as possible, and your controller is you can adjust a little valve to move them left and right now they have a bunch of smart ph. D. Es working on optimizing this so they thought they pretty much had it running as efficiently as possible but once they put Machine Learning all l gore rhythm against it got dramatically better. This is better of Machine Learning ftion on and turn Machine Learning own and 40 more efficient, and it was pretty and then they turn off again and the way it was beef. So machine figured out ho run Data Center Better than all of these geniuses at google were running it and it doesnt take imagination to say do it for data centers and all kinds of factories for steel finishing lines and makers of any kind of an october so theres room to apply these things to improve all sorts of categories. Many of the theres three big break throughs where Machine Learning has made a big difference that werent important as recently as about 10 or 15 years ago and we write about them in our book to some extent vision lag wage interacting with a physical world and problem solving so for instance, you may want to get some snacks gaff wards be care what youre reefing for. Theres sol muffins here not all are muffins, though. [laughter] sometimes we can mistake when is were seeing things and at stanford they have a large database called image net with 14 million images and had many painstakingly labeled by humans as to what they are, and back in 2010 when they tried to have machines see what they were machines were not very good they were wrong about 30 of the time. Today, theyre wrong about 2. 6 of the time. So got dramatically better. The steep curve which when they startedded using deep Machine Learnings algorithms as a Reference Point humans are about 5 they havent improved a whole lot. [laughter] so we still have pretty much the same hardware and software, and so machine have crossed that threshold now many tasks better with humans do them it is better to have a machine do them at least more accurate to have a machine do them. And that shows up in a number of areas for instance you can help them diagnosis diseases with that same and show examples of patients that dont have cancer and patients that do have kerns and machine carts figuring out as well with or better than a human pathologist it was a paper just published by sebastian throwing Company Looking at skin cancer and it did better than human so this is just happening this past few months. Past few years, i mentioned voice recognition, you can see progress there 8. 5 to in the past year thats not like over the past ten years thats just the past year. Since july 16th, july of 2016. Still humans are about 5 error rate too so sort of in that ballpark right now not quite better than humans. And that is opening up a lot of economic possibilities. Interacting with the physical world so once you can see and recognize things, very are us to recognize a peds or bicycle iit starts becoming feasible to give control of the car to a machine. When they first started doing these they made error one per 30 frame one per second not what you want to have in a car. Now, its once per 30 million frames so thats years you can go without making a mistake again better than human and so very is soon well see more and more of these on the road i have privilege of ridsen many a bunch of them now and more comfortable driving down the road being driven down the road making a left turn through traffic waiting and ultimately i think it will be much safer there are 30,000 deaths by humans. Drivers today, we can drop that by 90 or o 99 when machines probably not 100 so well have to face some ethical issues when machines drivers still make mistake, but it will be dramatically safer than what we have today. And theyre beginning to work in factories brod brooks when used to be Computer Science and run the lab has a company in boston called rethink robotics baxter works for about 4 an who are doing simple task you dont have to do any Computer Programming. You show baxter what you want it to do pick this up, put it in the box after a couple of example it is says i get what you want to do and it does that task and, of course, baxter can work seven days week. 24 hours a day, i was just on thursday watching another robot a little bit like this sort sorg thing hads soft objects like clothing with faster than humans did, and that will replace a great deal of work in those areas last but not least all sorts of problem solving medical diagnosis i already showed in legal area which tack to guys at jpmorgan down in new york they see a lot of relatively routine legal work with a set of 360,000 hours worth of legal work. So what does it mean nor the economy let me briefly touch on what before i toughs it over to max. First off theres good news go theres also some big challenges. It makes the pie bigger. But theres no law theres no economic guarantee that everyone is going to benefit. It is possible for some people to be that none of them share or to be made worse off than they were before, and saturdayly thats part of whats been happening the past decade and i think it could get worse if were not careful. Productivity has continued toughs grow and gdp is higher and Family Income is lower. Good news is they have reports up for 2016 in the last year, there was a up tick over 3 in last year and depending on how you do adjustment it may have matched previous high although if you normalize it it is still lower than the previous high back in 199 so were roughly flat during that period. Now, meeting how can meeting be so much lower than that per person and thats because median you guys museum of science here is 50 per percentile but person right in the middle half of the people are higher half of the people are lower, so medium can stay plat if you have a bunch of wealth going to the top 1 where the top one tenth of one percent thats basically what happened as kiewrts kicked in theres been biased technical change and youve heard about 1 , the 1 had their own 1 . The pnt 01 up here, and the share of income going to them is at a new record high only time it was close was back right before the Great Depression it thats any consolation. So were having a pie getting bigger and one happening but distribution is become more and more skewed and there are many reasons for that. Part has to do with tax policy part of it International Trade but most economists including me see the way that technology is being used as number one driver of that. Now thats not inevitable. Ultimately, we have an opportunity to rethink how we organize our economy. Where pie is getting bigger creating more wealth that means it is theory for even to get richer at the same time we can make the rich ripper, middle class richer, poor richer and better off at the same time that math mathically adds up but choices we have to make as a society of what we want to do many terms of taking advantage of some of this bounty. Business as usual is not going to solve the problem. Were doing a number of things that m. I. T. To address it and trying to understand, the drivers of this, do some research on it. Weve also organized something called inclusive innovation challenge and i invite you to another event if you dont get tired out by this one october 12th Boston City Hall plaza the governor eric schmidt other people are coming to talk about how we can use technology to create shared prosperity for many and not just for the few. So with that, let me leave it leave you with a closing thought that these technology its are wondering but they give us all sorts of opportunities to be used for good. They can be used to create vast wealth but they dont lead to a distribution that makes everyone else better off. So it is very important im glad youre all here because it is important for all of us to think hard about what we can do to change the kinds of society we have towards a better one and what we want to do to use these technologies for broadly shared prosperity so thanks very much with that let me turn it over to max. [applause] thank you so much for inviting me here and thank you so much eric if for your friendship and for your all too kind introduction and also for setting me up. So wonderful here. See technology cooperates if i switch over to this so im going to continue a little bit further forward in time. And talk about what will happen if it gets even smarter. What it will be like being human in the age of a. I. And what it should be like. So lets first go far become and look at thing big picture so starting years ago you know, i have to start here since physicist universe was very boring with just almost uniform plasma everywhere and nobody there to even witness it or enjoy it. Gradually, the law of the physic clump this into galaxy stars, planet and so on and about 4 billion years ago first life appeared here on earth. Satellites was it is pretty dumb, though, i call it life. 0 couldnt learn like bacteria here. Life 2. 0 just what i call us, however, we can learn. And if we use eric, we metaphor of thinking as a computer of sorts then learning is uploading new software into our minds if i want to learn spanish i can spend a bunch of Time Starting and uploading it and have a new skill and [inaudible conversations] bacteria cant do that, and its frankly ability to learn to change their own software which enabled culture evolution which mades us the most powerful species on this planet. Life 3. Would be if you can also design your own hardware, we humans are kind of trying to head into that correction and life 2. 1 right now to talk implant or artificial needs or artificial pacemakers but true life 3. 0 certainly doesnt it exist Artificial Intelligence might help pus get there, though. We heard from eric that a. I. Is getting smarter and we heard from eric that traditionally Artificial Intelligence used to work like when a character star of the world chess champion over 20 years ago used to work by people. Taking their own intelligence and coding it into a Computer Program which then beats simply because it could think faster an remember more than he could. With the recent progress as we heard from eric has been driven instead by Machine Learning where you just have very simple machines often simulated neuronetwork inspired by human brains and you just train them with massive amount of data you see it in pixels and out come this is caption. A group of young people playing game