Benign Effects of Automation: New Evidence from Patent Texts

By Katja Mann and Lukas Püttmann, PhD students, University of Bonn. Originally published at VoxEU

Researchers disagree over whether automation is creating or destroying jobs. This column introduces a new indicator of automation constructed by applying a machine learning algorithm to classify patents, and uses the result to investigate which US regions and industries are most exposed to automation. This indicator suggests that automation has created more jobs in the US than it has destroyed.

Automation technology is fundamentally changing the way in which we live and work. Machines are taking over tasks that were previously done by human workers. But new jobs are also created, so what is the net effect of automation on employment? So far, we have mixed evidence. For example, Acemoglu and Restrepo (2017) report negative effects, whereas Autor et al. (2015) and Graetz and Michaels (2015) find no effect. Gregory et al. (2016) suggest there is a positive impact.

This disagreement might be due to the difficulty of measuring automation technology. The literature uses proxies, such as the share of a job that consists of routine tasks (following Autor et al. 2003), or the use of computers at the workplace and investment in robots and computers. Ideally, it would be better to measure automation technology directly as the outcome of an innovative process.

This is what we do in a new paper using patents (Mann and Püttmann 2017). We apply a machine learning algorithm to all US patents granted between 1976 and 2014, and identify the patents related to automation. Our classification is comprehensive, as it includes both physical inventions (such as robots) and cognitive inventions (such as software and processes).

By assigning the patents to the industries where they are likely to be used, and to the regions where these industries are represented, we create an annual measure of new automation technology at the level of US commuting zones. In our empirical strategy, we make use of the fact that we measure innovation at the national level, and employment effects at the level of local labour markets. We find that automation has positive net employment effects.

A New Measure of Automation

Patents are a natural candidate for measuring technological advances. But while they are often used as proxies for innovative activity, they have rarely been interpreted as indicators of new technology available to firms – even though Griliches had recommended this as far back as 1990. This is what we attempt to do.

We take the texts of all 5 million patents granted in the US between 1976 and 2014, and sort them into automation and non-automation patents. We manually classify 560 patents using the following definition:

An automation patent describes a device that operates independently from human intervention and fulfils a task with reasonable completion.

Based on this sample, we train a machine learning algorithm to find automation patents drawing on a dictionary of several hundred word stems. Patent texts are written in a standardised, matter-of-fact language, which makes them well suited for text classification.

The most important tokens are ‘automat’ and ‘output’ (see Figure 1). While these are fairly general, the algorithm also puts weight on more specific engineering terms such as ‘microprocessor’ or ‘motor’ and action verbs like ‘detect’ and ‘execut’.

Figure 1 Words that indicate an automation patent


Source: USPTO, Google and own calculations.
Notes: Token size is proportional to the value of the mutual information criterion in a sample of 560 classified patents. We show only the 150 highest-ranked tokens, excluding chemical and pharmaceutical words.

Figure 2 shows the number of automation patents over time. We identify 2 million automation patents out of a total of 5 million. We find a rapid increase in the number of automation patents, from 70,000 in 1976 to 180,000 in 2014. Automation technology is clearly on the rise: in this period, the relative share of patents related to automation increased from 25% to 67%.

Figure 2 Automation patents, 1976-2014


Source: USPTO, Google and own calculations.
Note: Shows all utility patents, classified as described in text and paper.

Next, we link the patents to the industries in which they are likely to be used, drawing on Silverman (2002). These industries may be different from the industry of the inventor. For example, Microsoft might hold a patent on spreadsheet software that can be used in finance. Our resulting industry-level indicator confirms that, in line with the theories of routine-biased technological change, more automation patents can be applied in industries that had a high share of routine tasks in 1960. The link between routine-task intensity and automation, however, has grown weaker over time. This might be a sign that routine jobs have already largely been replaced, but it could also mean that automation technology has evolved to carry out non-routine tasks too.

We identify in which US commuting zones the industries that can use automation technology are located. This allows us to calculate how many new automation patents that individual workers in a commuting zone could potentially use. The resulting dataset shows a striking pattern. Figure 2 maps the automation indicator for 1976 and for 2014 (the differences between these two years are characteristic of the trend over the whole sample). In the 1970s and 1980s, new automation technology was concentrated in the Great Lakes area, which had a high share of manufacturing employment. By 2014, the areas where automation technology can be used have become more dispersed, a result of both the evolving nature of automation, and changes in local industry structure.

