tv Artificial Intelligence Automation and Jobs CSPAN September 3, 2017 12:51pm-1:49pm EDT
instead, many of us find information niches that reinforce our opinions. seemed polarization has to split us into two nations. >> watch this labor day on c-span and c-span.org, and listen on the free c-span radio app. now, a panel from the technology policy institute's annual aspen forum on whether artificial intelligence and automation are destroying or creating jobs. this is about an hour. >> ok. moving on to talk about artificial intelligence. not a day goes by where we do not see another story in the media about how artificial intelligence and automation either, one, are absolutely coming for our jobs, and you can see it already. just look at the relatively low and declining labor and rate.ipation
also, that it will allow for the creation of completely new jobs and industries, just as in previous revolutions. turny event, occupational is at an all-time low, so there's nothing to worry about. that is what this panel will talk about. our automation and aia substitute or koppelman to human to humancomplement labor? first, we have diane bailey, associate professor at the school of information at ut austin. her current research interests include remote occupational -- she conducts empirical studies, often involving multiple countries. james bessen is from the boston
university school of law. he is an economist. he does research on whether passion promotes innovation, how technology affects jobs, skills, and wages. his latest book is "learning by doing." at kamar is a researcher microsoft research. she works in several subfields of ai. she particularly looks at the real word applications. hal varian is chief economist at google. he has been involved in auction design, corporate strategy, public policy. professor at uc berkeley and three departments. a good place to start this panel is to talk about what we actually mean by artificial intelligence.
the preliminary discussions, ece brought this up. so i will ask you to lead us off and tell us where we are with a.i. ece: thanks. as a person studying ai and is not really involved in these policy discussions, it is a pleasure to be here and hear the discussion. a.i., artificial intelligence, is around 70 years old. it started with alan turing, when he wrote his paper on machines. then, he -- of the fathers of a.i. came together and wrote a paper that said this is what -- is. to they had a meeting where they discussed all the different applications of a.i. that we think about today.
the problem ended up being that it is more difficult than they ought it was going to be. out to bes turned easier, some things turned out to be harder. we are going through these ups and downs and winters and summers, where he realized that we are getting kind of better at this -- oh, no, i think this is still kind of hard. so what is a.i.? there are multiple definitions of a.i. one definition i like very much is the activity of making machines and intelligent. what we mean by intelligence is, for whatever the machine is divine best designed for, we expect the machine to react appropriately iss and the environment and reacting on it. that is what we mean by intelligent machine. but "intelligence" is a term we think of humans for.
humans are intelligent. other animals are not. a lot of abilities humans have we expect an a.i. system to have. so what is going on right now, why we think a.i. is so much in the press. i was a graduate student 10 years ago -- the advice i was given was if you are in the job market, do not say you are doing a.i. term that wasas a so kind of poisonous. people thought nothing ever comes out of a.i. companyle is an a.i. from the birth. what is really going on is there fivewo sectors in the last to 10 years that came together to make some of the algorithms we are ready had in the field very successful. for some tasks that humans are good at, like perception. data now coming
from sensors, coming from crowdsourcing, coming from human activities. conversation than ever before. that will increase as well. may things like deep numeral networks very successful. how to train existing algorithms. through this, now we are seeing great advances in perception tasks, like image recognition. so then, we see these tasks that are very associated with humans, like perceiving the world and where objects are, i just machines are really getting intelligent this time. scott: before we go into the labor aspect of it, a.i. has been around -- people have been working on a 70 years. it comes and goes in popularity.
