Automation vs. Innovation

We don’t yet know how to make computer software that is as flexibly smart as human brains. So when we automate tasks, replacing human workers with computer-guided machines, we usually pay large costs in flexibility and innovation. The new automated processes are harder to change to adapt to new circumstances. Software is harder to change than mental habits, it takes longer to conceive and implement software changes, and such changes require the coordination of larger organizations. The people who write software are further from the task, and so are less likely than human workers to notice opportunities for improvement.

This is a big reason why it will take automation a lot longer to replace human workers than many recent pundits seem to think. And this isn’t just abstract theory. For example, some of the most efficient auto plants are the least automated. Read more about Honda auto plants:

[Honda] is one of the few multinational companies that has succeeded at globalization. Their profit margins are high in the auto industry. Almost everywhere they go — over 5 percent profit margins. In most markets, they consistently are in the top 10 of specific models that sell. They’ve never lost money. They’ve been profitable every year. And they’ve been around since 1949. …

Soichiro Honda, the founder of the company … was one of the world’s greatest engineers. And yet he never graduated college. He believed that hands-on work as an engineer is what it takes to be a great manufacturer. …

One of the early Formula Ones, Honda was embarrassed because the car they put out went up in flames. … Soichiro … he went back to the man who designed it and interviewed him. This guy had never tested these pistons, had never talked to race car engineers, never been on the ground to test pistons in the conditions they would, never even talked to race car drivers to find out how the car felt. He didn’t go to the spot, as Honda would put it. That kind of intellectual or design laziness really bothered him. And he embarrassed this man completely. He made him go around and apologize to everyone in the company one after another, holding the burnt piston. Now this man, however, also went on to design the next year’s cars, and those cars came in first place. And this man became head of Honda in North America. …

Honda believed that you have to argue all the time. You have to see both sides of every issue. Most companies have a difficult time with this. … I’ve been at factories where suddenly you see a group of workers standing together and not so much yelling but talking over each other. … They argue over every little piece of why it’s not fitting right, what they’ve seen in other places, what they could do better. …

So today the thing that stands out is that Honda can make any number of cars on the same assembly line, one after another. They can have a Civic and then an Odyssey or Pilot follow on the same assembly line. No one else has a flexible assembly line like that. … every time Honda does something, like build a factory, they’re building from scratch. Again, it’s on the ground: Go look at what we’ve been doing. Learn from it. Then build something else. … The Lincoln, Ala., plant, … is one of the most productive and flexible auto plants in the world. Uniquely, instead of setting up assembly line stations, where one person puts in the dashboard, the next station will put the radios in, and the next one will put the steering wheel in, at Honda they have zones of workers, so the zones put in five or six things. ..

Because of the flexibility, they are one of the least automated factories. Because they need human beings to work on these cars. If you’re going to have a robot put in a dashboard that has differences from one car to the next, you have to change the arms of the robot for every car. That can take hours.

Some would say that Honda not being automated and having more workers would hurt productivity. But it just shows what they make up for in flexibility. Again, their profit margins are better than anyone in the industry.

They do automate things once they feel like it has become a commodity. But once you automate, you can never improve anymore. A robot will never tell you, “Hey, I could do this better.” …

In most [firms] R&D, design, engineering, all of those critical aspects of the vehicles are still determined at the home office. Honda sets up an autonomous subsidiary operation wherever it goes. So, Honda China, Honda of North America, Honda Europe are each independent. … Quickly, they’re running on their own and developing cars and designs for the local market. … Each market determines for itself which cars it’s going to sell, which new designs they want to make. How they’re going to run their operations. How they’re going to build their factories. …

Around every major factory they have a research and development team and an engineering team. So that if they need something designed, or if they make a new model, and the factory is beginning to produce it, the R&D team is right there through the whole process of the production of the car. In most companies, R&D designs the car and it gets passed to manufacturing and R&D is out of the process. …

Honda doesn’t want their suppliers to be exclusive to Honda. That would only create a sense of “give Honda what it wants.” What Honda wants instead is to get new ideas all the time — not the ideas they know about but the ideas they haven’t thought about yet. (more)

GD Star Rating
Tagged as:
Trackback URL:
  • J

    My guess is that the talk on automation eliminating jobs is a political smokescreen for globalization — the jobs actually went to developing countries. Technology makes a better scapegoat than the subsistence farmers who now have much more lucrative factory jobs.

