February 08, 2015
McAfee on automation: what will we need all the people for?
A Technology Review piece about the research that MIT Sloan School of Management profs Andrew McAfee and Erik Brynjolfsson do into how computer automation is destroying jobs at a high rate includes this line McAfee "when all these science-fiction technologies are deployed, what will we need all the people for?”
McAfee, associate director of the MIT Center for Digital Business at the Sloan School of Management, speaks rapidly and with a certain awe as he describes advances such as Google’s driverless car. Still, despite his obvious enthusiasm for the technologies, he doesn’t see the recently vanished jobs coming back. The pressure on employment and the resulting inequality will only get worse, he suggests, as digital technologies—fueled with “enough computing power, data, and geeks”—continue their exponential advances over the next several decades. “I would like to be wrong,” he says, “but when all these science-fiction technologies are deployed, what will we need all the people for?”
I'm in agreement with Brynjolfsson and McAfee about the impact that automation is having on employment and wages. I look at my daily experience, most notably in terms of dealing with people in order to deal with organizations. In a nutshell: very few of the commercial transactions I engage in (buying things, paying for services, signing up for services) involve interacting with humans. In the vast majority of cases I deal with computer systems. Click to choose. Click to order. Click to pay a bill or click to set up automated future bill payment and automated future delivery. Interacting with humans? Not so much. Less than once a week and dropping.
Once autonomous cars can make deliveries with robots to bring the goods to your doorstep where will humans show up in the chain? The warehouses will go lights out eventually. Long haul trucks will be autonomous. Robots will load and unload the trucks. Farms? Autonomous planters and harvesters. Factories? Fully automated.
As I argue in my recent post Employment-Population Ratio By Education Level the science fiction future is already happening right now. The official unemployment rate radically understates the extent of real unemployment.
Jim Clifton, the chairman and ECO of the Gallup polling organization, says the real employment rate is much lower than thought and he's got polling numbers to prove it.
Yet another figure of importance that doesn't get much press: those working part time but wanting full-time work. If you have a degree in chemistry or math and are working 10 hours part time because it is all you can find -- in other words, you are severely underemployed -- the government doesn't count you in the 5.6%. Few Americans know this.
There's no other way to say this. The official unemployment rate, which cruelly overlooks the suffering of the long-term and often permanently unemployed as well as the depressingly underemployed, amounts to a Big Lie.
Take a hard look at your job and skill set. Do you need to learn new skills to put you into a job that will last longer?
Update: The most common job in most US states is truck driver. Almost all those jobs will disappear in 20 years as autonomous vehicles replace human truck drivers. In 1978 secretary was the most common job in 21 states. Computers radically slashed the need for managers to have secretaries. Secretary is now the most common job in only 5 states. In 1978 machine operator was the most common job in 9 states. Factory automation devastated that position. It machine operator does not show up asa most common job in any state.
Software developer is the most common job in 4 states. That does not bode well for the blue collar workers of today. Few truck drivers will be able to make the transition to software developer. What will the laid off truck drivers do? What will be the big jobs of the future for blue collar workers? To put it another way: Is there much of a future for the lower classes to serve the upper classes? Or are the lower classes going to mostly work for each other? Will a parallel economy of the lower classes emerge?
Randall Parker, 2015 February 08 11:37 AM
Once autonomous cars can make deliveries with robots to bring the goods to your doorstep where will humans show up in the chain? The warehouses will go lights out eventually. Long haul trucks will be autonomous. Robots will load and unload the trucks. Farms? Autonomous planters and harvesters. Factories? Fully automated.
I'm the guy who does the automating. I can you that its going to be while before we get to what you describe above. Humans have dexterity and hand-eye coordination that are still difficult to duplicate in robotics. I can also tell you that their are machine shop jobs (welding, machining, etc.) that won't be automated for quite some time, and that the boomers doing these jobs are retiring. Automation demands increase numbers of electricians and mechanical fitters to install the systems and technicians to maintain them once they are started up and in production. One engineering job (that's me) in this field requires about 3-4 technician and electrician-level jobs in order to do the work.