Figure 3 Intensity of automation patents across commuting zones

Source: USPTO, Google, Silverman (2002), County Business Patterns and own calculations.

Employment Effects of Automation

We analyse the effect of automation technology on employment across 722 US commuting zones, over 39 years. We examine local economic outcomes of national patenting activity, in the spirit of Bartik (1991). This approach has the advantage that outcomes in local industries are unlikely to significantly affect national patenting. This suggests that most of the employment changes we measure should be due to advances in the available technology, rather than the other way around.

We regress five-year changes in the employment-to-population ratio on our five-year cumulated automation indicator. We use several controls, among them non-automation patents and demographic variables. We find:

  • More automation delivers more employment. More patents in a commuting zone leads to an increase in the employment-to-population ratio. A one-standard-deviation increase of our automation measure predicts a rise of a fifth of a percentage point in the employment-to-population-ratio per five-year period.
  • Weaker positive effects where more jobs are routine. The employment effect of automation is weaker in areas with a high initial share of routine labour, but remains positive.
  • Bad news if you work in a factory, good news if you want a service job. Manufacturing employment falls and service sector employment grows in response to new automation technology. In manufacturing, commuting zones with a high routine task share fare worse, whereas in the service sector, the interaction of the automation index with the routine task share is insignificant.

Our assessment of automation is more positive than most other research – automation, measured comprehensively and granularly through patents, has a benign effect on employment. Our analysis can be reconciled with previous findings. Routine jobs benefit less from automation, and manufacturing jobs (where more robots are used) tend to do worse. But this has been more than compensated by a rise in service jobs.

See original post for references

Print Friendly, PDF & Email


  1. dk

    Bumping labor over to service industries, even if it creates more positions, loses or fails to cultivate more specialized and valuable skills (and of course, service positions generally don’t pay as well). At a time where national infrastructure is degrading with increasing rapidity, technical training is declining.

    1. Disturbed Voter

      Law of commons … just because we are making more jobs in China/India, and increasing their standard of living … doesn’t necessarily increase the demand for our “services”, even cheap ones. Hard for all those new industrial workers to stop for lunch at a McDonalds in America. Want General Tsao Chicken with that? This applies even within the US … the immiseration of one region by another, is old hat.

    2. sgt_doom

      You put your focus on the exact problem — well stated!

      Without those production, manufacturing, developmental-type jobs, real technological progress does not happen – – but will occur in those countries those jobs have been offshored and/or created in!

  2. Jon S

    This doesn’t provide any causal relationship. Just correlation. Of course I didn’t read the Bartik paper.

    1. visitor

      Something bothered me about this otherwise interesting article: the way that patents are supposed to be an indicator of the sectors in which automation has taken place:

      By assigning the patents to the industries where they are likely to be used, and to the regions where these industries are represented, we create an annual measure of new automation technology at the level of US commuting zones.

      There are three aspects putting that reasoning step in question.

      1) Most patents do not result in any practical application at all — but are simply used as an armoury of IPR weapons to deter competitors from entering a market, or from daring to sue for IP infringement.

      2) The fundamental advantage of a patent is that it gives its holder a legally enforceable monopoly. In other words, truly valuable patents are kept for the sole use of a corporation whenever possible, and will not be widely disseminated across industries — or at least not without a considerable delay.

      3) Those patents with a very high impact are “essential” patents — those that it is uneconomic or even technically impossible to circumvent when implementing some product or process. So much so that in some areas like telecommunications, firms must declare those patents essential for implementing international norms, and make them available on FRAND terms to their competitors.

      Both (1) and (2) should muddy the link “automation-patent” – “automatable-industries”. In addition (2) makes this link time-lagged — because it takes time before a firm decides that licensing a patent is worthwhile, or before the patent expires. Aspect (3) is possibly the one that determines the actual uptake of automation in an industrial sector — and, regardless of patent density, it will very different depending on institutional arrangements (compulsory FRAND licensing vs. exclusive IPR).