is it the case that the last time it was popular, someone may have had a panel like this, where people were sitting, saying this is the time it is going to work? or is is there really something different this time? ece: that is the question we are here for, to discuss. however, when you look around -- there are, already, a lot of applications of a.i. people use every day, like search engines. was nott like a.i. there. athink if --a.i. was good some tasks that others were not good at. now, with new techniques, a.i. is inching into tasks a.i. was not good at by humans were good at. that is creating, first, a discussion about what other
tasks to sheehan will be able to do now that we have these techniques. second, what does this mean for the human job? that is something we need to see. however, there is another discussion, which is does this mean we are getting job loss because of intelligence? meaning i will have a.i. technique that will become so good that without much customization, i will be able to solve all a.i. tasks? i do not think we are getting there. scott: jim, you have been writing about labor and a.i. tell us your view. are we going to be jobless? [laughter] and is that good or bad? [laughter] james: not in the next 10 or 20 years. in the had a.i. workplace and marketplace since 1987. do. systems were used to fraud section in credit card
systems. but we had computer automation, some of which is not so we are seeing an acceleration in the scope of things that a.i. can handle, and maybe in the case in which it is addressing. it is about automation and the impact of automation. --perennially have a basic many people basically misunderstand what automation means for jobs. it is commonly assumed if a job -- if some tasks are automated, jobs are lost in the occupation. that is simply not true. we can look at manufacturing and we are well aware that lots of manufacturing jobs have been lost automation. in the 1940's, half a million textile workers in the u.s. now there are only 16,000.
most of that difference is from automation, something global trade. that has clearly had a dramatic effect on many communities, on many workers and their families. the thing to remember is automation can increase jobs. we only got to have 500,000 textile workers because for the previous hundred years before 1940 automation was accompanied by job growth. how can automation sometimes create more jobs and sometimes eliminate jobs? what is going on and what does that mean? it comes down to command -- demand. when you look at textile automation, in the beginning of the 19 century the average person at one set of clothing. automation met the price of cloth went down and people could afford more. demand was elastic so they bought more. they bought so much more that even though they needed fewer workers to produce the cloth, many were employed because they
were buying that much more cloth. in the 20th century people have closets full of clothing, further price decline is simply not going to produce much more demand for cloth. and the have automation net effect of eliminating jobs. what is happening with computer automation, we see lots of evidence of examples where computer automation, just like the early textile automation is creating jobs. one of my favorite examples is the bank teller. there is a great untapped demand, war there was a demand for getting cash at remote locations. the atm machine came along and people assumed it will wipe out the bank teller. we have more bank tellers since the atm was installed. the reason is it made it cheaper to operate a bank branch. thanks could open up more branches and serve more people. there was a market demand for
that. they built so many more branches that even though they needed fewer tellers per branch, they were employing many more. that is the pattern we are going to continue to see in many sectors of the economy. not in manufacturing over the next 10-20 years. that is what i think the immediate response is going to be to a.i. as well. diane, do you say accelerating change in a.i. will generate more jobs were quickly. that is counter to what others say. eric for example his book a couple of years ago said one of the differences is the rate of change is not allowing industries to catch up. by allowing the labor market to catch up. you are saying the opposite, right? the faster the changes, the better? james: yes and no.
a faster change will -- if you have elastic demand and you were making faster change so productivity improvements will ring job growth, if you make faster, you will have faster job growth. at least for the period of time that is occurring. it will be disruptive. i was may be overly optimistic in the way i described things. when the question is, are we going to be seeing mass unemployment, the answer is no. are we going to see individual jobs destroyed? yes, but others will be created. as that accelerates that is disruptive. the textile workers in north carolina need to find jobs, need to have skills. there are jobs opening up in the rest of the economy. and the same thing everywhere else.
we will continue to see jobs eliminated. the acceleration will put more stress on our ability to transition people, to retrain them, to relocate them. scott: diane, i know you were much more less optimistic. diane: i would just ask this. book04, a guy read a called "the job-training charade." what he talked about was that we were losing a lot of manufacturing jobs. the way we talk about that is these people need to be retrained in other jobs. we will train them to be acres and -- bakers. there is only so much bread and pastries we can eat and there were not enough of these other jobs are people to take. the language of job retraining started to change from teaching people new technical skills to enter different jobs, to
focusing on what they call soft skills. people started to be told the reason you don't have a job is because you are lacking technical skills, for your communication skills are not very good. you don't work well on a team. it started to put -- on workers for their lack of soft skills rather than recognizing we had a structural shift in the economy and what kind of jobs were available. i was asked about two weeks ago to sit on a national academies of engineering panel to talk about engineering workforce and how they needed to be more adaptive to survive in this new economy we are going to be seeing. i work at a large public institution, the university of texas. getting an invite to sit on the national academy of engineering panel is a big deal. that means i can put it on my tv in my get a 3% raise set of my 2% raise. i have some skin in the game. i turned it down.