You are correct about the lights out factories and probably farms being automated. Making an automated planter and harvester is much easier than an automated truck or car. Long haul trucks will be automated on the freeways, but will likely have to be driven by a human from the freeway off-ramp to the warehouse on the surface streets (I think automated freeway driving will come much sooner than compete driver-less cars on surface streets). Driverless delivery vehicles to deliver to your doorstep are a long way off. Local truck driving jobs are likely safe for the next couple of decades, I think. Warehouses will be fully automated when we make automated (driver-less) forklifts (which is something I would like to do personally, but is not an option on this current gig I'm doing).
Automation is incremental in nature.
The export of jobs to lower paying countries, labor arbitrage, is also contributing to joblessness in the US. If the foreign trade deficit were balanced, there would be a lot less jobs exported and this would mitigate the effects of automation. In particular, if the national impoverishment in the form of accumulating trade deficit were stopped, there would be a lot more resources to take care of those who lose their jobs to robots.
But there is still a ray of hope: one side-effect of automation is that it will uptimately make the labor arbitrage that exports jobs to foreign countries essentially unnecessary, and this would mean the elimination of the trade deficit. In this case, more taxes will be collected in the US, and this will make it possible to take care of the unemployed. The current state of labor arbitrage that exports jobs abroad, is continually decimating the middle class.
My take on problems like hand-eye coordination is that other areas have made faster progress than expected. 10 years ago self-driving cars seemed to be a lot further away than they turned out to be. Not commercialized yet but working remarkably well. Progress on many other ML problems has been faster than expected. Lots of money and start-ups going after robotics too. I think we'll see discontinuities on the rate of progress with surprises.
Agreed on driver-less vehicles on freeway before surface streets. Highways are much simpler environments. But the gap between highway and warehouse could be bridged for autonomous vehicles by putting warehouses along freeways and giving them dedicated off-ramps to autonomous-only zones to pull up to warehouses. The highway-side of the warehouses would have autonomous vehicles parked while the city-side of the warehouses could have human-operated vehicles parked.
Wal-Marts could be built near highways with the same relationship as well.
For transportation on the large if we really want to automate I think we should think from first principles . The aim should not be to replicate what a truck driver does on the truck and on the road and automate it , but it should be to transport large heavy goods over long distance safely and cheaply. Then we can get rid of the trucks and roads and other costly infrastucture meant for human drivers and come up with something new . Maybe we can have something like an overhead line hawling goods or something like a simplified railway but fully mechanised and only for goods. It will be much simpler and easier and easier to build operate and maintain. I think they already do this in many mines ,factories, plants etc. We can construct them
along side by side rivers or canals or oil, gas, water pipelines .
We are many years into this process right now and we can already see the dearth of what we used to call good jobs for the low skilled. 30 years from now, there may be few jobs available for the left hand side of the Bell Curve. What are those people going to do?
One of the major roles of a truck driver is to be the temporary caretaker of the products they carry. The contents of a truck might be worth $50K or $100K or even $500K.
If you watch the reality show Big Rig Bounty Hunters, you will realize that truck theft is already big business as things are. With no driver present, theft would be trivial. Just put an object in front of the truck and cart off the contents. It would be incredibly safe. No dealing with a pair of angry 200 pound dudes. And what's more, you could choose remote sections of highway, far from the police. Just put a houseplant in the road and the automated truck will sit there, permanently stuck.
Therefore you will still need an operator. An operator would flatten or drive around an object, report you to the cops, and perhaps come at you with a wrecking bar.
Aside from theft, there will be situations such as a construction worker directing traffic onto the opposite lane or a detour around a broken car or a board that fell off the back of a vehicle into the road that will leave an automated truck stuck while a driver could handle this trivially.
Also, there is plenty of go/yield face and hand communication with other drivers and pedestrians that is pretty key. And what about flat tires and mechanical breakdowns? These drivers can not only fix a bunch of things but judge when to ignore problems or apply a temporary fix. A given big rig probably has several minor mechanical issues at any given time. If it must stop for every little thing crossing the country coast to coast, it will be a long trip indeed.
An even better way to rob an automated big rig company might be to cause an accident with one. Let's put the poor human being against a software program in the courtroom, with an all human jury. C'mon Google, 500 mil is nothing to you, you earn that in a week. That poor guy in the neck brace may never be the same again.