      Surely enough, the authors hint at some of the problems in their original paper:

      Linking patents to the industries of their use is difficult. If we wanted to measure the actual usage of a specific patent in a certain industry, we would need data on out-licensing. But this information is not available, as firms and research institutions have incentives to keep their licensing agreements private. Interpreting patents more indirectly as a proxy for automation technology rather than a direct measure, we can use information about the areas in which patents can potentially be applied.

      In other words, the authors do not know whether automation patents actually led to automation in the industries concerned by them; they assume it is so, and that patents affects all industries with proportional intensities.

      They checked the plausibility of that assumption by correlating the automation patents per industry with the investment in automation per industry. The latter variable is again measured by a proxy — surveys about the investment in or usage of robots and computers in selected industries. They also computed the coefficients with lagged data “accounting for the fact that it may take some time to translate a patented innovation into the actual production of this technology.”

      When presenting the results, the authors state:

      All correlations are highly positive, which indicates that our automation measure captures both advances in robotics and in software, which are then translated into production and trade of computers and robots.

      A look at table 5 reveals some oddities though: all correlations for the “patents-robot shipments” (lagged and unlagged) are significant when taking into account “time and industry fixed effects” — but none are when taking into account only “time fixed effects”. Conversely, correlations for “patents – computer investment” (lagged and unlagged) for one of the two “computer investment” proxies are significant when dealing only with “time fixed effects”, and none are when dealing with “time and industry fixed effects”.

      The whole comparison of proxies with other proxies seems to me to lead to a fragile chain of derivations. Not taking into account the nature of patents (essential vs. non-essential) and the institutional arrangements (exclusive IPR for 16/20 years vs. immediate compulsory FRAND licensing) leaves me unconvinced at the strength of the computed effects on employment, and therefore I am not even tempted to discuss the direction of those effects.

      1. Jeremy Grimm

        This is a strange bit of research for two PhD students at the University of Bonn to work on. I wonder where their funding came from and/or where they hope to find employment.

        The relationship between automation and jobs revealed through deconstructing patent literature — what’s next? Will economists begin applying advances in bird auguries made through capture of members of the auguring flock, their sacrifice, and haruspices [through inspection of the sacrificial victim’s liver] to questions of the economy? [I hope such practice may be constrained to readings of starling flocks.]

        Job creation as a happy side-effect of automation? That notion and its warrants is as convincing as the studies made suggesting we can increase employment and grow the economy by giving more money to the megarich. Maybe the authors of this post can find some auguries in their patent literature to bolster supply-side claims.

        Also, I feel a little cheated by these authors. Where are their partial differential equations and Hamiltonian models? How can I put faith into a pronouncement on economics without several pages of unintelligible mathematics?

        I think this paragraph stolen from a recent post by the Archdruid may better explain the nature, impacts, and intent driving automation.

        A Few Notes on Nature Spirits, Part One: Nature as “It,” Nature as “You”

        “Modern industrial civilization is terrified of the I-you relationship, and goes to really quite astonishing extremes in its attempts to force all relationships into the I-it mode. The flight from the I-you relationship isn’t limited to the obvious. The frantic efforts to replace human workers with machines even when the machines cost more and don’t do as good a job—a fairly common occurrence these days—are motivated, not by the shibboleths of profit and efficiency that get bandied around so freely in such situations, but by the fear of having to relate to employees as human beings. The common corporate phrase “human resources” expresses the same fear in a different key: you don’t interact with a resource, after all, you just exploit it.”

  3. johnf

    I have to wonder how much noise their token clouds contain. I could imagine, for example, patenting an electric toothbrush. With little trouble I could hit many popular words in Figure 1, with: schematic diagram, user, force sensor, non-volatile store, automatic stop, signal LED, push button switch, oscillating system and wireless charging

    Perhaps the authors will address this in their theses, and I should visit Bonn and attend their defense.

    1. Larry

      You raise a good point, there is likely to be a high amount of false positives. I also contest their notion that patents are written with standard language. The claims section is very legal and skilled drafters will often write patents so that they are hard to find when searching.

