February 21, 2015
Thinking About Robots And ZMP Workers
Some robots do not cause a productivity explosion because they replace low productivity workers and the quality of the task done does not improve much after the initial introduction of robots. The automated equipment that turns a worker into a zero marginal productivity (ZMP) ex-worker does not need to offer a large and growing impact on the firm's total productivity. It just has to lower total costs.
I was thinking about that after reading what I think is a wrong argument from Larry Summers about robots and productivity. He's coupling rates of productivity growth with rates of worker displacement. Those two aren't necessarily highly coupled regardless of how coupled they were in the past.
How much money do robots have to save to replace human workers? Not much from a total cost perspective. Before we explore that point lets briefly look at costs. Calculating costs can be hard because you pay differently for a robot than you do for a worker. In an economic downturn you can't fire a robot you own. But you can fire a worker. So buying the equipment commits you to a long term cost: paying the interest on the loan to buy the robot. That's part of the cost calculation for whether to replace the human worker.
So there's a risk from ownership of a robot. But that risk is not always large and it replaces other risks. Sometimes the need to do a task can be projected for many years into the future, long enough to pay off the robot loan. Plus, you could always lease the robot and pay more per month to be able to easily and quickly stop using it. A large rental and lease market demonstrates the usefulness of this practice. Besides, human employees pose their own risks such as theft, illness, injury, and lawsuits. Plus, the human can quit at any time. The freedom of the employer to fire is paired with the freedom of the employee to quit or damage the firm.
Okay, so getting back to the lower cost robot: Do robots which replace workers have to cause a productivity revolution? Not necessarily. Could be that a robot comes into a position in the workforce, replaces humans from that position with an initial small productivity boost, and then the new lower cost stabilizes with little additional future productivity gains in future years.
Take bank ATMs for example. Once they replace some bank tellers they are going to keep getting used regardless of whether their total cost of operation stays flat or decreases slowly in future years. Ditto self checkout at cash registers. Once the switch has been made what's the potential for future savings?
What I suspect might be happening with robots in many situations is exactly that. If these displacements happen among workers who aren't very productive in the first place then the total initial impact on the economy's productivity is low and so is the future year impact. All the $10 per hour workers are responsible for a pretty small fraction of the American economy's output. Automation can wipe out large numbers of their jobs without causing a huge surge in overall productivity.
Another point: A substantial fraction of the cost savings of robots comes from reducing the work load on managers. Management is freeing itself from the need to manage others. I suspect it preferentially prefers to dump lower productivity workers.
Suppose you manage low skilled and low productivity workers. Suppose they aren't very conscientious. What is your day going to be like? Kinda like someone running around plugging new holes that spout in dikes. Will you like your job? For most people the answer is going to be "No!". So automate the jobs and stop dealing with people who are a hassle to deal with.
Think about it from the perspective of managers. They'd rather manage more fun and interesting problems and more talented people. Computer systems and robots are enabling them to do that. So I expect firms to develop technology that lets them shift their hiring preferences toward the sorts of workers their managers would rather manage. This might be why the labor market participation of high school drop-outs has dropped so low. Nobody wants to deal with them and the technology now exists to avoid it.
Randall Parker, 2015 February 21 02:44 PM
"Suppose you manage low skilled and low productivity workers...So automate the jobs and stop dealing with people who are a hassle to deal with."
Of course, the next step is to eliminate the guy managing the ZMP'ers, as well. Now THAT'S a productivity gain.
The first level manager is getting paid much less than the combined cost of all the people he or she manages. So the savings isn't that big. But sure, eliminate bottom ranks and some of the people at the next level up become redundant as well.
The nature of the work at the next level up changes too. Someone has to manage maintenance and repair techs who deal with the equipment. Also, in highly automated environments. Software developers have to develop info systems to detect when the automated equipment starts to go awry and be able to diagnose it quickly.
Productivity isn't calculated on a cost basis, it's calculated on the basis of work hours. So, if you eliminate 10% of your labor hours you've increased labor productivity 10%. If the average salary of the laid off employees is only 1/4 of the company average then the savings is only about 2.5% of payroll costs.
So, 2 implications:
1st, laying off low-skill employees will have the same impact on labor productivity statistics as laying off high-skill employees. If it's happening at a high rate we should see it in the productivity stats.
2nd, laying off high-skill employees saves more, so they may be a more attractive target for automation.