i turned it down because i told them i do not believe in the premise of the panel. they thought they were putting the onus on engineers to become more adaptable, be a quick learner. they are telling all of us these things instead of saying we'll start seeing fundamental shifts in the economy and maybe we ought to start planning for that. as all of us, not just us individuals becoming more adaptive and quicker learners and moving up the scale. there is only so much room at the top. if what we think about a.i. might be true, there will be only so many jobs and not all of us are quick to go up the ladder. that worries me, what will be left for everyone. i think there are a lot of problems with job training. we have issues with geographic
relocation. you see a lot of jobs appearing is not really see unemployment. you also have a great difficulty. my book is called "learning by doing." a lot of new technology related skills have to be learned on the job. it is not a matter of the classroom entirely. we have to come up with new ways of getting people experience. but i think we do see plenty of sectors where there are knit skilled jobs emerging -- mi d-skilled jobs emerging. nursing jobs have been in great demand for a long time. yet it is often very difficult for people to transition into those. -- we don't't have understand what is involved in making all these transitions. scott: you are bringing in the longer term talk. i want to say a word about
this job-training issue. jobs are here and the skills are here, you bring the skills of to the demand or you bring the job down. there is a lot of that going on through technology because it used to be to be a cashier you had to know how to make change. to be a taxi driver he had to know how to navigate around town. not necessary anymore. to be a veterinarian you had to identify one of 250 breeds of dogs. not necessary anymore. you can do that with a.i. or your phone for that matter. this cognitive assist is a big deal because it allows the on-the-job training you are talking about. you drive around town and learn your way. you learn how to make change because the machine tells you. the breeds of dogs on the phone. there are a lot of delivery mechanisms that are extremely efficient in the on-the-job delivery of education. look at you to. there are 500 million video views per day of how-to videos on youtube.
i will that you almost everyone in this audience has asked something in her house by going and looking at a youtube video. these are not just high-level cognitive skills like solving quadratic equations. they are important, manual labor skills that people can learn how to weld, how to replace a screen door, how to hang the window. scott: if i were to play devil's advocate, that is a lot to prepare people who were not called in to get work. -- repair people who did not get called in to do work. hal: when you look at the next part of my talk we talk about what happens to them. i want to talk about the theme that released to this discussion. we talk about the demand for label and the areas -- the theory is a.i. will reduce the demand for labor. on the other hand, if you look at the supply of labor, we can find a different story because
that is only one social science they can predict 10 years ahead. that is tomography. everything kind of pales besides that. 1946, that is when the baby boom started. 1946 to 1964. after that, there was a baby bust. then there was the echo of the baby boom. you can look through this whole series of population changes and add 65 years to it and we see what is happening now. when all those baby boomers are retiring, that is followed by the baby bust. what does that mean? the labor force is growing at half the rate of the population. 20202'sade of the you will see the lowest growth in the labor force since world war ii. if look at the labor force, you restrict immigration, it is actually going to decline. all those baby boomers are retiring. they expect to continue
consuming. you need some workers somewhere to be producing this stuff they need to consume. you have this race going on between automation, which is increasing productivity, and you have the supply of labor which is very, very low to decline. we have a good in the u.s. , japan, korea,a germany, italy. they are seeing outright declines in the labor force. from very, very worrisome the point of view of the future of their economies. look at robots. what countries of the most investment in robots? china, japan, korea, germany, italy. they have to have those robots. they have to have some improved efficiency and productivity to produce the stuff their population will be demanding. that is true is a worldwide phenomenon. unless there is
a really big surprises on the, you will see -- surprises on the automation side, you will see this in the next 25 years. scott: how far down the line you see that? hal: 20 look at the figures around 2060, bc the labor force growing at the same rate as the population. it is interesting to think this is all because of this huge shock of world war ii. it created this gigantic demographic event that does not work itself out for 100 years. scott: it is the baby boomers fault? hal: of course. scott: it seems like so far there are four issues. what is a general short-term versus long-term. is there anything you can do for people who might be -- i'm not sure what the right word is -- displaced in the short run?