Shopping for a jury for a human driver - automated car accident case would be easy. As in, every single jurisdiction in the entire country save for the Bay area. (It's unfair, really. That automated driver A/I should really be entitled to a jury of its peers. Social Justice Warriers, what's the hold up? #RobotLivesMatter)
To my knowledge Google has never let their automated car go around without a person inside, probably because their lawyers have a smidgen of sense.
In this article the entrepreneur Peter Thiel says that the threat to middle class jobs in the US is not technology but globalization that exports jobs to foreign countries. Peter Thiel also points out that the industrial revolution in England actually raised the standard of living without causing unemployment, even though production went up dramatically:
I would add to the latter claim the fact that robotization would actually stop the export of jobs to other countries!
Peter Thiel touched upon it, but didn't fully flesh it out in that article. Technology wont be the cause of the loss of the Middle Class. Technology increases productivity. Agriculture automation makes food more plentiful, people already have an overload of choice when it comes to entertainmemt, computer productivity tools. Elon Musks success with Tesla has provided a path towards less dependence on fossil fuels. New initiatives around LFTR's could cause an energy revolution. Think Peter Diamandis' Abundance book. All things being equal technology extends the purchasing power of individuals and lifts the poor into middle class lifestyles. It creates more of the Middle Class. True it kills jobs, but I'm sure most people would rather work 20 hours a week to maintain a middle class lifestyle and build savings than work 60 hours a week.
What is killing the Middle Class is 1) global monetary authorities systemically eroding purchasing power through manufactured monetary inflation and inflating asset bubbles and 2) oligarchs jacking up costs on specific industries like Health Care and Education. Why is the Fed so fearful of deflation? Its good for technology... why not for other goods and services?
My take on problems like hand-eye coordination is that other areas have made faster progress than expected. 10 years ago self-driving cars seemed to be a lot further away than they turned out to be. Not commercialized yet but working remarkably well. Progress on many other ML problems has been faster than expected.
Much of this is based on "deep learning" algorithms. Deep learning algorithms are essentially neuro-nets with 6-10 layers rather than the traditional 2 layers. They are rather difficult to "train". At least I've heard that this is the core technology of all of these breakthroughs. Details are sparse (intentionally so??), but I think Boston Dynamics' "big dog" and other robots are based on this technology as well.
In any case, I'll be more convinced of this stuff when it starts showing up in the controllers in the automation industry.
What do we need all the people for? Dude! They are the customer base! Who else is going to buy all the crap being hauled around by the automated long-haul trucks? Gotta solve the meaningful employment problem along with the automation problem. There is no need for automation if the customer base for whiz-bang products goes away.
My take on problems like hand-eye coordination is that other areas have made faster progress than expected. 10 years ago self-driving cars seemed to be a lot further away than they turned out to be.
I'm a big fan of the idea of self-driving technology, but it's looking like self-driving cars are actually a lot farther away than they currently seem to be. In the real world they do pretty well, but only with roads that have been precisely and exhaustively characterized in terms of everything from radar to lidar and that only until somebody moves a traffic cone or something. They're still very bad at handling anything unexpected. Will this difficulty last forever? Probably not, but right now they might actually still be pretty far out.
The thing that drives me crazy is that automation is often a BAD thing. Take the example of computers and secretaries.
Used to be that managers and executives and lawyers had secretaries. They would dictate a memo, or letter, or contract, or brief, or whatever, and the secretary would then type it at 90+ words per minute.
Now the executive, or manager, or whoever, has a COMPUTER and MS WORD! He doesn't NEED a secretary any more! So now the manager is typing his OWN letters - at 15-20 words per minute.
The secretary was paid perhaps 10-15 dollars per hour. (Probably not that much, but we'll account for inflation.) The manager gets 4-5X that much, but when burdened with typing tasks, is far LESS efficient.
So long as the traveling infrastructure is shared cost construction the implementation of a new system will not be done. Your tax dollar at work, so to speak. Besides there is already a system that could be automated relatively easily -- existing rail network. Surprising that Amazon does not see this opportunity.
The other factor is this. The first drive will be to automate out the driver to eliminate his labor costs. $5k of compute power can probably do it once all the sensors and controls are validated. The next will be to modify the tractor itself. Make the cab smaller, only needed for the short haul distance to the depot. Then modify the trailer since the tractor won't be in the way (Think the tractor now sits under the trailer)
What might come next would be some way to pick up electrical power ala a tram like system. Mercedes and Volvo are already testing this idea.
