Will we all be replaced by Robots?
Good. Evening ladies and, gentlemen. My. Name is Avinash Mishra. I'm. A proud Chicago, Booth alum and, I. Work for the school now I'm. Director of India and South Asia outreach. For. Executive education and. External. Relations for the school. It's. A pleasure to welcome, everyone. Here in this evening. Today, especially. Our colleagues from the. MHRD. Fraternity, thank you doctor the hinge a Singh for, all your support in this a. Fellow, booth alums in this room, Akshay. Wherever. You are thank you so much for putting, this together with us. Our. Colleagues, from the UChicago Center in Delhi and the, you Chicago alumni community, Aditi. Thank you so much for, all your support in bringing this together. A. Little, background about the Center in Delhi the. Center established in 2014. This, place that you hear in today in. Establishing, the centre in India, the, university. Draws. Upon a long and rich history of. Excellence in, scholarship. Research. And teaching related to South Asia, the. Center in Delhi in last, four years since it, has been set up already serves. As a base for. Undergraduate, graduate. And professional, students. Studying. In India in the. Last year itself the, center has hosted hundred plus workshops. Seminars. And, lectures. With. Our faculty students. And alumni in the region and, over. 60 faculty from various, departments, in UChicago, have, visited, the center since, our opening. As. Part of the ongoing endeavor, to bring best. Of units of Chicago. Intellectual. Capital to this part of the world we. Initiated the unit of Chicago leadership, the, roundtable series some time back and, in. The same. Line. Of thought the. Intention, is to bring, insights. Ideas, and, perspectives, from. Our current research for, the community to this part of the world. We. Are delighted to host this evening's, lecture by. One of my fav professors, professor Michael. Gibbs here who's. A clinical professor of economics, at Chicago, Booth and his. Study. Covers. A very relevant topic. In today's world will, we all be wiped. Out by auto sorry replaced by robots. Thought. About they call apocalypse. Professor. Gives studies the. Economics, of human resources, and organizational design. And. He's. The co-author of a leading textbook in this field personal. Economics, in practice, and. This. Book has been translated at, least five languages, and the Chinese version I think is coming down soon. In. 2007. Professor Gibbs received. The notable, contribution, to management. Accounting, literature. Award. From American, accounting. Association. And, has, received three Hillel. Einhorn, excellence in teaching abouts and. He earned his bachelor's. Master's. And, PhD in, economics all feminist of Chicago I trust. Me when you have a, an. Economist, and that too from Chicago. Booth talking. About HR the. Perspectives, are going to be very different and you're, going to walk, out with a different world view from this room I was. Very fortunate, professor gives to attend your class, well. In my program in MBA and one. Of your topics was, compensation. And difference. Between stocks. And options, one. Of your insights made, me a little wiser a little. Bit richer after. That and I hope I had paid more attention I would have been much richer now. Thank. You so much for, joining us this evening and over, to you it's. A pleasure to be back at the deli Sara this is the third time I've been here I was, here when it opened I was here for pie bar and this is and. I'm sure this won't be the last time. It's. A real pleasure to see a as far as I can tell, at least seven of my former students, in.
The Room from our ADP, axp, and exp, programs, and some, other alumni and, I. Appreciate, those of you who are not alumni, coming, as well making the, go into the trouble of coming to our Center and I hope you come often, before. I get started I want to thank NH Rd for, sponsoring. This program, I appreciate, that and our. Alumni club and, finally. Also want to mention that I'm here in Delhi on. Funding. From a grant received, by our Tata. Centre at the University of Chicago. The. Reason I'm here is to, try to start, a new research project with, a local company I can't, reveal the identity, company because. We'll be doing confidential. Research. But. I do appreciate the funding and that's another reason why I expect to be back in again so as. Well and of, course thank you to the Delhi center, into a blush. And. A DeeDee all. Right let's get started, let's. See is this a touchscreen. Alright. This. Is a classroom at the University of Chicago which, means that, you. Shouldn't be quiet if you have questions, or objections. Or. Perspectives. That you want to add in don't be shy don't hesitate, to. Raise your hand and talk at any time don't wait till the end okay, how. Much time do I have three hours three and a half okay. Just. Kidding all. Right so what I want to talk about is, I'm. Used to walking around when I teach so it's a little frustrating but I've been. Asked to stand here so it's a little awkward but I will I will I will. Comply, since this is being put on Facebook, I should. Probably have a couple of former students watching. Right now hi guys I want. To talk about the, labor market robot apocalypse. This is something, that is in the news a lot has been for about the last five years and it's something that I think we should all be. Legitimately. Worried. About or concerned, about at least, and I want to try to provide some perspective on if. There's going to be an apocalypse, or not and why this, this, is based. A little bit on some research I've done but Laura what I'm mostly, going to talk about is a lot of research done by other economists, and academics and sort, of tie it all together, okay.