does job-training play a role when we learned about the effectiveness of that? the distribution affects, short-term and long-term. and over time the demographic and demand for labor, which will swamp everything. diane -- and the inequality issue, whether it benefits just accrue to a small group. diane, you turned down this position at the academy. their projectink should have focused on to address her concerns? -- your concerns? diane: we should be paying attention to power dynamics. you hear a lot if you read about books on a.i. and predictions about jobs, look at the bureau of labor statistics data that describes jobs. describes the tasks and jobs. based on that description we will tell you some percentage of
jobs are going to be automated or replaced by a.i. within some period of time. they are doing that based on description of what people do. no job is just what you do. every job takes place in an environment that is surrounded by, for example, all kinds of occupational norms and perhaps regulations. i spent a decade studying how engineers are using new computational techniques and software. things like finite element analysis. i wanted to understand how it was changing design and analysis in that field and what it was doing to the workforce. i will give you two quick examples that point out issues of power and when workers have power and how they control technology choices made for them and when they are not in power. it people who chose power, is power held by the government
somewhat. if you look at civil engineers to design building structures like the one we are in, their solutions are governed by a strict law and regulation that involves things like peer-reviewed and accounting review of plans for building. because of this building were to fall down, those of us who survived would sue. is theson responsible senior engineer who put his professional stamp on the drawings for this building. and because that person faces professional liability, they are very careful about using automation in their work because they know computer software programs can yield unrealistic unfeasibleased on assumptions. for civil engineers everything is about those assumptions of how a load travel through buildings. that guides designs.
they rejected a lot of automation. it is not like they use the techniques, but they don't use any automation between the steps of engineering design and analysis that we see in other fields. now i got automotive engineers. scott: are you portraying that as a positive thing about engineering? does that necessarily make us safer or less safe? diane: i think it makes as much safer. i would hesitate to write in an elevator that a computer had designed, and all the civil engineers i watched would tell you the same thing. it is because you have to watch -- i can go to countless examples. i study things at the micro level. i spent hours sitting at the elbow of engineers while they are designing. i talked to them. the computer crashes, we have time for a short interview. i asked why did they do this
thing? at the automotive firm they had a secret laptop with secret software. they were not supposed to use it anymore. they had it locked in a desk. -- they werey restricted in the software programs they were supposed to use. i asked them, why do you use this technology? i think it's a good thing that the legislation was there. --il engineers took safety they did not take it as a harness. they took it as an ethical obligation and something they were proud of. something that made them different. automotive engineers do not have the same kind of things guiding their work. yes, they have the national butsportation safety board, there is no stamp that is to be put on the vehicles. they just have to pass government test. their work has been rationalized.
their work has been digitized. their work is been computerized in ways changing what is happening with the workforce. very quickly i will just say the engineers i studied since about 2004 have a hiring freeze on analysis engineers in the u.s. they only hire analysis engineers, simulation engineers in india. they have a big center. my team spent months at that center and months in michigan. at the center they have offshored the work of building the models. the reason they were able to offshore that work was because we had digitized the models. we digitized and magnetize. ize. me -- mathmeth people in india can do the work for a fraction of the cost of a u.s. engineer.
my point about adaptable. how can u.s. engineers who wanted to do that work be adapted in the u.s. when it would happen with the india? you can't do that job anymore in the u.s. it came about because those engineers did not have the same type of power to control what they do. radiologist -- yes what. all your medical scans can be read abroad for reading and they are. they are often sent at night in your radiologist does not look to be woken up. they send it to india. in india a radiologist will examine it and it comes back to the u.s. the ara, their professional association lobby for legislation that sent those scans have to be signed up in the morning by a board certified u.s. trained radiologists. scott: hal, a response to that? --: i think the radiology not only can be done in india,
but now can be done automatically. that is true not just recently, but it has been true for a decade or two. recognizingcases the malignant cell is really pretty straightforward. it can be done by even relatively untrained labor. there are border cases and lots of things where you might want have some of the supervision you were describing, but it can be turned into basically exercising exclusionary power to keep a privileged position. i think that is true. i have often said we would have driverless vehicles, autonomous vehicles on the right now if it were not for the darn human drivers. not to mention the pedestrians that are even worse. you have a controlled environment like a freeway, and expressway, a controlled
environment is really possible to have autonomous vehicles right now. it was possible to have autonomous vehicles at least a decade ago. it is dealing with all the exception handling that is the problem in many cases. scott: feel free to jump in. ,ce: k discussions about a.i. there is discussion about super intelligence or taking jobs away. i worry about something more short-term than that, which is as applications enter society these a.i. algorithms have their problems too. how are they going to be handling the shortcomings of a.i. will be the determinants of halley get value out of these value -- how we get value out of this technology. there is a new excitement about a.i. however that excitement comes with the downside.
