How computers turned machinists into problem-solvers
Computers execute flawlessly and never tire, but people are the ultimate GPT
Welcome back! Last week we took a break from the history of technological disruption to talk about some recent evidence on the rapid adoption of generative AI. Check that out if you didn’t get a chance! This week, we’re talking about machinists.
A machinist is a trained professional who operates machine tools. Most machinists work in machine shops or factories where they operate machinery that produces precision component parts. There are currently about 357,000 machinists in the United States, and their median pay is about $53,000 per year according to the Bureau of Labor Statistics, and experienced machinists can earn upwards of $100k. The job typically does not require a college degree, although machinists undergo extensive on-the-job training and often enroll in apprenticeship programs that provide formal certification. Many machinists belong to a labor union (mostly the International Association of Machinists and Aerospace Workers, or the IAM).
Machinists use lots of different tools and – well – machines. This includes tools for altering the shape of raw material (usually metal) like drills, mills, and die grinders, as well measuring tools like calipers. One particularly important tool is the lathe, a machine that rotates a piece of material around an axis to facilitate various shaping operations like cutting, sanding and threading. Lathes existed as early as 1300 BC, with clear archeological evidence dating back more than 2500 years in far-flung places like Italy, Turkey, Egypt, and China. Here is a carving from a Pharoah’s tomb of two men operating a primitive wood lathe that was rotated by hand:
This simple design was refined in various ways such as the pole lathe, which allowed a person to operate it solo by turning the axis with a foot pedal. During the revolutionary war, the English manufactured cannons by powering massive lathes with horses.
Precision metal lathes were developed during the Industrial Revolution and were powered by steam - and later, electricity. Here is a picture of workers operating industrial lathes from the 1950s:
The key skills required for industrial machinists were mechanical know-how, critical attention to physical details, and the ability to reliability create precision products that meet exacting quality standards.
The Rise and Fall of the Machin(ists)
The machining occupation rose greatly in numbers, compensation, and status in the post-World War II period. This was due to a combination of factors, including increased public investment in national defense, the rise of the U.S. as a global manufacturing power, and the growing popularity of labor unions.
The figure below shows the trend over the last 80+ years in employment and inflation-adjusted wages for machinists. I downloaded the data from IPUMS and used a crosswalk I created in previous work to get a consistent definition of the machinist occupation over time.1
After adjusting for inflation, machinist wages doubled from $14 to $28 an hour between 1940 and 1970. The IAM rightly claims a lot of credit for this success, but just as important was growing employer demand among manufacturers. Over this same period, machinist employment increased by 43 percent as a share of all jobs in the U.S. economy, from 2.3 percent to 3.3 percent.
As it turns out, 1970 was the peak. Employment declined modestly in the 1970s and then dropped by 34 percent (from 2.9 to 1.9 percent of all jobs) between 1970 and 1980, with another large decline in the 2000s (from 1.9 percent to 1.2 percent). Real wages for machinists have also declined steadily over this period, from $28 in 1970 to $22 in 2022. Usually when quantities (employment) and prices (wages) move in the same direction, you’re looking at a demand shift rather than a supply shift. In other words, firms had less need for machinists and were hiring fewer of them.
Falling firm demand for machinists had an unlikely culprit - computers.2 Specifically, the shift in manufacturing away from manual machining and toward Computer Numerical Control (CNC). CNC machines replace manual cutting, boring, drilling, and sanding of equipment with coded instructions that tell a machine what operations to perform where, and in what order.
The code to operate CNC machines can be written by a person or, increasingly, generated automatically from a 3D prototype built with computer-aided design (CAD) software. Here is a picture of a modern CNC lathe (notice the programming interface on the right):
Computers telling machines what to do
Methods of mechanically controlling machine tools preceded the invention of the computer. They were called NCs rather than CNCs, and they were controlled first by punch tape, where combinations of holes were punched in specific patterns that got translated into binary code to govern the machine’s movements. Later NC systems were more sophisticated and allowed operators to manually enter coordinates and instructions.
Around the mid-1970s, manufacturers of NC machines started replacing the dedicated hardware modules that controlled machine tools with microprocessors and control interfaces. CNC was far superior to NC in terms of speed, flexibility, and reliability. While NC relied on physical handling of punch cards, CNC stored machine instructions digitally in computer memory. This allowed for easier real-time adjustment of instructions and reduced human error in punching.
These clear advantages led to rapid proliferation of CNC machines in the early 1980s. Enghin Atalay and coauthors scraped the text of job ads from major newspapers to study the evolution of work in the U.S., and they chose machinists as one of their case studies. The figure below, from their paper, shows the number of mentions of “CNC” in machinist job ads over time.
The scale is a little hard to interpret, but it went from basically zero in the 1970s to very frequent afterward.
How CNC changed valve manufacturing
In one of my all-time favorite papers, Bartel, Ichniowski, and Shaw (2007) study how CNC technology changed the internal operations of valve manufacturing plants. I love this paper because it takes the details seriously. I learned more about the impact of computer and information technology from reading that paper than I’ve learned from dozens of papers with mathematical models of what economists call “skill-biased technological change”.
A valve is a metal device that gets attached to pipes to regulate the flow of liquids or gases. Valves are sold in bulk for standard objects like air conditioners, but also have specialized use in high-stakes settings like submarines and chemical processing plants. Valves are highly technical machine parts with multiple components that must all work well together. They often have careful geometric shapes and tight tolerances, so precision is key.