So Here's, a couple of example, headlines. Merrill. Lynch warns of robot apocalypse in The Wall Street Journal, last year in the, feed is your job at risk of the upcoming robot apocalypse. And, then Reuters millions, of jobs may be lost to automation, in the next two decades, and you can find many many more of these and I'm sure, you've there there have been some in the indiatimes and so forth as well. And. Many. Of these headlines traced to a specific, academic study, that was done by artificial, intelligence, researchers. At Oxford University that, I mentioned later, I think that kind of spawned, this. Media. Attention. But. In any case, automation. Is something, that has been. Proceeding. At a relatively, rapid, pace recently, I'm familiar with some executives, at HCL we have two executives, from HCl in the room today and, I. Was talking to one of them yesterday and part, of that the conversation, was about an, automation business, it's, one small part of HCl as many lines, of business so. You guys in some sense are creating the robot apocalypse. You. Know you're working for Cyberdyne, systems. That's. The that's a Terminator movie reference but, this is an old idea, in. Eighteen, in the early, 1800s, in England the Luddites were, a movement founded, by Charles Lud who was concerned, that textile. Workers were losing, their jobs to automation because the first automated. Textile, machinery, was, being installed and, of course textiles, are a big part of India's economy now, in some, areas right and here we have a picture of them destroying, the machines and the factories, because. They couldn't convince, the, companies, to stop firing the workers installing, machines, and they only thought, they thought the only thing we can do now is rise. Up in rebellion. John. Maynard keynes is one of the most famous economists, of all time in 1933, he, was talking about automation. Outrunning. The pace at, which we can find new uses for labor and he was concerned that we'll end up with mass unemployment. In. The, 1964. In the united states the. US government, set up, one of these blue-ribbon. Commission's, of, important. Mucky-mucks. Called. The ad hoc committee on the triple revolution, I'm not sure with the triple revolution, was but it probably was telecom in computers, and something. Related to both of those, it. Had several Nobel laureates, and distinguished. Academics. And government people, and business people and so forth and so on and in. Their report, they concluded, that the cyber nation revolution, which. Is a great, name a great movie title the. Cyber nation revolution, would lead to almost unlimited, productivity. With progressively, less labor and then. Finally, as, I, was looking, around for articles related to this I was amused to find an, article from 2009. In which French workers had seized their Factory and we're trying to blow it up just like the Luddites and France. Being France they. Did it again in 2017. So these issues have recurred, throughout the, last two or three hundred years and I'll come back to that point as well. Before. I get into that I'm going to provide a little perspective, and this is actually from some research of mine I did. Some research with some colleagues one, of which works. For the United States Department of Labor which is government agency, which collects data. On the US labor market and she. Was able to get us access to an unusual, data set it was a 10% random, sample of the entire US labor market very large data set which. Included. Information on how the jobs, were designed, and, research. Of this kind on how jobs are designed, is fairly. Rare, and so. We had information on, character, on four different characteristics. Of 10%. Of the jobs in the United States the, first was the extent to which the jobs were. Were. Ones, in which your work is closely, dependent. With your colleagues, or you, work relatively, independently, the. Second, was a, measure, of the extent to which, someone. Performing this job does, many tasks, or has a very specialized, job so, specialization.
Through Multitasking. Continuum. All of us have jobs with where, we do lots and lots of multitask, but that many factory workers for example do, very, repetitive simple. Jobs the. Third dimension was the extent to which in your job you're given a lot of discretion, about how to do your work think about that is decentralization. Or. The lack of it is, this a job where you have authority to, try, new things to, choose, your own methods. And. So, forth and so on or it is a job where you're essentially told what to do and how to do it and then. The last was a measure of the skills that would be required for someone to perform this job, okay. So we did something very simple, we. Took these jobs, these, information, about these jobs and we, we within. An industry and within an occupation. In that industry we had a set. Of PEEP of jobs which were, relatively. Similar to each other and then we just ranked them against each other so. For every, secretary in, banking, we, we. Ranked. Your your. Measure of independence interdependence. Multitasking. Discretion, or skills compared, to other secretaries. In the banking industry. So, the way we defined it very simply was if, you, were at the median we call that middle, level if you were below it you were ranked low on that dimension and if. You were above it you were ranked as high on that to mention so. Now we have, some, simple metrics on the job we have four dimensions, and we have low, medium. Or high, OMH is what I'm going to refer to the mass here and. Let's. See if there are four. Dimensions, times. Three dimensions, so let's see what do we have sixty-four, different combinations. I think I, think. I know that's right because you're shaking your head yes you. Did the math right oh okay. Sixteen, I think it's 64, 64, different combinations. Of jobs, where you could be low on one dimension middle, medium on another low, on a third high on the other and so, forth and so on right now. If these four dimensions of how jobs are designed, are, unrelated. To each other then. You can calculate the expected. Probability. That, a job will have will, be each of those 64. Categories. Based. On the. Probabilities. That. On each of those dimensions a job in this group of people is low. Or medium or high all. Right so it's a very simple, nonparametric. Statistical, test so, we come up with predicted. Likelihood. Or a predicted, distribution of, job design, assuming. These are independent, from each other and then, we compare that to the actual, designs. And what we found was a very striking pattern, of those. 64, different, types, two. Types were. Extremely, common, they were about 30 times, 240 times more. More. Observed. Than, would be predicted and, those. Two were job designs that were low in all four dimensions, or, high on all four dimensions. The. Other job design, which was more which, deserved more than would be predicted was. The one where job was medium, all of those dimensions so these are jobs which what, we see is jobs fall into patterns which. Are in some sense coherent, you tend to be low, on all of these four dimensions in the same job or medium, or high and especially, be. Extremes. All low or all high I'm. Going to call all low jobs classical, job design and I'm going to call a high, modern, job design, alright. So. Think about it is we have four different measures. And if they're statistically, independent ability or uncorrelated, Oh.