they are hard for people to understand. with thetes very much comments about, if i know when my algorithm is doing the wrong thing,-override it. statistical techniques when the algorithms are learning from large amounts of data, it is quite impossible to understand what the algorithms are going to be doing for each case. the algorithms get updated pretty often. for example, tesla car. there are questions about how they will be driving if they change every week. as the driver you were expected to understand. there is a transfer problem between the ai algorithm and the user or the controller or the supervisor of the algorithm. you need to do a lot of work to make that later transparent in the sense we can actually create a transfer a partnership between
the human and the machine so they can work together. so we can get to that case where, ok, i know might algorithm is not doing the right thing. i should override. one of the cases where a.i. algorithms have been used in public space is decisions that happen in the legal system. for a decade now statistical learning algorithms have been employed in courts for making sentencing an important decisions. -- and important decisions. there was an article that discusses issues by having been working with judges. that's a great thing for all of us to understand, the social issues that come through using a.i. if you're a judge under time pressure and need to make these decisions, wouldn't it be easier
to just agree with the machine decision. what you gain by overwriting it? if you override and a person commits a crime, you are at fault. you need to really think hard about the balance of opinions and he was responsible for these decisions when you actually have a team of a.i. engineers making the rest -- a.i. algorithms making the decision. algorithms can be quite biased. they were analyzing statistical techniques for making decisions for african-americans and whites in society. scott: sorry to interrupt. these are huge issues. let's try to stay more narrowly focused. that theme, how in your work do you try to bring --
try to make humans and a.i. complementary, whether in labor or understanding? ece: we need to move away from the thought that a.i. will automate and humans will adapt. that is not likely to happen. is wee need to work on a are going to be working on automation for getting these capabilities to a.i., only have to think hard about the design of the thing in the middle later in terms of the a.i. explaining itself of the human, humans having transparency with the machine, and working on the coordination. that's the only people get to a situation we like. think about the exceptions in self driving vehicles. black swan events are hard to
predict. you need this huge amount of data to come up with a reliable estimate about how safe the vehicle is. you think about self driving cars. n enormousa amount of proprietary data. but the insurance companies or government regulars with the public in general don't have access. it is hard to understand what the actual risk is without some kind of data sharing. new set of issues emerging in terms of data transparency. hal: i think you are exactly right. they will be the man's for interoperability and data sharing for safety reasons. look at the airline industry as an example. when it is an airline event there is immediately an investigation by several different parties with constituencies and interests and they try to resolve the cause and make sure it doesn't happen again. i think that same set of procedures will be carried over
into this context without much objection. who is going to stand up to object to that kind of investigation? one thing about the news, when you look at the journalism a.i. they were picking the interesting cases. if you want to get an idea of what is really done in the ordinary sorts of cases, go look at tangle. that's a company that sets of machine learning competitions. one company might say we will offer a prize of $1 million to the party best able to predict hospital readmission within 90 days using this data set. two qualifications. and itn angel investor was recently acquired by google. nothing to do with each other, those two facts.