    • Doug

      I don’t think most of the jobs eliminated since 1985 are in manufacturing. IT really did replace a lot of solid middle to lower-middle class jobs: travel agents, bookkeepers, telephone operators, receptionists, tool booth attendants, stockbrokers, secretaries, video store clerks, television repair, book store workers, etc.

      Imports only make up about 15% of the overall economy, so it seems unlikely that globalization alone could have such a substantial impact on labor markets.

      • It’s manufacturing jobs that seem usually the main concern.

        So, we might ask what proportion of the manufacturing jobs eliminated was due to each cause. Anyone know?

      • Doug

        I don’t know the answer, but even asking the question in the right way is tough. For example what about a job that shipped to China, but in the absence of globalization would have been automated anyway due to cost pressure? What about a job that moved overseas 20 years ago, but then subsequently automated 10 years ago due to rising third world wages? What if a heavily human-based American carmaker had market share taken by a heavily automated Japanese carmaker?

      • These ambiguities also make dubious any unequivocal overall calculation (like 80/20), don’t they?

    • vaniver

      They’ve actually done the math on this. It’s about 80% automation and 20% globalization, if I remember correctly.

      Want to know how you can verify this easily? Look at manufacturing’s share of the American GDP and labor force. Manufacturing *must* be getting more efficient per worker to see what you see, and that implies automation.

  • TheBrett

    The people who write software are further from the task, and so are less likely than human workers would be to notice opportunities for useful change.

    I don’t know enough about robotics, but . . . more modular robots, maybe? And you could teach workers how to change the programming of them, although that would add yet another skill barrier to hiring more.

    But aside from that, I agree. That’s always been the general trade-off with more automation, from water-powered trip hammers in the Middle Ages to modern robots. You’re trading flexibility, comparatively high capital costs, and some maintenance requirements/costs for much better work done at a particular routine task.

    • Name

      “And you could teach workers how to change the programming of them,
      although that would add yet another skill barrier to hiring more.”

      Such workers would probably be assigned to write/modify programs for tasks not directly related to robots, rationalizing from a management (or from someone who allocates resources to such) standpoint not having to spend more for someone to work on those non robot tasks, and knowing that they would probably not be compensated more for doing such non robot programing work and thus taking away from worker time that could be spent thinking about innovating on robot related tasks.

  • Human physical dexterity and motor ability is much lower hanging fruit than our mental ability. I take your point, Robin, but we will get to super flexible and dexterous hardware soon and then human technicians can stand in for the brain, by doing the job themselves and in essence training the robots. Trainable robot arms and such are already on the market.

    • Upcoming hardware might be physically flexible, but I very much doubt it will be mentally flexible anytime soon.

      • Bob

        Human beings have no reliable definition of “intelligence”.

        So it’s quite obvious that you’re blowing hot air. No human being can search through information as fast as google can. Machine learning already outperforms human beings in many area’s.

        Humans aren’t special, in fact the state of our world (war, religion) shows the limits and defectiveness of our animal mind.

      • Doug

        Google’s search engine follows a set of instruction that any human being could mimic, given infinite time (read documents in universe, highlight search term, tally pagerank score on subset of matching documents). Google simply does it at much faster speed and larger scale. Its existence isn’t a justification of software flexibility, but rather raw microprocessor calculation power.

        As for machine learning, despite decades of developing very advanced blackbox algorithms (e.g. deep learning), using simple algorithms with heavy human feature engineering still outperform on the large majority of domains. In other words combining human flexibility and domain knowledge with machine processing power beats pure software solutions.


    Isn’t automation just the next step in specialization? Specialization has the same tradeoff of local efficiency versus overall flexiblity (and robustness) but in the end it has enabled tremendous economic growth since the time of the hunter gatherers. Sure there will be cases where automation backfires but it is somewhat likely that are descendants 100 years from now will have both a higher standard of living and a 16-28 hour workweek largely because of automation.

  • Charles

    Robots can aggregate experience much better than human workers. So software programmers will have access to much larger datasets from which to derive process improvements than any individual human worker.

    Also it is much easier to copy software between (identical) robots than to transfer similar skills between humans, allowing process innovations to be applied more rapidly and more uniformly. That also means that a single robot can, in theory, perform a much wider range of skills than a single human. In that sense robots have greater mental flexibility than humans.

  • Venu

    Those are some big claims to make on the basis of one stylized observation about a Honda plant. And Honda is not the only successful car company in the world, so it hardly is the case we can draw universal truths about automation and productivity from whatever works for them.

