The stage we are welcoming back to new america, andrew mcathe and erik brynjolfsson. Theyre back testimony with a followup. That book is machine, platform, crowd, harnessing our Digital Futures which we have copies of for sale forking the conversation, shy mention. By further introduction, andy this Principal Research scientist and erik is the director othe initiative on Digital Economy at m. I. T. , and both of them are the top headed scholars on how technology is changing business and economics. Joining them on stage, alice son griswold, a reporter where she covers the economy, another sharing economy and strata conflicts. She comes from slate magazine where she wrote about business and economics for in the money box column. So we have copies of erik and andys book for sale. Following the conversation and they will stick around and autograph them for you. Without further adieu, Andrew Mcafee and erik brynjolfsson, and alice son. Thank you for coming. Tell us bat your book. Well dive right. In you start. The book, called my been, platform, crowd , available for sale, and people drive down the resale value of it by signing it for you arewards. This is about three great rebounds. Mind and machine, product and platform, core and crowd. We couldnt fit all the words on the cover so just used the last three and thats the direction that the world is changing right now. Not all the way, but partway from decisionmaking, moving more from human minds toward machines, both data driven decisionmaking and Artificial Intelligence, moving from product towards platforms like the five most valuable companies on the planet, apple, amazon, alpha get, google, microsoft, and facebook, and moving from the core to the crowd, the core being the people at the center of the company or organization that traditional hey have made most of be important decisions towards the crowd is people out of it and now can include billions of people connect evidence with a Digital Network in a way they never were before. We should tell you the book came from. Ways heard, erik and i wrote a book that came out in early 2014 called the second machine age all about the period of crazy tech progress were living through, and as soon as we publishedded we notice this really interesting phenomenon, we would good to conferences and talk about the book and then we would have really interesting hallway conversations over and over again, and the same hallway conversation. Somebody would come up to and say, i run company x, and i believe the story youre telling, now what . What do i need do differently . How do i need to think about my business model, the industry im competing in, this world were heading in. I know thursday are changing. Need a guidebook for this, and so the more we heard this the more we real realized this would be the next book. The central problem ways didnt know the answer to that question when we first start hearing it over and over but we turn that bug into other feature and this he can intellectually rich territory to write a guide book to the new world. And in addition to doing what agency does which is Read Research and digest papers and do all of that, but we do that is a little different and thats much more fun is we go out and talk to the alpha geeks, this is a term we learned from our friend, tim opretrial reilly, in oreilly, the future of the world and well talk to them and our homework was to try to distill down what they were saying and come up with some kind of framework or way to think about the changes that are happening so quickly, and thats where erik the three part structure that erik just described came from. We kind of heard that machines were getting ridiculously more powerful. Couldnt leave out the Artificial Intelligence and Machine Learning revolution. We kept on seeingwards companies appear and disrupted industry after industry and they all looked like platforms to us. We kept on seeing these examples of companies and individuals and things like look at dish wikipedia, very Diverse People though crowd was a powerful phenomenon. So that leads to threepart structure in the book. So, i think one of the an an neck e neck an net dote is about alpha go. Er who here is familiar with the alpha go story . How much a lot of people. Okay. So those who arent, do you want to explain what happened . Sure. Alpha go is an example of a Machine Learning algorithm that wag trained to play the game of go. The ancient game of go is that sort of a little board game but vastly more complicated than, say, chess, which was beat by machines back in 1997. There itsthere are more possible moves in a game of go than atom inside the universe. And the tradition principal programming techniques of trying to sample even a tiny fraction of the moves is completely infees able, and for that reason, as recently as 2015, computer scientists were surveyed when the thought machines would be able to beat humans at the game of go, and they predicted, probably around 2027. That was the median estimate. These were insiders, we polled this poll of Artificial Intelligence experts and asked the question. As some of you know, a machine did beat the World Champion in go just last year, and they did it by using the new technique, not trying to exhaustively search the move buzz by looking at the board and seeing patterns in it. Thats the way human does it. Humans when they mail the game of go, they dont even know how they do it. If you ask them, you you play go, they cant explain it. They just say, look at the board and i see a move that seems right but i cap explain how. I cant describe the strategy, cant tell you what im doing. Okay cant explain it, how do you program it, code it . But the new approach to Machine Learning where the machines look at lots and lots of examples of successes and failures, the machines can figure out the patterns on their own, and by applying a technique called deep natural net and reinforcement learning the minnesota learn the game of good to the level where they can beat the humans at the game of go, and now the human champions, when they see what the machines are doing, they say, you know, we have not even touched the edge of the