Pick. Discretion, this is extent to which you have a lot of authority in your job or you're told what to do it these. Are, so. We have three too low, medium high those. Probabilities going to add up to 100 and then you, go from there yeah. Okay. The. Second thing that we found the second pattern was jobs that are high in all of these what I'm going to call modern, job design, or. In, industries. Which are. Relatively. Rd intensive, so. They're changing rapidly there, and, job, and industries, where, there's relatively, high spending on information, technology so. There's a relationship between these, and both. Change and, in technology, and I could talk about this all day, I've, done. A lot of writing on this particular, idea but I want to link it back to this question about robot, the robot apocalypse. Alright, so why, two different approaches, to job design, extreme. Ones classical. And modern the. Way I think about it is classical. Job design first of all what is classical job design this, is a job where. Someone, does very specialized, work. 1, or 2 or a small, number of tasks, over and over and over again so. Factory. Workers a classical, example that but very very simple, clerical, or secretarial, work would be similar. Their, work is relatively independent, of others, they. Have little. Or no discretion. Over how they do their work they are told how to do their work and as. You might expect in a job like that such. A person doesn't need very many skills and they don't need high-level skills including. Higher-order thinking skills cognitive, skills because, they are, told how to do their work all. Right so where are those kind of jobs going to make gonna, be seen they're, going to be seen in cases where we've already figured, out the best way to do something so the classical. Way to do this is to bring in consultants, who do Industrial, Engineering who. Help a company figure out what are the best processes. For. Designing, a factory or designing, some kind of, insurance. Company's, process, to deal with claims or something like that Frederick, Taylor was the first industrial, engineer one of the first business, consultants. In the world he. Was famous for helping the Ford Motor Company set up the first mass. Assembly, line in Detroit and he, used exactly two these kinds of principles. And, when. He did that what when you're trying to optimize a business, process that's a very difficult problem, to solve and the, way you tend to do that is you break up that business process into. Different pieces and then you try to perfect each of those and if. That's still too complicated you break it up even more and you focus and you focus and then so you've taken the whole business process you, broke it into two discrete steps, and then.
You Try to perfect, how to do this step and that, step and that, step in that step and in, some kinds of work, that is very easy to do UPS, does a lot of this kind of Taylor istic approach but, UPS is a very simple business take, a package from one place and bring it to another and they've, been in business for about 120, years and. Very. Little technology, has changed during that time they went from bicycles, to, motorcycles. To trucks they, added airplanes in the 1970s. And then there was barcode, scanners, and tracking, of packages, in the 1980s. And so forth but the business is pretty simple and they've had a long time to perfected so, they literally tell, the drivers how to step into the trucks so, they could be faster, alright. That's Taylorism, so, when you've done that when you've broken up the process, into, discrete, steps and broken more and more and more and then you figure out how to do this step exactly, the best way, it. Doesn't, take much to then say okay we'll have one person to do this and do, it over and over again and do it exactly the way we've figured out is the best way to do it and, what. Does that lead you to a specialized, job in which. We, don't want you to think and figure out a better way to do it because we've already know the best way and since. We know the best way or something close to the best way we. Don't want you to try it any other way because the way we figure it out is a relatively, efficient, way so we're going to give you a, specialized. Job with low discretion, and you're, not going to need a lot of skills then and since, we've perfected. Your job sort of independent, from everything else you're. Probably not going to be too interdependent. With your colleagues either and that's classical, job design alright. And, here's a quote from Frederick. Taylor's book about his practices, which gives you some idea of how. He thought about workers he described. Factory. Workers in. Steel. Plant handling, pig iron as stupid. And flah mattock. Resembling. And mental make up the Ox it's very inspiring quote, I'd like to give us go to my students, you know just to make you really feel good about yourself, you know. What's. He saying there I don't want skilled workers I basically. Want someone's gonna do repetitively, what I tell them to do because I'm smart, and I figured out the best way and and, they're, not and you, get the idea now let's, talk about modern job design modern job design is the opposite, when. You can't figure out best practices completely. In advance that means there's a lot of opportunities. For further, learning and, optimization. As you, go. Okay. And when that's the case you. Can use the employee, as someone. Who does that learning so this is continuous, improvement. This. Is ex-ante, optimization, this is continuous, improvement and for, continuous, improvement what are you going to want to do you're, going to want to attend to design a job so it's relatively, complex, so that closely related, tasks. Are being performed by the same person, or by, that person's, teammates. So. That the, person, understands, how the tasks, work together and can. Solve quality, problems that often occur when two parts of a process, don't mesh perfectly, so. You're going to move from specialization. To multitasking. You're. Going to give them much, more discretion. Because what you want them to do is think and observe. And develop. Hypotheses. About problems, they see quality. And so forth and so on and then. Develop. Experiment. With ways to do. Their job better and. Once. They find a way to do their job better you want them to implement that way so you're gonna have high discretion, as well so you have a broader job there's going to be more interdependence.