Summers' 2nd question was: if *overall* labor supply & demand are in balance, then a shortage of low-skill jobs should mean an excess of high-skill jobs. In other words, employers should be having a hard time filling all the high-skill jobs, and wages for those jobs should be exploding as employers bid for employees.
Is that happening? Is there a *large* sector of the economy where there are plenty of jobs going empty, and employers are begging for applicants?
If labor productivity is calculated by hours saved then economists are using a really messed up productivity measure.
What we are seeing in the labor stats: a huge difference in the percentage of people in the workforce as a function of educational level. 41.5% of high school dropouts vs 71.5% of those with at least a bachelor's degree is a huge gap. We've also seen development of a much bigger wage premium for those at higher education levels (and higher IQ levels). The rise in the value of higher skill levels is understated by mixing together advanced degrees in economically valuable and not valuable subjects. The Gini coefficient has soared as well.
If labor productivity is calculated by hours saved then economists are using a really messed up productivity measure.
Well, labor is traditionally measured in hours. The value produced by the economy is measured by GDP. GDP divided by labor equals labor productivity. If that ratio isn't changing, then it's hard to say that we're getting more value out of people's work.
What we are seeing ...: development of a much bigger wage premium for those at higher education levels
Summers disagrees: "The premium to higher education has plateaued over the last 10 years. We see evidence highly skilled workers have less rapid career trajectories and are moving into less skill occupation if anything. Productivity is not growing very rapidly, and a lot of the employment growth we’ve seen in the past 15 years has been in relatively low education, in-person service occupations."
So, do we have data to support the idea that the difference in workforce participation is a new development? Or that Summers is wrong about the higher education wage premium? Or that the increase in the Gini coefficient isn't primarily caused by the shift from wage income to non-wage (ROI) income?
Yes. I work in Silicon Valley in finance. I see all hiring reqs that go out, and have done so for ten years. The premium to higher technical skills is higher than it has ever been. We are putting six figures on new college hires plus six figures of restricted stock. And we still can't get the hires. If you slice the numbers by technical positions only, I believe the premium for talent has gone up. Take out the PhD's in English lit and you will see a very different picture.
This problem is only going to get worse because it's not just about robots. "Knowledge workers" like Mr. Summers himself have been safe from the effects of technology so far. Indeed, technology has made them more productive and made it easier to do what they do. So far so good, right? And wouldn't everyone rather do something better and more productive than the menial jobs that robots do?
The thing is, at the beginning of the automation revolution, the only work that could be automated was the simplest of the simple work. Think simple robotic arms screwing in a single screw in the same place on every single car that passes through the assembly line. Or calculators and spreadsheets replacing the army of human calculators that used to be employed to do everything from menial accounting work to working the arithmetic out for engineers. All pretty straightforward stuff, very easy to program a machine to do.
But things have progressed since then, a lot. More than most people outside of the software industry understand. More than many inside the software industry understand. The day is very near when these knowledge workers start to have their jobs replaced by machines as well. The technology is already here to do it for the most menial knowledge work, but it's only barely made it out into the wild. And that technology is improving rapidly. As a software engineer myself, this is good - for now. It means more work for us, and a bigger share of the pie. But eventually we're going to create the very software that can do what we do ourselves and then we'll be out of work, too. I'm old enough that I'll probably be able to coast to a nice retirement before this happens, but if you don't think the day is coming then you're absolutely fooling yourself.
If we don't bomb ourselves back into the stone age, in a hundred years or so we're going to reach the point where it's unproductive for actual humans to do much of any work.
"In an economic downturn you can't fire a robot you own. But you can fire a worker."
Except for those LMP and ZMP workers who have union protections or some claim to minority status and are, therefore, unfireable.
"Could be that a robot comes into a position in the workforce, replaces humans from that position with an initial small productivity boost, and then the new lower cost stabilizes with little additional future productivity gains in future years."
That stability is the important part. You aren't going to have a robot claim that stocking shelves in Wal-Mart should be a productive and fulfilling career that deserves $50,000 a year. You aren't going to have a robot say that the other robots were creating a hostile work environment. You aren't going to have a robot file a lawsuit over the policy about working during state-mandated break times. You aren't going to have a robot claim that it slipped and fell and now has permanent debilitating back pain.
"If we don't bomb ourselves back into the stone age, in a hundred years or so we're going to reach the point where it's unproductive for actual humans to do much of any work."--Russell Newquist (10:51 AM)
At that point I expect machine-made artifacts will become unfashionable, uniqueness will become a premium price feature, and the Art Majors will take over.