To understand how CNCs changed valve manufacturing, the authors surveyed all valve manufacturing plants in the U.S. in 2002 with at least 20 employees and asked about the firm’s production, hiring, profits, and other details.
They separated production into three processes:
1. Setup - figuring out what machine tools you are going to use to make a set of valves for a customer, and in what order;
2. Production - actually machining the raw materials to create the final product;
3. Inspection – checking the valves to ensure they meet technical specifications.
The plants became faster at all three processes between 1997 and 2002, with the largest reduction coming from reduced setup time. With CNCs, all machining tasks could be programmed at once, even across multiple jobs and tools, and features like 3D computer-aided design sped up the transition from idea to execution.
Efficiency gains in production were smaller and were mostly driven by the increased flexibility of CNC programming, which reduces frictions like switching across raw materials mid-job. Finally, sensors embedded in the CNCs automatically detected anomalies, which sped up the inspection process.
OK, so CNCs allowed valve manufacturing plants to operate much faster. So what?
Interviews with plant managers made it clear that faster turnaround times for jobs greatly increased the possibilities for customization. Valve manufacturing is a “batch” process, with each batch representing a customer order and a set of technical specifications. Faster turnaround time allowed plants to move up the value chain, away from valves as mass market commodities and toward bespoke custom jobs.
Broad labor market data wouldn’t reveal this impact of CNC technology, but it turns out to be very important for understanding the changing demand for machinists.
How did greater customization affect the machinist workforce?
Plants that adopted CNC technology hired more CNC operators, about one per machine. However, the number of non-CNC machinists and other shop floor employees declined by 14 per plant. So overall, the valve manufacturing plants in the Bartel, Ichniowski, and Shaw (2007) study employed fewer workers after adopting CNCs.
However, the machinists and factory workers in these plants may have found employment elsewhere. That is the conclusion of a recent paper by Boustan, Choi, and Clingingsmith (2024), who study the impact of CNC technology on aggregate manufacturing employment using a clever research design.3 They find large employment losses in more exposed industries, but also that employment fully shifted over into related fields. As it turns out, CNC technology was initially much more useful for metal manufacturing than wood or plastic, because those materials have lots of irregularities and tricky thermal properties. So manual machinists just shifted over to non-metal products. However, this is a short-run impact, and we can see in the aggregate data above that technology eventually destroyed a lot of machinist jobs.
Equally interesting from my perspective is how the job of machinist changed after the adoption of CNCs. When machining was manual, it was important for workers to be careful and precise and to have extensive mechanical knowledge. With CNCs, skill demand shifted toward knowledge of computers of course, but also toward greater planning and problem-solving. Customers demanded a high degree of intricacy and customization, which required more planning and coordination on the shop floor.
To understand changes in skill demand, they asked plant managers which skills had become more or less important between 1997 and 2002, and they compared the changes in skill importance among firms that adopted CNCs to firms that did not.
68% of valve manufacturing plants who did not adopt CNCs over that period said problem-solving skills had become more important, compared to 85% in firms that had adopted CNCs. For engineering knowledge and programming, the shares increased from 52% to 72% and from 14% to 43% respectively. CNC adoption caused rapid upskilling.
They also found that plants adopting CNCs were 75% more likely to say they organized their workers into problem-solving teams (73 percent vs. 42 percent). 93% of CNC adopters had regular shop floor meetings compared to 69% of non-adopters, and the CNC adopters were twice as likely to offer technical on-the-job training (67% vs. 36%).
Computers and soft skills – like peanut butter and jelly
Overall, they found very strong and consistent evidence of skill upgrading and changes in workplace organization because of CNC adoption. It is not very surprising that CNCs would increase demand for technical knowledge of computers. The more surprising and interesting finding is that CNCs pushed workers toward problem-solving and teamwork.
Frank Levy and Dick Murnane argue in their 2004 book The New Division of Labor that the complementarity between computers and soft skills was a more general trend throughout the economy. Using case studies of jobs like financial advisor, customer service representative, and auto mechanic, they show how computer and information technology pushed work toward complex communication and problem-solving.
Autor, Levy and Murnane (2003) formalize this idea in their famous paper about the impact of computers on jobs and tasks. They argue that computers substitute for people in routine tasks, meaning anything that can be accomplished by following explicitly programmed rules. This includes manual tasks like cutting valves, as well as information processing tasks like storing data in a spreadsheet. Computers flawlessly execute complex instructions that are specified in advance.
If a machine exists that can perform any one task better (and cheaper) than people, what’s left for us? Our advantage is flexibility. We can program a computer, walk up two flights of stairs, catch and throw a ball, and have an unscripted conversation, all in the span of a few minutes. When work is unpredictable, we can still beat out the machines, because humans are the ultimate general-purpose technology.
For the wonks, this is Standard Occupation Classification (SOC) code 514, which includes machinists, tool and die makers, and some other categories of metal workers. We then map this on to older occupation codes through a laborious process that I am happy to describe to you if you send me an email.
The IAW gives a fascinating timeline here, which catalogues how union membership has changed over the years and their putative explanations for the changing fortunes of machinists. Nowhere do they mention technology.
They observe that the machine tools replaced by CNC technology were more common in some industries than others, and they use this pre-existing variation to predict exposure to technological disruption using what is often called a “shift-share” research design.