With Your colleagues, and we're. Going to expect, you to think and yourself. And innovate, so. We're gonna need a higher skilled person, and. Every job is on that continuum, to some extent all of our jobs our modern job designs where. We're doing lots and lots of innovation, and continuous improvement and, so forth and so on but. Even within an occupation, there's variation, and we tend to see those two extremes and again. I could go on and on but basically think about this as cases where we already pretty, much know. Have, a pretty good idea of how to do things the right way you. Know maybe this is a restaurant, where you've got recipes, that you perfected and so. Forth and this is where, the, environment. Is changing. So. You're gonna do a lot of R&D or IT spending, and things like that and you need to learn as you go because you haven't already figured out the best way or you're trying to innovate and create the future like, at HCL or something like that okay. All. Right so now let's go back to the robot apocalypse. They. Through there. Has been basically. Two ways in which automation. Has affected, jobs. Historically. In, one case jobs. Have been. Replaced. By. Machines, and that's. The case where machines, are able to substitute, for humans they're able to do the work that humans do, but. There's another case and this is the case where technology. Has enabled. Has. Complimented, the work of humans and in, this case it's made humans more productive, instead of replacing them it's made them more valuable so. To illustrate this. Distinction, I'm going to use an interesting, example from, a book by economists, to focus on this issue. It's, actually an old example, now but it is a nice example so I'm going to talk about it and this is aircraft, design so in 1962. Boeing came out with a 727. Which, had about a hundred thousand parts and could handle, about 130, passengers. This. This development. Process took 81 months and 5000, engineers, if you, were an engineer at Boeing in 1962. Designing, the Boeing 727. Think. About what your job was you. Were drawing blueprints. And I mean you were drawing them you're, using paper and, pencil. And erasers. And rulers. And compasses. And protractors. And, you. Had a calculator but it was a very crude one with a hand crank or something like that and it, you couldn't do very complicated, calculations. And so. You spent a lot of time, drawing. And then. Redrawing. And so forth and so on. And. Your. Output would be a stack, of paper. Alright. And. As. You can imagine this was not a very efficient process this, is why they needed 5,000, engineers and when, the work of these 5,000, engineers was combined to actually build the plane what. They had to do is take my blueprints, and hablas his blueprints, and a.d these blueprints, and so forth and reconcile. All of them and since, they were all on paper and they were done by hand calculations. And, you know humans, make mistakes and so forth they. Had to build a full-scale model of the plane out of balsa wood to, try to make sure everything fit together all. Right and then they had to go back and erase and, fit through the calculations, and fix the blueprints again before they finalize, the design and then, they finally, were, able to manufacture the 727. Now. This process wasn't, very, it wasn't perfect. And. Even. Though, they tried as best as they could when they constructed, the actual planes out of metal, they, found that sometimes the parts didn't quite fit together, right there, would be gaps between them if, you, look at an old plane you. Will see a lot of what they call shims, this.
Is A you know a metal, strip that they use to cover the fact that there's a hole where the two pieces are supposed to fit together or, some, foam rubber or, something like that filling, in the gaps okay. So these planes weighed 44, tons, and. Since we're in India I think I can use the, non-metric numbers and you'll understand it and of. Those 44 tonnes half, a ton was these shims. Basically. A fancy word for duct tape now. Think, about that next time you fly in a 727. All. Right and, they're. Acceptable, tolerance where. There would be a gap between two parts except. For the fuselage was a half-inch now. You go to 1994. Which is a long way in. The past from, today but it was a big, change for 1962, the Boeing 777. Was. A much larger, plane it took 52 months to develop I don't know how many engineers, it took but it was far less, what. Was different is engineers, were using, sophisticated, cabs, can, software. That was developed by an aviation, company in France so. Now think about your job as an aviation, engineer, you used to be drawing a paper and pencil and crude calculators, and rulers and protractors. Now all, of, that stuff is done in computer for you so. What does that mean this, frees, you up from a lot of the drudge tasks, that you've been doing and it allows you to experiment more what. Would happen if I change the, curvature of the fuselage, what, would happen, if I move this part a little bit that way or a little bit that way could. I think about customization, where, I have several different models now because I don't have to spend all this time doing the hard part the, computers, doing that for me I can. Think about I can think more creatively I can think more innovatively, and I, don't have to focus on one design I can focus on I can do multiple. Designs and in fact bong was able to customize their planes far more so. What happened, is we needed fewer engineers, but the engineers, that we used, were, far more productive because the computer didn't replace them it replaced some of it tasks, but. Those are the lower order repetitive, tasks, the calculations. And so forth and that freed. Them up to, focus more of their brain on, the tasks that were hard to have to automate. The. Design, the creativity, innovation and, so forth and that's. The pattern we've seen over and over and over again most. Of us were not replaced, by computers, in any, of our previous jobs, and all of us are much more productive now because, we have computers, and telecommunications and. I can have. Teleconferences. With executives, at HCL in Chicago, and you know collaborate with them even though they're seven. Thousand miles away and so forth and so on oh and, by the way they needed many fewer shims and the tolerance, was much much lower, and I'm sure it's even better 2018. All. Right so I just described how technology, has two, very different effects on two very different types of jobs and employees. And. This. Is leading to some dramatic effects on the labor market so that's not the main part of my talk but I'll talk about it for just a minute. So, on the left here I have a figure from the textbook that hablas mentioned. That. I that I wrote with a colleague co-author, of mine, and. What it's showing is. Two measures of the. Relative. Compensation. Of high-skilled people compared, to lower skilled people, plotted. In the United States from 1970. Through 2005. And this, textbook was published a couple of years ago I could probably, update it now to 2016. Or 17. So, I'll just show.