      Finally, I would have made a further restriction and looked at granted patents with at least one foreign equivalent. A patent after all only gives you exclusivity for your invention, it does not mean that you have to make an actual product or code some software. If the US patent also has an equivalent in another market its a better sign that the assignee actually plans to do something with the intellectual property.

  4. The Rev Kev

    This article, I am afraid, has raised my hackles so here goes. (Rant mode enabled). Far be it from me to rain on the parade of two young Phds from Bonn (a lovely city to visit) but this has got to be challenged. At least trying to use patent numbers/types shows some original thinking but is hardly a penetrative one. Where to start?
    OK, how about the fact that there is an assumption of equivalence in that a job taken over by automation is the same as a service job. So a guy that has spent twenty years on machinery would automatically transition to a service job like flipping burgers or being a Walmart greeter. Perhaps the correct orientation is to ask how many jobs that pay a living wage have been automated to jobs that no longer pay a living wage. America has become notorius for some of the biggest corporations paying wages so low that American taxpayers are now making up the wage difference in the form of food stamps and the like. That’s right, taxpayers are paying more taxes to help some of America’s biggest corporations be even more profitable. Disgusting!
    There is also the matter of transition time between the old and new jobs. It is not exactly immediate as when jobs have been automated it takes time to rebuild a new economy that will play host to new service jobs. And what sort of service jobs anyway? Shamelessly stealing a fact from another article from NC, it was reported that “Nearly 95% of the jobs created when Obama was in office were part time or contract work.” The study showed that the jobs were temporary, contract positions, or part-time “gig” jobs in a variety of fields with women getting the brown end of the stick as usual. What sort of taxes will they be able to kick back into the economy. Where will their consumption power come from? Buying a house? Buying a new car? Forget it.
    This brings me to the subject of externalities (bear with me here). That is, costs that do not appear in a balance sheet for automation as we still use primitive economic models. How about wrecked lives, marriages, dislocation through automation. Someone has to pay those costs. Opioid abuse and deaths I would judge as something else that really should be factored in but like this study, modern economics only takes a very narrow slice of economic activity and ignores the matrix that the economy is embedded in. Others may suggest additional costs.
    I suppose one reason this riles me all is that I have an ancestor that was transported out to the colonies at His Majesty’s pleasure back in 1831. He was one of a group of rioters that was smashing up machinery in the 1830 Swing Riots ( as the machines replaced the work that the workers had over the winter which for them meant no money, no ability to buy food, no nothing over the cold winter months. Charity was stingy and at the sufferance of the very people that were bringing in automation with no money for either internal or external emigration. Yeah, we’ve been here before. (Rant mode disengaged)

    1. Normal

      You’ve identified some effects of automation. Further progress in automation is inevitable. The problem has not been automation, but our failure to manage its effects on people. In the past, we have relied on the “next wave of innovation” – first industrialization, then IT – to replace the lost jobs – first in agriculture, then in industrialization – and we had been lucky until recently. There’s no evidence that another wave is coming, or coming in time to save the economy.

      We need to envision a world where most work is automated. How do we distribute the proceeds?

      1. diptherio

        I agree that automation isn’t the problem in itself; it’s the matter of who owns the machines and who benefits from their produce….i.e. we need to democratize the ownership of the means of production.

      2. Chris

        Thank you Kev and Normal.

        I lost the article when they said AI had increased jobs…

        Does anyone remember all the people in the US and here (aust) who used to work in the auto industry. Australia had GM, Ford, Toyota, Mitsubishi… now nothing. Our govt wouldn’t kick in $500m in subsidies and automation (and lower costs for overseas makers) are squarely to blame. The job losses from all those wages including those from suppliers is more than 200,000 by some estimates. Adelaide was particularly hit and our generous govt offered them a submarine making deal, costing a mere 300b from the taxpayer (so we are led on of course) – compared to just 500m to keep our auto industry. These are the family bloggers we are dealing with here. Grubs, scum, you name it.

        In US, at least the politicians agreed they needed an auto industry. But, you will all remember when the US makers were all bankrupt, instead of nationalising or just saying to the stock and bond holders stiff cheddar, the taxpayer bailed them out with huge amounts of money, which the auto makers used to modernise their plants with machines…

        Don’t know how many men and women lost those well paying jobs and were tossed into the rubbish bin and forgotten. Perhaps someone here will know

        As for distributing the profits from an automated world where the only jobs are maintaining the machines… we already can see how that is going to pan out.