game of go. We have been studying this for 3,000 years but the machines so far ahead of us we have no conception of what is possible. The reason erik and i think were still underestimating the impact and to power of Machine Learning, despite the fact we all read about it in press all the time is exactly what he gist said. Here is the domain of intense human study for three millenia. We build a body of what way thought was exhaustive technology, this machine shows us how much more headroom fliss this particular domain. Thats amazing. When we think about all of the other incredibly complex things that were trying to get good at, whether that is solving the mysteries of cancer, whether that is unlocking the secrets of protein folding, that is figuring out what to investy, whenever that is dealing with the silly complexity over the human genome we now have a colleague for that work that is not recommended be the knowledge we humans possess, might be able to show us some ways forward. This last points is really the key one and one reason we call it the second wave of the second machine age. The second member age is this idea that machines starts doing mental tasks the way that the first machine age that started doing physical tasksment but the way for the past few decades we taught machines to do fundamental tank is we programmed them. Codified our own knowledge and encode it so the machines can do how to do tax preparation or a spreadsheet. Stuff we any what to deand unhad to be very precise and the machine would do the same thing. Thats great. That generated trillions of dollars worth of value. Thats great. But theres so many problems that it sound paradoxic cal that we dont know how to do them. I wouldnt know how to code how to recognize my mothers face or how to ride a bicycle. These are thing wes just know how to do intuitively but can explain them. We know more than we can tell. With the Machine Learning we just described, we now have opened up those possibilities. These machines are learning on their own how to solve the problems, not by having a human teach. The seven by step but instead of giving them examples what works and doesnt work. This is a cat, a dog, a human face, this is the word yes or no in french or in english, and the machines give them enough samples will figure out the pattern on their own. Its really a remarkable transition that has really become important the past fewer years, since he wrote the last book, the second machine age. There was an example that went viral on twitter a while back. Could the machine tell whether its a blueberry muffin or a dog . Because it had the little dots on the face. One thing love about that another ones that came out at the same time, the labradoodle the kentucky fried chicken. Its surprisingly hard to tell difference. I was like, dog, dog, chicken, dog, and not getting it right. A story that works better if you look at the picture. Rite now you dont see that someone would accidentally bite into a piece of labradoodle. If you see the pictures, they can be hard to tell apart, but we humans can often do it and what is really remarkable is that seven years ago, machines had about a 30 error rate on a big database called image net. That was using the best technique. Now the error rote is lower than five percent, which is less than a error rate of humans on the same data set. So we have kind of crossed an infection point. Jot just didnt inflection point, when you raise water past the boiling point its a fundamental transformation. Every machine can do a task better than a human innings youre an entrepreneur or manager and have to make a decision, who do i assign this task you pick the one that does the job better. One thing we cook talk about ised the idea that in the future or even now, people will be learning from machines in addition to teaching them, and one thing liked about the alpha go story is that after the World Champion was defeated, he said, so beautiful, so beautiful, about the moves. It wasnt something heed ever conceptualized before and thats something we can sunny a lot of field its. Seeing it over and over and youre exactly right about people learning sidebysaid misminnesota. Lisa dahl, a guy who was bet my alpha go. A reason that where was a downward lip in the korean economy. It was big deal. The think two really important things happened along the lines of your question. One of was that sedal was beating 75 offers of his top level human opponents before he played alpha go. After he got done playing with alpha go he was beating 90 of his human opponents so he made a big quantum leap in his game, already very high level. The other thing that happened is the salutes go boards boards ann korea and asia skyrocketed and little kids started playing the games at much higher volume than before. So one thing we hear, isnt taylor wish no longer the best at this particular activity and doesnt that reduce human value or dig it . I think thats dead flat wrong because if we define ourselves by the thing were best at, thats to silly thing to doment we have this neat thing happening that people want to be part of. If technology is part of that neat thing, thats good. That do you think . I think its good. Let me understand scone pint that andy made. He got better because he played a bun of games with the machine . No its because this machine came at the game from a fundamentally different way. When they saw some of the moves the machine made, the go experts said, i at first, my god, that is idiotic. Theres because ear. Every beginner knows not to do what the machine just did and then a few later the mon won and they are like, oh, i see what he did there thats really innovative. So for 3,000 years humans have been learning from other humans, from their masters. If if the made the move the machine made their mast we are slap their wrist and sky, dont do that. But the machine came at it with such perspective they have a different way of thinking if you want to call it that and came up with innovative approaches and not just true in the game of go. Its true in all kinds of problems. The machines righthand side just better in certain ways but most importantly