You The one that's closest. To me this is the, average, hourly, wage of somewhere the college degree divided by the average hourly, wage of someone with a high school degree in the United States so, it's a measure of how much does a labor market value. People with more skills compared to less skills and the. Point of this figure is. You. Can see. The. There wasn't much change until about 1980, but starting in 1980. The. Relative. Pay of high-skilled, people has, risen, compared, to the relative, to pay of low-skilled or medium skilled people in, any, way you measure this any. Other measure of skills you find this pattern since, about 1980, and this pattern has continued, through 2018. In this, pattern is global it's, true in India it's yet it's far. More true in India I would guess in the United States, it's, true all of Asia Europe Middle, East Latin, America, and Africa, to. The extent we can get data from all these parts of the world is a global phenomenon and if, I had pushed this graph backwards. In time the, last time we saw any kind of change. Of this significance. In relative. Compensation, of high versus low skilled people was. Back in the first Industrial, Revolution for. Most of the 20th century these ratios. Sort of bounced around but didn't have any trend okay. So this is a big deal. All. Right so, and, the reason is because if you're in a job where information technology, complements. You because you do higher-order, thinking. Cognitive. Work creativity, innovation. Leadership. Things like that, computers. Make you more valuable but, if you do simple classical, job design types, of work those. Are the kind of things that are easy to automate you can you're likely to get replaced or at least you have to now compete, with machines, or software, right. The. Second, picture is a, measure, of the change in. Employment. Within occupations. So percentage, change in employment. In different occupations. And. These. Are measured by in, by, breaking, people in the same occupation it's, a low middle. Or high levels of compensation similar, to what I did with the job design data earlier this is a picture I, borrowed. From a study by an MIT economist, and, this. Is data from the EU. Plus. England. And. What, we see and so, this is 1993. Excuse me of 2010, and what, we see in all these countries is, the same pattern, we, see a relative, increase in, jobs, for high-paying people, some. Increase in jobs for low-paying people the green and the. Middle is dropping, out so. We, used to think based on pictures like this that what automation, was doing was destroying, jobs for those with low skills but, it's more complicated than that what.
It's Doing, actually. Is replacing. Jobs for those with middle, levels of skills. Okay. So on the high end our. Kinds, of jobs are hard to automate but. On the low end there there have been, until. Recently at least. Jobs. Which, have skills, that are or tasks, that are hard to automate - -, in particular certain kinds of physical jobs, okay. So if. I was allowed, to walk around this room, randomly. In this unmapped space that would be hard for a robot to do until recently, but it's really easy for humans to do robots. And machines have, been varied have not been able to do fine motor skills, although. They're getting better at it recently so, if. Jobs involving, physical work have, been ones that have been hard to automate those tend to be low skill jobs and the, second thing is service jobs where. Some kind of human interaction, is important. So, teaching would be an example of a low skilled job involving, human interaction, and services. And. Those, have so, that's what you see so if you combine these you see rising, inequality. And, pay for those who have jobs high. Skill versus low skill or medium skill and also. The middle has fewer jobs putting. Those things together, tells. You why we've had growing, inequality. Dramatic, increase in economic inequality. Globally. Since 1980. By. The way why since 1980, because. That's when computers really, started to be useful personal. Computers started being deployed by corporations. All over the world. Ok. And, we're. Seeing some political. Fallout from that no. Doubt you're seeing some of that in India what I'm most familiar with is. Breaks. It in the UK because I teach in our London program, and. A lot of that is based. On fears of my, being replaced. Some of it is misguided and is is, related, to fears about international, trade but it's really based more on automation. And, in the United States we have. Donald. Trump as our president and Bravo was a serious candidate needs basically communist, you. Know and both, of them, were, successful. Politically. Because they appeal to those in. The middle skills who are fearing that they're gonna lose their job or have lost their job and their, compensation is stagnated, if they're still employed very. Successfully, okay so I'm, guessing, you have similar dynamics, and if you don't you will. All. Right now let's go back to the robot apocalypse and, away from politics, so there's. Probably a couple of you in the room who, are working in artificial intelligence, are there out, of curiosity a. Little. Bit okay so I, I. Don't understand, it very well and if I miss date something please, speak up okay so, you can clarify so, basically, what, I want to do now is talk about recent. Years so what I've given you a sort of the story up until 2010. Or so. The. Reason, things may, be different now is because artificial, intelligence, while. We've been talking about it for about 50 years it's, only in the last ten years or so that it has started to provide tangible. Applications. In the business world. Largely. Speaking, and. Artificial. Intelligence, is attempts. To make computers. Mimic. Human. Thinking, human, cognition, and. Therefore. Might start threatening us in. Our high skill jobs as well so I want to talk about that that's where the real fear is coming from are, things different now because of AI all. Right so there's different flavors of AI the simplest, version is called, data mining so. AI is becoming powerful now because we have vast amounts of data and we have really really cheap computers, and we have developed techniques, and statistical, methods and so forth to put those two together and. We. Didn't have that stuff 10 or 15 years ago so the first is data mining where you take vast amounts, of unstructured data and, you basically look for correlations. Patterns. There's. No theory, behind it but you can find patterns you, those, patterns, may be able to help you predict it may have applications. Good, example that might be interpreting. A radiological. Scan like an x-ray or something like that, machine. Learning is the next step up where you're actually trying to get, the machine to make some decisions, not just make some rough predictions, or look for correlations, what. Machine learning does. Is it defines an objective, for the computer, to maximize, or, minimize sometimes. Called a cost function as an economist, I would call the utility function you define an objective, for the computer. You. Give it some initial, data and some initial ways, to process that data to make decisions if, you will and. You. Allow.