it is interesting because you see things like housing appraisal. ow put up a data set of housing features and values and put up the best model for pricing houses. put up 4.5 million videos and has a contest to predict what people are doing those videos. even with still images we have good technology for that now. we don't have the technology for videos, for people moving around. are the exercising, dancing, fighting, whatever. readmission to hospital example i mentioned, recognizing leaves, counting plankton. there is 230 something applications to get an idea of what is going on in ordinary business practice. it is quite interesting. i think we will see this diffusing. there will be the really exciting cases like a driverless
beinges and the humans the go champions. there is a lot of automation going on as well. ece: a.i. has a long-term problem. this comes to your point we use a there are a lot of cases. in the real world, a.i. has to do with a lot of edge cases. if you look at the cases were you have a lot of data, the distribution of the data, for some cases we have a lot of data. the common techniques we use, the statistical learning techniques are good to learn part of the distribution. we really need to think hard about how to get it right. it could be for collection of edge cases, creating data sets cases, that we
need to think about some kind of collection of techniques working together. not only sticking to one technique, but a collection working together. and other human supervision to get those cases right. 95% accuracy on the data set. that is not necessarily mean the application will be providing value to you and the short-term. that cap see your question about why we had nothing productivity from a.i. yet. when you think about those, you need to think about, what is the point where i'm going to be able to get value from this technology? that is a different question. scott: i want to go to questions. return -- oneto thing to bear in mind is to pay attention to the rhetoric used around a.i. on the the media focuses
fascinating cases. but i also think the tech companies without a lot of rhetoric other on. self driving vehicles, private all of you can tell me how many motor vehicle deaths per year we expect to save. 1.2 million lives. you might not note that as the number of motor vehicle deaths per year worldwide. the number in the u.s. -- we don't care about the rest of the world, the number in the u.s. is 35,000 deaths are year. to let you know what the number said i was situated between what is right below it and right above it. right below it, 5% lower, deaths by falling down. right above it, 25% higher, death by poisoning. solutions for.i. poisoning.n for motor vehicle deaths reached a
peak in the 1970's and we have been decreasing ever sent. why? mechanical and electrical improvement in vehicles that have been on grind and on regulations. things like dui laws and sobriety checkpoints in mandatory seatbelt. yourself why are -- therefore so interested in self driving vehicles? the thing about sitting lines is not really it. that problem is solving itself. if you look at commercial truck drivers, the first group -- from: the number of deaths automobiles is going down, but that is not me self driving cars but further reduce that. diane: they might help to reduce the number, but the number we are talking about realistically is 35,000 in this country. then you look at other countries. university of michigan's
transportation institute put out a map of the 25 countries with the most motor vehicle deaths per year. if you look at that map, you get a sense of what the roads in these countries look like. there is no way a cousin things up, they areght chaotic roads. and india there is no way you will have self-driving car is in india unless the chairman and suzuki says they will have it because there is chaos on the roads. we have other ways we can bring these numbers down. i don't think the real motivation of why we are going after self-driving vehicles is to reduce deaths. i think there is another motivation and i would like to have a conversation. we could really redesign our cities and improve and and some climate change problems. would not that be wonderful?
let's have it conversation. say if you look at the examples you gave, they have this common theme. advances in modern. people can fall over, be saw a pilland i just bottle that warns you about what you are taking and whether you already took this this morning and all sorts of other things. i don't know if i would call it a.i., but i would say technology. technological advances in one form or another that help people live safer, more productive lives. about selfis issue driving cars, the problem is dealing with all these exceptions. i asked the team if you break for squirrels? that is a decision. avoiding a squirrel could cause
lots of damage elsewhere. they say no. do you break for dogs? they say how big is the dog? [laughter] the break for deer? of course because there are huge number of accidents involving deer. toneed an automated deer avoid the cars in the will have a safer road. [laughter] scott: we have a little time left. larry? >> i was a cheerleader in high school. is it on? topic,ome back to the the demand for labor. back in my graduate school winner nobel prize threw out to the class the horse. october 1 story -- or horse
story. before 1940 the population of forces and gone way down. why? because of the internal combustion engine. ess vehiclesl rather than personless vehicles for personal transportation outside of urban areas, , hauling freight in urban areas. cars, truck's, tractors. did not totally reduce or eliminate all horses. there is still a small population left, but clearly the market clearing wage for horses went way down. low reproduction sustaining levels. fortunately horses have shorter lives. fortunately we feel differently
about how to deal with surplus force -- horses. hal, iconfident, and think your point about how labor issue may modify, but the -- anddemand for labor arguably we have been so lucky over the past two centuries that technology shifts have not really diminished the demand for labor. if anything it pushed the demand for labor, but we can't relet the possibility of a technology what theing to humans technology change between 1900 and 1940 digital horses. scott: what happens if we are horses or the population have not decreased? >> right. that is the market clearing wage
falling way low. and then what? >> it is difficult to retrain horses. [laughter] james: it is not a direct comparison. we can look at occupations that have largely gone away. we have taken those people and absorbed them and other capacities. that is generally true. why has it been true? if you look at individual occupations, it is a story about the man. there is job growth, but when demand starts getting inelastic, further automation least the job declines. they can be dramatic. in 10 or 20 year timeframe, we will not see a dramatic change in the nature of demand.