theyre gift in certain ways and they come at it with a different perspective and the possibility that humans and machines can Work Together and help each other because theyre not just two clones of the same thing. Their two different kind of intelligence. What is cool about the stories one thing that is cool is that you probably needed to be a player somewhere knee the level of seddal to get the insite why alpha go made the move it did. You need to be good at this domain to start to unpack why the machine was behaving in this way. The machine was mute why it was doing what was doing. We need really high level humans to look at that and say, think i know whats going on here. Lets test that idea. Where are some companies parts of the economy that were already seeing this. So people come to the website and have a chat with sales reps, is this course right for you, human sales reps. Stay tuned. Some of them are better than others and answer questions effectively and convert people and they are a good course for the sake of it. Others not so much. They recognize the transcript of all these chat become the data for supervised learning which is giving examples of failures to a machine and let the machine give the transcripts and their Machine Learning outlets are patterned, successful sales reps have ways of answering questions. At that point they do not turn it over to the boss because there are too many unstructured questions that there is an 8020 rule, very common questions that account for a big percentage of queries and a lot of humans interacting with incoming traffic but one one of these common questions came up, heres a phrase you might consider using. It has worked well in the past and humans pick up on that and the not so good sales rep, performance improved dramatically and the sales closing rate went up by 50 . A tremendous improvement but part of the with the ability to combine humans and machines in a way none of them would have been successful on their own. In some cases we will automate work, from mind to machine, that is the appropriate thing to do but naive to think that is the only force going on. We love that story because it shows fans of technology and automation realizing humans involved in this process but we amplify their abilities or turbocharge their abilities with Smart Technology in the background. Each of those, we are not seeing turn the dial all the way to one extreme. We are saying you got to about the balance. It makes sense to automate the platform but in most cases you need the balance and the balance is more toward machine platforms than most companies or individuals realize but not as simple as the first part. This question comes up a lot, and the advent of machines, we were talking about, would you like to quote it . The only difference between the unlinked series of observations is on the part of a songwriter. He is lowballing really badly because it would be trivially easy to program a computer, might rhyme and might have words associated with love and loss in him. We are incredibly confident that the lyrics would be not making a lot of sense or activate any emotion on the part of the rest of us. Great human lyricists have a barely linked observation in a way that resonates with other human beings and that is the skill we have not seen Computers Come up with yet. Theres a concept of linguistics i find incredibly helpful here called the intuition of the native speaker and what that means is if i look at any length or complexity in english, i can tell what is grammatically correct or not. Before my brain processed its meaning or know what it is trying to convey, that is kosher or not. I speak some french but i cant do that infringe to save my life, english is a second language, they can do as well as i can or any other native speaker can. We humans have native speakers intuition about the human condition, the social world we created, the physical world we move through. Trying to teach computers to have that native intuition. If we do that i believe that is decades away. They are secondlanguage speakers about the world we have created. A lot of confidence human ability to invoke an emotional response via words and language, if that is automated quickly, they are surprised. It is hard to tell apart from the music of famous composers. Not hard. The interesting question is what is the difference between music and lyrics and what about that makes it automatable and not . I am not a musicologist but the aesthetics of music are well known. Music, we know what sounds good to our ears and what sounds like sirens clanking on the new york street. We want to encode that stuff and we know the rules of these kinds of things so relatively easy to spend those out. Machines that are doing a good job are not encoded with rules or courts. There are patterns that emerge. We dont have to know anything about this deep structure to see these patterns. Machines are good, shadow pattern recognitions but the lyrics to be evocative, you need to understand the meaning, not just two words adhere to each other in sequence. If you want to go beyond that you need to understand the human condition and the machines although they can recognize patterns in words, they do not really understand the words. Machines can translate one language to another, but they are not understanding what was going on. Machines can be good at certain video games, Quick Reaction shooter video games, shoot the missile, great at that but if it involves deeper planning and structure they are hopeless and i think language is like a complicated videogame and music more like pattern matching. We include in the book an excerpt from a Machine Learning system that gets trained on jane austen novels and generated jane austen pros blues not that it is gibberish that makes your head hurt. It is unpleasant to read. Unless we can make a go 180 different i dont think that will happen anytime soon. It is often said when you look at jobs and what jobs are most likely to become automated, the divide isnt high school low skill so much as highfrequency tests versus variable unpredictable tasks and the broader lesson is machines are not just good bu