"In an economic downturn you can't fire a robot you own. But you can fire a worker." But surely, you can sell the partially depreciated robot. That reduces your interest expense. Of course, you may not get much for it in a time of general economic distress.
"You aren't going to have a robot say that the other robots were creating a hostile work environment."
And the suits will be able to chase their female robot-secretaries around the desk with impunity.
Now that's progress!
I think there's a confusion here between labor productivity (output/hours worked) and total factor productivity (output/a combination of labor, capital, materials, business services, and energy). Your argument works for for TFP, but it doesn't for the more traditional labor productivity, because when you replace a person with an equivalently productive machine, the person's hours vanish from the denominator, which ought to goose labor productivity.
It doesn't seem to do that. The change in labor productivity has been very noisy in the oh-ohs and teens, but if you do a 5-year smoothing on it there's a maximum in 2003 and then it's pretty much downhill from there, with a small dead-cat bounce in 2010. (Note that we're now talking about the second derivative of absolute labor productivity. Numbers are here. Summers's argument for this is that labor productivity doesn't rise as fast with slack demand, and he uses that as a basis for a standard demand-side argument.
I'm not at all convinced that we could be successful with a Keynesian-style goosing of demand, but I buy the argument that productivity is meaningless in times of slack demand. If that isn't true, then something weird is happening.
Robots are property, and are written off against any taxable profit the company makes. The interest charged for any loan taken out to buy a robot is also deductable. This cannot be said of the human employees, at least not until SCOTUS overturns the 13th, 14th, and 15th Amendments on some specious technicality.
And as far as managers go, with less human staff there is also less need for them. They can hope that their former underlings won't turn on them when they join the bread line at the soup kitchen. Their replacements will be safe in India and China, managing the business over the Internet.
@Russel Newquist foresees the end of jobs for all except perhaps the owners of machines (and maybe for them, as well). This has been predicted at every technological revolution in the past, yet has not occurred. Instead new jobs were created at much higher productivity levels, with standard of living benefits for all, usually after some transient periods of dislocation. It is claimed that what is different this time is the machines are programmable/flexible and with algorithms that will eventually surpass all but the most intelligent humans. But, this is only a matter of degree. The robots used in manufacturing have been reconfigurable for decades. Not as fast as software-driven machines, but still a kind of programming.
Now it could be that software is completely different. But, we should be careful predicting the ways in which humanity will adapt to the new reality. History has demonstrated that we are unable to imagine the pursuits that would become possible in a world where machines are producing much more of the stuff we need and want at much lower costs. I can't offer any ideas, but that's my point. And, in the limit, perhaps what happens is that many humans stop doing jobs at all. The machines make plenty available at near 0 cost; scarcity is a corner case. People play music, hang out, view art, make love, play games, whatever. My guess it that the vast majority would be fine with that.
For my bona fides, I run the research at a company that applies automatic speech recognition and natural language processing to producing captions for video. We use a hybrid automated/manual approach, but, of course, are trying to do more and more with software. We're not there yet, but I do understand the technological possibilities here.
Some day we'll have enough goods and services (including housing, transportation, perfect healthcare, complete environmental remediation, childcare, education, eldercare, etc., for both the developed and the developing world), but it's obvious we're not really close. So, we don't have to worry about finding enough to keep everyone employed, at least not anytime soon.
The real problem is: can we redeploy our resources from old industries to new ones, fast enough to make full use of all our resources, including people (i.e., keep unemployment down)?
Summer's argument is that we're facing an output gap, and it's not caused by insufficient workforce education or intelligence, it's caused by insufficient aggregate demand.
If so, how do we goose demand? QE only helped a little. Maybe we need to deal with income inequality, to reduce the problem of wealthy people who don't spend enough?
"In an economic downturn you can't fire a robot you own."
Oh yes you can. You ebay the sucker. Take the proceeds, pay off the note. Book any loss, which would probably not be much different from the unemployment costs incurred from firing a human worker.
I will grant one thing -- many an IT project was started to simply get around HR issues. Management did not want to face a long negotiation so they automated and re-classed the job. In doing so they got to pick the workers who would get the new positions. The other workers were left behind.
I was involved in another project to replace CRTs with LCD displays. ROI, at the time, came down to flipping a coin. It would have cost a little more to buy the displays but they had a longer life than CRTs. But the ultimate page turn was the realization that LCD's were half the weight of a CRTs. According to the union contract any device over 40# required two techs at union scale. It was `LCDs for everyone` within a year. Labor dollars went down thereafter.