Re Dan at 2/11 12:27pm:
The first trucks to be automated will only travel on interstates. Local driving will remain human for a longer time, partly for the reasons you describe. Also, it'll be awhile before a computer can reliably back a rig up to the right loading dock at the correct location.
Second, trucks will all have drivers. It's just that the drivers won't be riding on the trucks. But they will instantly know when something goes wrong, and will have the resources to send help quickly. So no, the truck will not be sitting on the road indefinitely because somebody put a flower pot in front of it.
(Each driver will be operating a couple dozen trucks at a time, so there will still be a huge labor savings.)
Software developer is the most common job in 4 states. That does not bode well for the blue collar workers of today.
It also doesn't bode well for software developers. A lot of hard work will be put into automating their jobs.
Technology will do what technology has always done: make humanity wealthier as a whole, but widen the gap between good and excellent. E.g. with the technology to record and transmit sound, instead of watching some guy play a poorly-tuned piano at the local saloon, we could bring the world's best pianists into our homes at a fraction of the cost.
More technology = more inequality. Nations that accept this outcome as a fact of nature (e.g. Singapore, Switzerland) will prosper, while nations that try to make people equal (e.g. Cuba, Venezuela) will become starving, pestilent slums visited only by sex tourists, and even they will stay away when sex robots get good enough.
"Will a parallel economy of the lower classes emerge?"
Not a legal one, unless regulatory capture and crony capitalism are both knee-capped by radical libertarianism. OTOH, the more people who are imprisoned, the more people they can ignore.
Mike Street Station,
The left hand side of the Bell Curve will be unemployed.
As Dan King points out, the automated trucks will start out on major interstates where theft is less of a risk. But I can think of many technological ways to reduce the risk of theft. Here are some:
- Have each truck carry a UAV on the roof. The moment it stops the UAV takes off and watches what the robbers do and follow stolen goods. Other UAVs take off to join the hunt. Images get sent to cops in real time. Cops converge.
- Put defensive systems on the truck, active ones. Tasers for example.
- Give the truck a remote control system. A remote human watching all the video feeds can take over and move the truck.
- Move the trucks in packs. They all stop in a group and defend each other. Their UAVs work together. Cops can take over a UAV (which carries payloads they can activate) and use it to sprinkle objects on the road that pop tires to stop the bad guys.
I figure automation will make theft less likely, not more likely. Bigger UAVs will move cops to scenes of crime progress much faster than can be done today. It will become harder to hijack a payload.
The customer base are the people who have money to spend. If you are a supplier you do not need large numbers of customers. You need large amounts of cash demand. That can be done by selling more stuff, fancier stuff, bigger stuff to a smaller number of people. How do I know? Because it is already happening.
Coming out of the last recession the middle class retailers (e.g. J.C. Penney) struggled while the upper end retailers enjoyed a great recovery and the bottom end (dollar stores) saw growth too. People are either on the down escalator or the up escalator. Fewer people are hanging out in the middle. Plan your life accordingly.
What I see in software development is a rising premium on being one of the very best. There is especially a premium on big data analytical skills (math, ML) combined with the ability to do software engineering to make that big data move into models for training. Also, the ability to rapidly learn lots of domain knowledge is very important. Curiosity and an excellent memory are very valuable.
Maybe in some areas trains will travel along at slow speeds with lots of UAVs flying off them to deliver goods along the way. But trains do not go to enough places for that to work everywhere. Still, maybe it could work in some areas.
"Almost all those jobs will disappear in 20 years as autonomous vehicles replace human truck drivers." *cough* No they won't. Not even close.
The necessary and expensive improvements to in-cab automation technology and radical changes to infrastructure, civil planning, and distribution logistics -- coupled with the cut-throat cost competitiveness of the freight industry -- will mean truck automation will be slow and gradual.
On the other hand, much of the demand for I.T. workers can be handled by automated scripts and overseas contractors who are willing to work for much, much less than American counterparts. Now. Today. Over a VPN connection.
Since we're making predictions, try this one: Over the next 20 years, expensive benefit requirements and corporate cost-cutting will send I.T. jobs offshore in the same way manufacturing jobs were shipped overseas in the 70's and 80's. (It's important to note "automation" did not erode the U.S. manufacturing base; a globalized economy, operating much more cheaply than US unionized labor, took those jobs)
I completely agree with you that improvements in tech will drastically change the employment picture of the future, but I believe you're significantly missing the mark as to how these changes will play out.