It To take, that data make. Those decisions, and that will lead to some outcomes, and then. Allow, it and then you keep feeding in new data and you allow it to test. Out new versions. Of those decision, trees if you will and evolve. Them over time. Ok, so the machine is learning, in that sense, it is. Changing. Its algorithms. To better, maximize. Whatever objective, you have set up for it make sense all. Right yes. Sir. That'll. Be my next one yeah. Absolutely. Today, twirl this is something that a eye researchers, are trying to figure out as I understand I don't think they have figured out they know there is this problem you. Know if. Technologists. Are men the, machines are gonna have thinking like men or something like that or yeah, you know. And. They're. Trying to figure out ways to develop, the algorithms, to avoid these biases, and I don't know whether they're going to be successful, it's hard to say, part. Of the problem is the utility function of I as I've called it here, are the objectives that you are supposed to pursue, are ones. That we have to define and we define them as humans so, but I don't know much about that but it's an area of active debate, and concern, right now, ok. So, by the way let me give you an example of utility, for and what I'm talking about our cost function if you're designing or, if you're trying to design. Autonomous, vehicles. One. Of the things that you have to do is tell the computer, who to kill and who did not kill in certain situations, if. There are two pedestrians, which, one is going to be doomed because. That decision is going at we made you have swerved left you're gonna have to swerve right and the, computer, is gonna have to have some guidance on how to do that so that's an example that's a dramatic example but that's the kind of thing you have to think about if you see how human biases, might come in there right okay. So neural networks are, the more the most advanced, and exotic version of this where you're. Actually trying, to mimic human thinking, neural, network scientists. Work. With, neuro. Scientists. And try to understand how does the human brain actually work and can, we use that to guide our design of. Algorithms. Basically. There. Are two parts to that the first is some, level of abstract, thinking and the second is some hierarchical. Thinking and let me explain what I mean by that so. It. Is common now for computers, to be able to, scan. A signature and interpret, and figure out this is my signature it's not my signature, ten. Years ago this was giving an example of something that computers probably would not do in our lifetime but. Because of neural networks, they're now able to do that so how do they do that so, first. Of all. Abstract. The first, thing that the computer. Needs to do is visually, look at a piece of paper and, then, it needs to say here's an area where there's dark, stuff with, a light background so that might be ink on paper and focus. On those areas and, then focus in more so that's one. Level and then take a look at that and decide okay this is a signature, and if, it's signature, then you have to try to parse it is this a letter is that, a letter is that a letter and so forth and so on that's a next, level of. Abstraction, it's a different algorithm, in some sense but it's a different step hence, the term hierarchical. And then once you think you've defined a letter then, you have to have an algorithm that tries interpret, what, letter is is a B C D and so forth and so on and and. Neural, networks kind of work in that direction there's much more complication, to it and I don't understand, it very well but that's basically what's going on and that's. Interesting, because that allows computers to do things like. Have. Visual, capabilities, like. Recognizing. His signature halves language, capabilities, because exactly what I described is how computers. Parse language, to.
Be. Able to speak a sentence, it may be an imperfect sentence, but in some crude, sense communicate. With humans and so. Forth and so on so the developing, Croods social, skills if you will some. Visual, skills and the visual skills allow computers, to more, easily move around an unmapped space or pick something up so they're getting better at physical tasks too so. This is actually, threatening, low-skilled. Jobs to not. Just high-skilled jobs but to accept they're good at these things they can start doing some of the things that. High-skilled. People would do so let me just here's some examples, of the applications, we're starting to see in there many many more, most, of you are familiar, with Siri that annoying thing on your phone if you have an Apple phone. Which. Speaks to you and. Tries, to interpret your language and often as mistakes or Amazon's, Alexa or so forth and so on. Another, example is anesthesiology. Anesthesiology. Not, anesthesiology. Well yeah anesthesiology. Is an example it is common now for computers, to make recommendations about how to treat. The pain of patients, during surgery, so. For its recommendations. Than are making decisions but they're making recommendations, for the doctor. The. Example I wanted to talk about is. Radiology. Cat, scans MRIs, x-rays and, so forth and so on. A college, student decided. For a class project about a year ago to try to develop a program that could interpret x-rays, he spent about a week on it, ran. His program with a lot of data from x-rays did machine learning techniques and so forth and so on and almost. Immediately, his algorithm. Was more accurate than professionally. Trained and experienced radiologists, at detecting, say a tumor, on an x-ray, imagine. If you are an anesthesiologist. I'm sorry, a radiologist. Who has, stood. Owns to pay off because you went through expensive, medical school and you've been doing this for 10 years and coming an expert and this college kid has come up with a machine. That can do it better than you so. Our in our radiologist being replaced by machines.