    2. cnchal

      So a guy that has spent twenty years on machinery would automatically transition to a service job like flipping burgers or being a Walmart greeter. . .

      It could go either way, with a career as a Walmart greeter at least one would have an operable body a few years hence, unlike the ex Amazon warehouse worker, as the consequences of extreme physical and mental duress while working takes it’s toll. A steady source of new victims for the medical system.

      Characterizing service jawbs as low level any monkey can do that work, leaves out the high end service sector, comprised of bullshit jawbs in the FIRE sector. The problem for the machinery guy is that isn’t what he wants to do.

  5. Jesper

    Using the past to predict the future is usually a good idea but as the disclaimers usually state when buying financial products: Past performance is no guarantee of future perfomance. If past performance was a guarantee then I’d buy bitcoin….

    What we do know is that after the automation/mechanization of agriculture then most jobs disappeared from agriculture and it is unlikely that those jobs come back. People then went to work in manufacturing.
    It seems to be proven that after the automation in manufacturing then most jobs disappeared from manufacturing and it is unlikely that those jobs come back. People then went to work in offices.
    Office work is now being automated. Some jobs have almost disappeared, other jobs are disappearing as we speak and we’re now being told not to worry. Jobs will be available, we just don’t know anything about those jobs just keep the faith and your jobs (rather living standards) will be safe. Just spend a lot of time and money on ‘educating’ yourself to become more ’employable’ as there is a high risk of losing income unless you do. But don’t worry about losing your job because in the future there’ll always be new well paying jobs and we know this as for a blip in time in human history that was true. That happened when markets didn’t rule supreme as they do now, but don’t worry about that insignificant detail. Just worry about yourself and your individual ’employability’, the government might be elected by people but it is now run for business.

    All that being said it is fascinating how IT has changed research. Without IT their thesis would never haven been produced as the number of hours spent in collecting and analyzing the data would have been prohibitively huge. So yes, in one sense their jobs were created by automation/IT, whether or not their jobs and findings are useful is more of an academic question.

    1. FreshOH


      These are PhD students. Not employees of the University of Bonn, right? Their work is existential at best. Sure their work relies upon automation/IT. Which did require a variety of employed citizens to build. But the issue there is for how long are those citizens employed?
      — The Rev Kev, “Shamelessly stealing a fact from another article from NC, it was reported that ‘Nearly 95% of the jobs created when Obama was in office were part time or contract work.’ The study showed that the jobs were temporary, contract positions, or part-time “gig” jobs in a variety of fields with women getting the brown end of the stick as usual. What sort of taxes will they be able to kick back into the economy. ”

      As humanity marches on toward the singularity/self aware machine learning. The speed in which work like this paper will become increasingly simple to accomplish.
      How much computer effort (time, electricity, etc) was utilized to accomplish the task?
      For now it is up to the human to interpret the data. For PhD students, that is what three long weekends at the bar theorizing with the local riffraff?
      I’ve been informed from some college bound friends that Universities are offering accelerated classes. What used to be taught in 16 weeks is now offered in 6-8 weeks or even one week of day long lectures.
      Is it any wonder that those 45 and older consider these college students less educated upon graduation than just the previous one or two generations of college students.
      Behold the dumbing of our nation.
      In order to obtain that supervisory McDonald’s service industry job the employee must have at least an associates degree from the local community college. Which in 5-10 years will be replaced by a service technician/contractor for the automated corner McDonald’s with automated deliver to your GPS coordinates. The consumer just has to remain motionless for 15 minutes after submitting the order.

  6. Nat

    I don’t buy it. If automation created more better jobs then no company would invest in it because it would inherently increase overhead and costs. In such an imaginary world only non-profit organizations would use it.

    1. Jon S

      Automation may increase the number of jobs given specific circumstances. Suppose you would like to purchase a product, but it seems twice as expensive as what you think it is worth. The company then applies automation that allows them to reduce the cost (and price) by half, perhaps by cutting 50% of manufacturing staff per item.