People who think automation would be cause of less employment day by day they are actually afraid of progress of science! In fact if they think that this progress is opening chances to start personal business and now more and more software engineers are required then the problem they keep in their mind will be addressed.
Automation really did cut the number of jobs in manufacturing for less skilled workers. See chart 4 and chart 5. Employment of Ph.D.s by manufacturers is way up. I bet a lot of ML specialists and roboticists. Domestic output is up sharply. But big lay-offs for people with less than a college degree.
IT jobs started getting shifted to India in the 1990s. I've managed people in India remotely while writing code here. Yet in spite of cheap intercontinental video conferencing and lots of software development outsourcing to India the demand for software developers in Silicon Valley is higher than ever. I am surprised by this. But propinquity still has enormous value.
Slow truck automation: look at Wal-Mart. They've got the cash resources needed to replace all their trucks and install automation equipment in warehouses. As soon as the equipment hits a positive ROI they will throw out the old and bring in the new. Jeff Bezos will spend big to cut hours and dollars out of Amazon package delivery time. This can only be done with aggressive automation. Bezos is working on UAVs to do delivery. He already bought a company to automate movement of packages around warehouses.
VC funding of robotics is way up. Lots of VC-funded companies are getting bought up by big Silicon Valley players. That's because computers have become powerful enough and algorithms have advanced enough to make lots of problems finally capable of being tackled.
Mapping out the coming collapse in work is pretty tricky. Truckers or nerds or nurses go first? That doesn't change the bigger point. The collapse is coming.
We have no clue what that era will look like or how it will affect our work-oriented culture. What does it mean when goods and services are basically free and only land and attention remain as rivalrous goods? What does it mean when the basis of human dignity - taking care of your family - doesn't mean going to work every day. Will we become like Saudi Arabia with robots taking the place of the Pakistanis and Filipinos who do the work?
(Did you see the same ad on this page that I did? Truck driver school?)
Globalization: Works well if you can "fire and forget" a project (e.g. mass production from blueprints). For coordinated tasks (e.g. half here, half in India) it's awful; language, culture, and time zones make a hash of things.
Globalization cause: Unions (and labor costs generally) are a part of it, but the morass of regulations and taxes are not insignificant, either.
opening chances to start personal business
You've obviously never tried to grow. Hiring that first person is horrendously difficult and expensive (see regulations, morass above). Hiring the fifty-first is hell on earth. Try opening a barber shop in your home; tip: have bail money ready. Try selling milk from your farm; tip: bail money won't help.
No one has mentioned the "sharing economy" that has potential (despite, not because of, the hype) to help in some situations but is also being strangled by rent-seekers.
We are being destroyed by our own government (huge unemployment but we need guest workers?!?). Hopefully there will be a tipping point before all hell breaks loose.
Good responses to my skepticism on truck automation. Regarding bots chasing down thieves -- how will that work? It's only property crime. Now you are going to take things up a level by going robocop on living, breathing people? Minorities, no doubt?
And what about the accident liability? Nobody has answered that one. Some 30,000 people are killed in car crashes each year. Suppose the death rate involving automated vehicles is only 5% as much. At 600 deaths a year, that would still be unimaginable robot carnage, politically viewed.
Imagine a news story on each automated car death. Drip, drip, drip. That would be wall-to-wall coverage. So far the total is evidently still zero, but that is bound to change soon enough.
And if Google cars are seen as meal tickets on wheels, that death rate will higher still. Good luck to Google billionaires in a courtroom in a poor southern state after their robot drives over someone's grandchild.
Lol, 5% of 30,000 is 1,500. Two percent is 600. Either way, major problem, no?
Randall, I respectfully apologize if this is long-winded, but there's a lot of ground to cover. I previously responded to your reference to the decline of US manufacturing jobs since 1978, and pointed toward the offshoring initiatives of the 1970's and 1980's as to the reason of for the decline in American manufacturing jobs during that time. The article you provide points to automation affecting manufacturing jobs, globally, from the late 1990's onward. While the phenomenon of offshoring and automation occurred consecutively and to a degree concurrently, it's important to note that automation is not the sole (or, arguably, even primary) cause of American manufacturing erosion in 1970's and 1980's... the period in which we saw the largest declines of American manufacturing as compared to GDP.