So. Far the answer appears to be no what's, happening, is. First. Of all. Radiological. Scanning, is incredibly. Cheap now, I remember, when there's one MRI in the whole city of Chicago and it was a huge deal and you could never get access to unless you had the best health insurance, and some, really serious disease, but now, any local. Clinic can have an MRI machine because. They're very cheap all, right so we have massive amounts of data coming, out of these radiological. Scans radiologists. Have more data than they can have time to look at but. If we can pass all those through a machine we, can process them. We'll have a lot more information that the radiologist can, take, into account and. The. Machine can then flag the anomalous, ones and say take a look at this one specifically, because something's strange here and then. The radiologist. Who's still better at the sophisticated. Interpretation. Of those, scans can. Focus their higher order skills on those, and let the machine do, the the. Processing, it's very similar to the aircraft engineers, so. Radiologists. Start debating all of this and there are journals right now but it looks it may be the case that we actually have more value, for radiologists. And we may actually, see an increase, in the demand for human radiologists. Because. They're more productive and cheaper at the same time or give better information. So. I have to put in a quote here from Sachin adela who's. From Hyderabad, and a. Graduate. Of our MBA, program because. He's the the. CEO of Microsoft currently. Talking. About artificial, intelligence I gave a talk at Microsoft's, Innovation, Center in Milan. Last. Summer, so, these were the AI researchers. Who are. Heading. Their efforts in Europe and one of them claimed, during, the workshop that within 50 years he thinks that the techniques, they're using will, be able to do anything. Humans. Can do anything. Like, give this lecture and, do my research. I'm. Skeptical, but that's what he thinks anyway. For. What that's worth okay oops, yeah. Now. Let's I want to go back to that Oxford. Study that I mentioned from 2013. So artificial, intelligence, research just Oxford University. Wanted. To answer the question. Using. The technology, that we are developing, now, and we see coming soon. What. Fraction. Of the labor force do you think we will be able to replace. By. Automating. Their jobs in the next say, 10 years, so. They, got a, collection. Of descriptions. Of different, jobs 700, different jobs describing. What's involved in doing those jobs, secretary. Or something like that professor. And they, randomly, selected, 10%, of those and they had their AI researchers, read those descriptions, and answer.
The Question do I think I could automate this job soon yes or no and then they. Fed those 10%. Algorithmically. To process, to, come up with some rules machine, learning so. Then process, the other 630. Occupations. To make decisions. About those as well so they were using AI in some sets to replicate their own thinking and from. That they. Then took. Those occupations and, looked at the percent of jobs in the workforce that, are currently, in those occupations and, asked. What. Fraction, of jobs are likely, to be automated not, like it be automated by occupation. In the next 10 years or so and what, they concluded was, about 50 percent of jobs will be eliminated, with about 10 years maybe 10 to 20 years that. Was terrifying. And that's, where all the headlines came from, half. Of our jobs will be gone done. By machines autonomous, cars and autonomous, professors, and everything else in, between this, plot is taken, from their paper. And, it. Shows on the horizontal, axis the probability. Of computerization, of, a job and the. Different colors are different kinds of occupations, almost. Everyone in this room is the light-blue management, business and financial, really. Good news your, jobs are really hard automate even using cutting-edge AI so, you're probably gonna be okay you. Know I figure I'm gonna work another 15, years maybe 20 so you, know this. This I I might be okay -. That's. Me education, the light green you. Don't see much automation, what's, easily, uh what's, likely to be automated they think a lot of service jobs which, have been hard so far sales. Jobs. Now. That's not going to be the high-end sales where you're selling things. Long sales cycles, complicated. Contracts. And so forth and so on but maybe sales in, in. A restaurant, or a store or something like that in fact we're seeing kiosks, where you scan things now, and, so forth and so on okay so we see high likelihood, of unemployment, low, and medium, here. Yes. It is yes. It is yes. Yes, it has. In. Some sense a lot has happened in some sense I'm not so sure the funny thing about artificial, intelligence. And. And you can speak, this maybe more than me as I said it's we've had artificial. Intelligence, as a field for about 50 years and lots, of excitement, this is all going to dramatically, change everything, and, for. 20, or 30 years there was almost no progress it was this theoretical, thing that computer scientists did and they couldn't, get much to do and then, they started making some progress you know they could make little robots that could roll around a room or something like that but. Are but what has. Happened in the past is artificial intelligence will make a leap, forward and then it'll stall and so. What I suspect, we're seeing although we won't know until there's a little bit of retrospection, so we made a leap forward with these neural. Network techniques and machine learning techniques, and now we're trying to absorb and figure out how to use them we're, not really refining those techniques at a rapid, rate right now although we are refining them so. I suspect we're at another stall, for a while but I could be wrong I'm basing, this a large part on the assumption, that would happen in the past is a pretty good guide, absent.