      Now you are willing to purchase the product at the new price, as are millions of other folks. A ten-fold increase in production will lead to a five-fold increase in jobs (in this scenario) thanks to automation.

      This is what happened in the early industrial revolution and is the basis for most right-wing economic theorists. Frankly, most automation today does not lead to significant increases in efficiency and cost reductions, resulting products are produced overseas, and lead to no jobs in the USA.

  7. Normal

    The paper needs much interpretation. The authors define a convoluted equation for “number of national automation patents that can be used by a single worker”, and I’m have no idea what the meaning of this calculation is.

  8. mpalomar

    The study’s focus on patents seemingly excludes relevant factors that might contradict their conclusion regarding tech growing jobs. Others note the shortcoming of the study for not looking at the quality of jobs supposedly created. Why if more jobs have actually been created, have wages flat lined during the same period for the average worker? Is the employment calculation done using new definitions of employment that fails to differentiate part time, gig work and consider falling rates of worker participation?

    During this same period trade agreements have been used to promote flows of technology and capital across borders while labor has remained more anchored to region. Tech has made industry less reliant on skilled and semi-skilled workers, contributing to the weakening of workers’ position overall, allowing business to beat up labor. The hi tech sector of US industry was until the mid twentieth century what is now the rust belt, new tech (since 1976 in this study) facilitated the movement of factories to (race-to-the-bottom) right-to-work US states and foreign countries.

  9. Wukchumni

    Above the land, across the sea,
    They’re everywhere, we need to be.
    They’re brothers of a special kind,
    A better non human brand, you’ll never find.
    Band of robot brothers, that’s what they are,
    Fighting evil, near and far.
    Band of robot brothers, that’s what I said,
    Baptized by electricity, powered by lead.
    They’re lean and mean, and fit to fight,
    Anywhere, day or night.

  10. Jessica

    “More patents in a commuting zone leads to an increase in the employment-to-population ratio.”
    (1) “Leads to” rather than “correlates with” is such a rookie mistake that it makes me wonder if this research was somehow supported by a party with an ax to grind.
    (2) This does not prove that patents (=automation) increases jobs. It only proves that it increases them in the area where the automation is applied, which might be different from the area where the jobs it destroys were located.
    Think of the well-known example of agriculture. Patents might correlate with increased jobs in the area where John Deere headquarters is located and at the same time correlate with massive loss of employment on farms.
    Or there could easily be an increase in patents and (software) jobs in Silicon Valley combined with a loss of jobs across the rest of the country, for example the data entry jobs that disappeared in the 1980s.
    Moving forward, if truck drivers are replaced by AI (or trucks are turned into drones are operated by remote control), there will be patents and jobs in a few concentrated locations and massive job loss across three countries, but their technique would read that as automation increasing employment.
    The same pattern could of course also apply to Europe: patents and jobs in Germany and massive job losses in other EU nations.

    1. Rosario

      Good catches. I had my initial suspicions, but on second pass I’m beginning to think there was an agenda behind the research.

  11. Mystery Man

    Can someone explain to me how number of patents granted actually relate to technology being used in the real world and thus employment? Not all patents get created and sold, surely just a subset – could even be a very small subset of overall granted patents. Am I right in thinking that rather than weighting for this, they simply accepted it was a valid measure based on that survey from 1990?

    1. Yves Smith Post author

      The big criticism of the use of patents as a proxy for innovation is that the most common use is for cross country measurement, when legal practice (the importance of patents) and the legal regime (in particular, what it takes to have a patent issued) varies a lot. See this slide show:

      This is a within US comparison, so the biggest complaint does not apply.

      Second, the authors tried to screen for relevant patents. I’m not able to judge the validity of that approach.

  12. Stephanie

    I wonder how many of these automation patents are fracking related? That was my first thought when I saw the boom in patent filing in the Dakotas, Nebraska, and Wyoming that seems to have taken place in 2014. I can see where that would correlate to increases in employment in those areas, for a while at least. After that?

  13. Fastball

    There is more to consider than just job effects from automation. There are also environmental damage effects to consider.

    What happens to the ecosystem even at full employment when robots are doing all the work and demand becomes an abstraction?