but it is a good question further out. are we going to basically be able to satiate demand in one market after another so there are no jobs left? that raises a broader philosophical question about what is it that humans want. keynes-- john maynard by thisti talked about time we were going to have huge amount of leisure time and technological unemployment. if you look at what is happened to leisure time, yes, the work week has declined fairly dramatically since the 1800s when you have a 72 hour work week. now we are down to 34 hours. it is slow. you have to ask because some level people are getting value. things theyand for will either consume or demand
for how they obtain their leisure. the technology of leisure has dramatically increased in terms of video games, movies, and all these various things. the question comes down to a fifth -- in 50 years, will the be anything humans want their machines can't deliver? -- we are at a philosophical level. do humans what interactions with other humans? do humans want -- of the personal role an important aspect of what humans want, both in terms of what they are doing for gainful employment and in terms of how they want to consume? i don't think i would gain to answer that -- dain to answer that, but i can point a very bright people.
they had a hard time imagining all the additional things humans might want that might cause them not to have a 10 hour work week. hal: i just wanted to say a couple of problems on that. one is 80% of the u.s. economy is service. people want to be involved with other people. there are lots of services you cannot make pretty easy. i don't really need it person to lead me to my table at a restaurant. you can flash little arrows on the floor to make it happen. but we want that. people want those services. the work week, nothing is written in stone about a five-day work week. mexico, the work week in mexico is 45 hours. in the u.s., 37 hours. in the netherlands, 30 hours. france is lightly behind at 33. you can take some of that increased productivity and leisure. no question about that. you would get very few objections if people said, hey,
let that every weekend the three-day weekend. which is pretty much what is technologically possible today. that could easily happen if people chose to move in that direction. one last word about the first invasion of the robots. the first invasion was around the 1880's to 1910 or so. washing machines, dryers, dishwashers, lawnmowers, sewing machines. all those mechanical innovations that made homework far more productive. particularen in shifting from the household into the labor market. and you see this tremendous increase in output from household basis because of that kind of innovation and automation. >> i introduced myself as an economist historian.
and besides the secret handshake we are supposed to tell people about how things were the same in the past. i don't think a.i. is distinct arrow,y the bow and which is a substitute for spears or the horse, which is a messengers on foot. of course these technologies have implications. the horse was an instrument of because theycy were expensive to maintain for a while.
whole thesehe robots, which is what a shovel is. robot.l is a viewed withto be optimism. which ought tor be in everyone's mind. it is not true the number of jobs created or lost in the united states is what is reported every month. about 200,000 jobs when things are good. minus 200,000 jobs when things are bad. jobs arer 20 million lost in the united states. workforce of 150 million.
there is a tremendous amount of turning. in 2000, 130,000 people employed in video stores. of goodou all to be cheer. [laughter] scott: probably have time for one more question. that comes back to something ece said at the beginning. we are moving towards a general artificial intelligence. it is a general-purpose technology or a shovel? there is a huge difference. they have argued that if we look back at other technological innovations like the mccormick reaper, it transformed the culture and through a lot of people out of
work. a because we do not have simultaneous innovations going on in other industries, there were places for them to go pretty quickly. his argument is that what we will see with a.i. is simultaneous innovations across all industries so there is never left to run. would you say you don't agree with that idea? >> no, and i would like to know how -- diane: howdy knows they were happening across -- i think he is looking at the present. >> had his you know the present [indiscernible] [laughter] it's a highly unpredictable thing. knowing of the invention -- no one knew the invention of gunpowder in china would thecally change aristocracy.
up -- we needrap to wrap up. we will take a short break, 10 minutes and start the next panel. if the next panel goes over we will miss the eclipse and they will let anyone out to see it. join me in thanking the panel. [applause] [crowd noise] >> from the same event, california u.s. congress and darrell essa, david young, and other technology company officials discussed the