See the Percentage of Private Sector Output in the article above. From 1978 to 1987, manufacture of goods shrunk from 31% of GDP to 25%. Then the decline tapers off; it would take more than two decades to shrink another 6%. If American jobs had replaced by automation in that decade, those American goods would still have been manufactured... right? This correlates with the graph above it; the percentage of jobs producing goods declines from 35% to 27% from '78 to '87, and -- as with the previous graph -- it would take another two decades to see the same drop. Arguably, the largest drop in US manufacturing, in terms of both jobs and output, happened in that decade... when sending manufacturing to cheaper facilities offshore was a popular business practice.
The graphs in the article you provide article start in 1987, after the more precipitous drop of domestic manufacturing jobs and output during the prior 10 years. The article goes on to postulate that increased trade with China came at the expense of other Pacific rim countries, and presents a graph of decline that begins in 1994... without mentioning the effects of NAFTA, which began in 1992.
The trade deficit of manufactured goods, beginning in 1978, correlates with the drop in U.S. manufacturing jobs. Consider the following:
Though America has moved toward a service-based economy, outsourcing and offshoring continues to threaten American in the service and IT industries.
Telecommuting over the internet allows employers to tap into thought-based employees around the world, often in countries with weaker currencies and lower pay requirements. Reciprocally, aspiring global employees now have access to information and learning resources online that allow them to be viable candidates for companies looking to save money.
Demand in Silicon Valley may be high, and Google and Facebook may compete vigorously for the same rockstar professionals, but the vast majority of those working in the I.T. industry will never step foot in Silicon Valley or an A-list company -- much less work there. That vast majority are those whose jobs are at stake by automation and outsourcing.
"[Bezos] already bought a company to automate movement of packages around warehouses." Indeed. A bellwether might be automated yard jockeys shuffling trailers around private property in a safe, reliable, and less expensive manner year after year. From that point perhaps we can begin anticipating over-the-road automation in a few decades. UPS or FedEx may be better candidates for such tech than Amazon; that retailer's expertise is marketing and inventory management, not transportation logistics.
As for Walmart, they run ~30,000 trailers... only a quarter of which, general merchandise, is pulled by ~7500 company tractors and drivers. The tractors, drivers, and distribution centers for Walmart's grocery goods are outsourced to Werner, Schneider, JBHunt, Crete, and others. Why? Because those are the transportation logistic experts, and outsourcing means Walmart doesn't have to shoulder the cost of operating expenses in expanding and contracting markets. As for their cash on hand, googling "Walmart liquidity" and looking at their cash flow cycle suggests they're not in a position to build, buy, or renovate dozens of distribution centers and start purchasing tractors at double the current price... at least if they intend to keep their product prices at the very low level they are now.
Cost, when compared to return, drives business decisions... including those that affect employment. Just because something *can* be done doesn't mean it will be profitable to do so, near future or otherwise. Just some thoughts as we ponder how tech will reshape labor markets in the coming decades. (Again, thank you for tolerating the length.)
Bots chasing thieves: flying cop bots will be the eyes and ears, not the arresting officers. The cops will save huge amounts of time because thieves will be caught in the act with video and other sensors providing a history of every moment from beginning of crime until arrest.
Insurance? That's what will drive autonomous vehicles. An autonomous truck going 100,000+ miles per year will get into fewer accidents than humans driving the same distance. So autonomous truck insurance will cost less than human-driven truck insurance.
Japan shot themselves in the foot by running away from nuclear power after the Fukushima Daichi incident, and that had zero confirmed casualties. Just bad PR, and they made choices that were very bad economically. Heck, even Germany ran away from nuclear and instead went heavy into coal. Coal is 4000 times more dangerous (http://www.the9billion.com/2011/03/24/death-rate-from-nuclear-power-vs-coal/) but no matter.
In this age of sensationalist news, a death by an autonomous vehicle on the roadways will be a very exciting story. Several of those and you have a big PR problem. Like with nuclear, I suspect that the standard that the public will demand of autonomous vehicles is absolute perfection, which is almost impossible to to deliver.