  14. Vernon Hamilton

    the Automat?! what f’n century are we living in !?!
    i just can’t tell if irony anymore. my sense of irony has faded along with my sense of smell.

    1. Mel

      Automate Automated Automating Automation.

      Nothing to do with P.D.Q.Bach’s Concerto for Horn and Hardart, discovered by Prof. Peter Shickele of the University of Southern North Dakota at Hoople.

  15. TomDority

    Laborers knowing that science and
    invention have increased enormously the
    power of labor, cannot understand why
    they do not receive more of the increased
    product, and accuse capital of withholding it.
    The employer, finding it increasingly
    difficult to make both ends meet,
    accuses labor of shirking. Thus suspicion
    is aroused, distrust follows, and soon
    both are angry and struggling for mastery.

    It is not the man who gives employment
    to labor that does harm. The mischief
    comes from the man who does not
    give employment. Every factory, every
    store, every building, every bit of wealth
    in any shape requires labor in its creation.
    The more wealth created the more
    labor employed, the higher wages and
    the lower prices.
    But while some men employ labor and
    produce wealth, others speculate in the
    lands and resources required for production,
    and without employing labor or
    producing wealth they secure a large
    part of the wealth others produce. What
    they get without producing, labor and
    capital produce without getting. That
    is why labor and capital quarrel. But
    the quarrel should not be between labor
    and capital, but between the non-producing
    speculator on the one hand and
    labor and capital on the other.

    Written sometime in the 1920’s

  16. ewmayer

    Other readers have already pointed out the fallacy of using patents as a proxy for automation-vulnerable industries – allow me to simply note an interesting confluence of articles today on this theme. First, in today’s 2PMWC we have a piece on how the Deplorables seem to not be buying into the “automation is at worst neutral, and at best great for jobs!” propaganda being peddled by the MSFM and the legions of Useless Eaters employed at various economic think-tanks:

    Shipping: “Contract talks between the International Longshoremen’s Association (ILA) and the United States Maritime Alliance (USMX) broke down on Wednesday over union concerns about potentially job-killing automation” [Splash 247]. “Union officials fear employers aim to use automation to wipe out dockworker jobs entirely rather than have automated features that would be operated by workers.”

    Now gosh, where might these benighted-heathen dockworkers be getting crazy ideas like that? Perhaps from stories like this one in today’s Links:

    Not in my lifetime: a dangerous sentiment in maritime? Splash24/7. Chuck Roast:

    The operative phrase is…”And before shipowners say that automation will never happen, they should visit Quingdao’s new automated container terminal that operates 24/7 with only 9 employees.” Yikes!!

    But I’m sure most of the dockworkers whose jobs went poof in Qingdao quickly “retooled their skillset” and moved on to high-paying software-developer jobs, right?

  17. Rosario

    1) This:

    “Bad news if you work in a factory, good news if you want a service job. Manufacturing employment falls and service sector employment grows in response to new automation technology. In manufacturing, commuting zones with a high routine task share fare worse, whereas in the service sector, the interaction of the automation index with the routine task share is insignificant.”

    Big one here, Employment (not =) good, unless it is Good Employment (living wage, benefits, duration/security). Service jobs tend to not be this. As our current regime is showing. Are these job quality metrics left out intentionally? Because I’m noticing a trend among economists when talking about employment. Despite my issues with his simplistic thesis, I can only think of the intuitive observations in David Graeber’s “Bullshit Jobs”.

    2) From what little I remember from my education, using machine learning to rouse 2 million “automation” patents from 5 million using training words is a far cry from distilling actually utilized patents in manufacturing applications over patents claimed for purely protective purposes (as noted by visitor above). Maybe they clarify this in their full technical paper? In addition, there is nothing to indicate that an automation patent is exclusively for use in manufacturing. There are a massive number of automation tasks done in software, and I am willing to bet many of those are patented. In that vein, I’m also willing to bet a large number of post 1998 patents (the big jump) are part of the growth in tech and software patents, and those are very contentious among patent theorists and tech professionals generally (i.e. I don’t know if software patents are the droids they are looking for).