    “Software is harder to change than mental habits” – evidence please.

  • Viliam Búr

    Seems like automatization often comes with premature optimization. In that case, collecting more data before automating everything can bring better results.

    This leads to other question: why is the automatization done wrong, and why is it not fixed? But that can be explained by the company politics. You know, someone made an important business decision to buy some automated system, so if you criticize the system, it’s like criticizing the person who made that decision. So the system stays there, even if it would be better to replace it with some other system, or to replace parts of it with humans.

    • Silent Cal

      Also the gains from flexibility are near-impossible to quantify in advance. Also d(profit)/d(automation) was probably positive for most of the twentieth century, since it started at almost zero automation, so automation=good might be a bit baked in to some mindsets (story of Henry Ford and all).

  • David Allen

    The flexibility of “mental habits” is generally a problem for manufacturing, introducing time-to-time, person-to-person, shift-to-shift variation that can impact productivity, production quality, and rework costs; as a result workers often follow standard operating procedures that describe precisely how the work is to be done, and these can be as institutionally bound and as hard to change as monolithic software processes you alluded to.

    Software can be built with granular composable domain specific aspects making it flexible, and provide the domain level interfaces that will align the perspective of the programmers with the domain experts and workers, and allow those domain experts and workers to modify the system directly–while still enforcing change control and necessary institutional communication.

    Automation does not have to be a problem and can be an important way of capturing institutional knowledge; but the level and type of automation that will lead to optimal results will of course be context dependent.

  • slocum

    You should check out:

    In particular, take a look at the section titled “Machines as Children, Humans as Intestinal Fauna”:

    “First, machines are like children. The opposite of the overlord personification we’ve been encouraged to adopt by science fiction.

    Like parents, we have to let them have the fun while we child-proof the environment (sanitize their inputs) and clean up after them (do whatever they are too clumsy to do and clean up any messes they create).”

  • Axa

    Software/robots are not smart yet and probably is never going to be “smart”. However they excel at predictability, doing the same thing with the same accuracy millions of times. Final car assembly as explained by the article needs to be flexible to maximize productivity and at the same time avoid unsold car inventories. What about spark plugs, windshields or brake pad manufacturing? The actual car components, those manufacturing plants also need lots of flexible humans or can rely more on robots?

  • Pingback: Friday Links « Healthcare Economist()

  • The issue is not that automation will replace human innovation anytime soon. The issue is that a large number of lower and middle class jobs mainly require only performing a finite set of repetitive tasks – retail salespeople / cashiers, warehouse movers, customer service workers, waiters, office clerks, taxi / truck drivers, mechanics, accountants, legal workers, cooks, etc. – and that puts them at high risk of being automated in the near future.

    Sure, some level of process innovation can occur within these fields that might save a business money in the long-run. But businesses are notoriously short-sighted, and from an employer’s perspective, is it worth the additional cost of hiring an un-specialized laborer on the off-chance that they have a business insight over a robot that works 24/7/365 and doesn’t require insurance policies?

    As Moore’s Law drives down the cost of hardware and computing power by half every 18-24 months, automation is going to get cheaper and cheaper. It’s only a matter of time before it costs less than $30,000 per year to hire a robot or algorithm capable of performing the vast majority of a low-level worker’s daily tasks in a particular field. 2 years after that, $15,000. 10 years after that, $1,000. Sure, the employer might lose some level of innovation as a tradeoff. But does the average employee contribute $29,000’s profit worth of ideas to the business each year to make himself a viable contender?

    Especially when that business’s competitors start automating and they need to compete on price, there’s a tipping point where a human worker simply becomes too expensive to employ relative to a machine with 90% of the same capabilities. And even if a machine can only handle 90% of the work that occupies an employee’s day, the employer now only needs to hire 1/10th as many employees to take care of the remaining 10% of those other 9 who got laid off. In other words, robots don’t have to completely replace human workers in order to cause a huge disruption in the labor force.

    As someone who works in the technology sector (in fact, a large part of my job is in automating things), I think this is possibly the biggest social problem we’ll see in our lifetimes. In the long-run it will be great for humanity, but I think it’s going to be a very rough transition.

  • Philip Goetz

    Humans are more flexible than software only if they are motivated to make improvements. Otherwise, they are less flexible than software. Software doesn’t resist when you try to change it.