To Other kinds of information so. I'm not sure I'm. Not sure but in any case as. I said yes. Yes. Well. Descriptions. Of 700. Different types of occupations intending, to to. Cover the whole UK, labor market, Willie, remarked at a big modern economy. Correct. Correct. Yeah and it's basically focusing on the tasks, that have to be performed not the skills which. Is what you want for this because we want to automate those tasks, right yes. So. Here's the OECD, study which was done by economists, not AI researchers, so you know it's you just sort of have different perspectives so, thank you for that, perfect. Segue I'm glad we planned that ahead so. That's so you got a huge amount of attention so. Others. Started collecting their data and, trying to look at it with other methods so some, economists, at the OECD said now wait a second, you don't automate, all of a job usually what happens think. About the aircraft engineers, is you automate some parts of the job the, calculation, of blueprints but. Other parts, are difficult, to automate so, jobs evolve, think about a job as a bundle of different kinds of tasks and some things will be automate others won't and, when. They took that into perspective and, use the same data and methods, they came up with a much. More. Much. Less terrifying. Prediction. That's something like 10% of, jobs in advanced. Economies this is OECD. Economies. I don't think India's in here. Would. Be likely to it would be at high risk of automation soon, exactly as this gentleman said. Right. Ok, so I, think these artifice, thus first, study was too alarmist, it. Was a good study it was provocative got, people thinking about an interesting question now let me go back through history so, look at this picture this picture is the distribution, of the US labor market from 1850. To 2010. Last. Year I could get data for this and what. We see is in, 1850. The blue was. Agriculture, over half the US market, was farmers. That's. Probably India today, right right. Massively. On inefficient, farming in India from what I've observed driving, outside of Delhi is what we were discussing before, yep.
So. Automation. Already has occurred in places, like the u.s. in 2010. About 1 percent of our labor force is in farming, and as you can tell for me we're not starving, we, have a too much food we export a lot of food there's been massive improvements. In productivity there's, mechanization. This is the first replacement, of people by, machines. Labor. Service, and service jobs have been a kind. Of constant, this is skilled blue-collar but, what's really growing is these white-collar, jobs job where there's thinking. Creativity, cognition. Leadership. Teamwork. Collaboration, those. Are the kind of jobs that have grown over. Time and the, one thing we have never seen the United States is mass unemployment you. Know, despite all of these dramatic changes in the workforce and then, there's another thing 1/2 in the u.s. workforce where. We had introduction, of a new technology, that was able to perform. Higher. Order cognitive jobs, as well. As lower level jaw lower. Skilled, jobs as well, as those who are currently in labor market that's when women enter the labor market, you. Know my mother had a PhD in physics. No. And women, started dramatically, entering the labor market and we didn't see mass unemployment oh man somehow. The labor market, absorbed all of this new talent, and we. Were fine, what. Happens, is we. Find other uses for the talent that we have if, we. Can do something more efficiently, with something. Else or someone else or a machine we'll, do so but, that doesn't mean that someone's unemployed that means that the aircraft engineers, can focus on designing. More. Complex, airplanes, or ones that have a nicer. Look and feel or more comfortable, or more efficient, engines, or something along that line all. Right and we can also design new things like, computers, and projection. Screens and phones and smart phones and so forth so we build new industries and so demand, for. Labor can continue, to increase it just how it shifts, to different parts of the labor market, so. You can see where I'm going here I really don't think that, we're likely to see mass unemployment. Result, even of artificial intelligence even in 50 years we'll, see a lot of disruption, but history. Has told us that that wasn't the case, but. What I want to leave you with is. Basically. Some advice for those of you who have children and that, is how should you prepare for your children because I as I said I'm probably gonna be okay I got 10 15 years left maybe 20, if. You know. Depending. On how my 401k, retirement plan, does and my health holds out. But. You know what about children if you're saying my college or you're trying to figure out what should they focus on basically. What you should be saying is, anything, that can be automated eventually. Is going to be automated as costs. Of computers. And other kinds of Technology fall so. Focus on the things that are hard to automate and there's basically three of, those things, all. Right one is, cognitive. Skills so we're always told your kids. Should study stem, which. I think should actually be spelled steam. Statistics. Science. Technology. Economics, engineering. And. Math. Right. Those, are the things that help you. Problem-solve. Analyze. Diagnose, a problem come up with hypothesis. Develop. A way to experiment, to test your hypothesis. Analyze. Your data and so forth and so on, the, second, is social skills it turns out that social skills have also been increasing value in the labor market, in the last two. Decades it's, just something that, social. Scientists, had not realized, until recently, it's not just higher order cognitive thinking, skills but. Social skills and by. The way the most value people labor market are those who are high on social, skills and cognitive. Skills. My. Son. Is a computer, engineer and I'm, an economist, so those are two professions, which are high on that cognitive thinking, skills and I can tell you the, average person, in both, of those occupations is low on social skills finding. One who's high on both of those that's that, that's the key that's, a person's going to run HCl someday all. Right and, then, the last one which I didn't put on here is creativity. This is the one that's really hard to inculcate, into someone, but you, know you, know. Coming. At things from a surprising, angle, so you know reading some poetry taking, in heart class things like that as well as stem. Classes, and as well as joining. Clubs and and being, in sports and interacting, with other people those are kind of a portfolio, that is going to be valuable in labor more going, forward I think, I have ended three minutes early I didn't leave much time for Q&A but I'll stay here as long as you want.