    3) I’d have to see their deep research but I don’t really see how this would pass the sniff test on a first read among policy level people, unless it is what they wanted to hear. In academy, it would probably get a pass and published in a journal to show their modelling and analysis chops, even if the correlations are all over the place or nonexistent.

    I’ll believe modern automation (the last 40 years) in the Capitalist framework is good when I see the social benefits broadly applied. So far Capitalists, by design, tend to utilize automation to displace human labor if the cost is right, and how that has translated by the magic of free markets to morer betterer jobs, right now, is not apparent (see 1). Last I checked, wages were still stagnant or decreasing, and people still worked more than they enjoyed free time (or they have plenty of free time and no money). So when do these benefits, “kick in”? There are plenty of utopian wonders to be had from automation such as decreased work day, more fulfilling (creative?) employment, etc. etc., but the benefits don’t appear to be getting passed to the working class unless one counts inexpensive consumer goods bought with debt.

    Finally, and I can’t stress this enough, even with machine learning and mountains of “data” the last 40 years is not a bellwether for the next 40 years when you are talking about complex social and technological development. You’d think the unpredictable utilization of every technology in the 20th century (particularly the Internet) would have buried this approach in favor of something a bit more adaptive/stochastic rather than predictive/deterministic.

    1. Yves Smith Post author

      Manufacturing jobs are not what they used to be. Your fond take on them is from the days when unions were widespread, and even non-union shops were not much below the level of unionized manufacturers. See this article….from a manufacturers’ trade publication:

      There remains in the U.S. a significant number of manufacturing jobs—production jobs—that earn the negative impression. These jobs involve tedious manual tasks, often involving dangerous machinery, and are low-paying. One recent report found that manufacturing production wages now rank in the bottom half of all jobs in the U.S., and that a third of the families of those workers “are enrolled in one or more public safety net programs.”

      Moreover, the idea that manufacturing jobs are more stable is questionable in light of offshoring and factory closures.

      My brother’s relatively well-paid job in one of the few manufacturers that is still unionized is literally killing him. The bosses are trying to drive older workers out by imposing schedules of multiple days running of 12 hour shifts, in combination with requiring workers to switch shifts from night to day and back over the course of a month. And they’ve greatly crapified his health insurance. He’d quit for less than half his current pay if he could get health insurance, but at 56 and very much overweight (despite not being diabetic), I can’t see how that happens. He insists he’s quitting next year at worst and I don’t see how he can afford that either. And yes, he has very good computer skills (he runs the computer operations now) but that’s not enough of a calling card, particularly since he lives in the boonies.

      1. Rosario

        I’m sorry to hear about your brother. I only hope he has some pension sloshing around that can be tapped. My mother is in a similar position as a teacher in KY, Gov. Bevin is not helping vis-a-vis pension security.

        Yeah, what you say is true, though I’m only fond of the strong-union manufacturing of yore because I’m unable to see a viable solution to the current economic environment. Unless, of course, strong unionization takes hold again in modern forms of employment (tech, service) with less conventional organizational structures. Some variation of IWW membership comes to mind, where members utilize the organization as they see fit and membership is very fluid. I imagine tech workers would be more sympathetic to this model.

        I can see precariat employment (service, periphery tech, and self-employed/”entrepreneur” work) being viable for many people my age (millennial) if they a) don’t have any kids b) don’t mind renting their entire lives c) plan on not getting old. I don’t necessarily think a) and b) are problems, I’m for reducing population and have no issue with renting (I’m still renting in 30s), many people would prefer to rent, but the real killer IMO is c). People can’t live like a 30 something or 40 something when they are in their 70s or 80s. In addition, the above only states that precariat existence can be viable (even sexy) for youth, but it says nothing of the fact that the 1% are bloodsucking society while selling this lifestyle to my generation to, in a way, buy silence. It may not hurt much now, but it will kill later.

  18. Madmamie

    Seriously? All that just to point out what we already knew: automation, over time, will leave us with only care-giving to fill our days! And even that can eventually be accomplished by bots. I thought this was article from the Onion at first but the tongue in cheek is too subtle.

  19. d

    maybe dollar general is the canary in the coal mine for how automation and out sourcing has impacted the middle class, it has added stores when most retailers are closing stores

Comments are closed.