Bill Knight, an assembler at General Electric’s plant in Grove City, Pennsylvania, demonstrates how to torque a bolt on the flex plate of a 15-ton locomotive engine. Instead of lifting a heavy tool, weighing 25 to 40 pounds, like he used to do, he almost effortlessly guides a robotic tool. Once the tool is in place he uses a single finger to activate it, and it torques the bolt perfectly. Then it tells Knight where to place the machine next.
“How many bolts do you have, Bill?” GE’s plant manager, Jeff Smith, asked Knight during a press tour in March.
“Sixteen times,” Smith repeated, before using the metric to illustrate how much wear and tear the job would cause if done with a heavy tool rather than a robot. “Think of the repetitive motion injury in the shoulders, the back. The other thing is pinch points. You could get your fingers in there and have a potential fracture or amputation.”
Knight stopped manually tightening bolts on engines five years ago when GE introduced the robotic tool to the then-brand new Grove City plant, where it sends locomotive engines to be rebuilt and repaired. He still knows how to do it by hand, but he no longer needs this knowledge to do his job. If Knight doesn’t torque the bolts in the exact right sequence, the machine won’t let him continue. If he twists a bolt halfway, the screen turns red and the machine stops. GE also can promise its clients that every bolt has been torqued perfectly. Knight likes it, too. “It’s way more ergonomic, mostly safer, and easier,” he says.
Here’s the problem: Knight’s knowledge about how to do his job manually—his memorization of the star-shaped pattern in which he fastens the bolts and the understanding of how to do so —still has value. It allows him to understand if the machine is making a mistake, and when its process could be improved.“If there’s something wrong, I need to know that”
As GE has automated its factories, its executives say they still value experience like Knight’s, if only because it helps them catch machine errors and, in some cases, better understand how to program machines. “If there’s something wrong in the calibration of this equipment, I need to know that,” says Denice Biocca, GE’s head of HR for supply chain and services. “I need to have some sort of understanding of what [the process] should be in order to know, ‘the machine says it’s tight, but I know it’s not.”
Knight, however, won’t be around forever. Future workers who do his job won’t have his experience and won’t be able to double-check the machines. So how will they be trained?
It’s not just GE or manufacturers that face this potential problem. Repetition and apprenticeships are popular training strategies in many types of businesses with high automation potential, including restaurants, client service companies, accounting firms, investment banks, and law firms.
Chefs typically start out as line cooks, keeping busy with routine tasks like chopping and combining that may be soon be automated, before they master the art of a recipe. Prior to promotion to senior positions, junior lawyers and paralegals spend time on repetitive tasks like combing through documents to predict what might be relevant to a case or drafting simple documents, both tasks some law firms have begun to automate. And junior auditors often take on routine, boring work before they manage a team and meet with clients.
Deloitte CEO Cathy Engelbert says new technology capable of scanning and reviewing thousands of contracts–an entire year of human work–in an hour will mean the firm needs fewer entry-level workers and more workers with experience and judgment. But, “where do [middle-level employees] get that experience and judgment?” she wondered in a May interview with Quartz. “That’s probably the number one thing I worry about as we shift our model.”
Nancy Altobello, EY’s global head of talent, says she expects technology to eat into much of the routine work that EY’s nearly 65,000 annual hires complete during their first years at the firm. Entry-level employees will likely have to handle more complex work earlier in their careers. Training programs won’t have to teach them the automated processes, but they will need to identify and teach skills that they would have learned by doing manual processes over and over again. “How do we make sure that they know how to evaluate and supervise and build relationships and be skeptical and lead complex teams?” Altobello says.
Nobody has an exact plan for how technology will impact future training, because technology itself is constantly changing. But it’s clear that the impact will be great. “It’s moving so fast,” Altobello says. “I have plans to have plans.”
The airline industry has already reckoned with a need to update its training processes for the age of automation. With autopilot technology, most commercial pilots touch the plane’s controls for only minutes each flight—if at all.
But pilots still need manual know-how. While autopilot has made air flight safer, with the number of crashes declining even as the number of flights increases, accidents in San Francisco, the Atlantic Ocean, and elsewhere have been blamed on pilot’s over-reliance or misunderstanding of autopilot maneuvers.
In 2014, the FAA issued a safety alert saying that autopilot could lead to the “degradation of the pilot’s ability to quickly recover the aircraft from an undesired state” if the pilot didn’t practice flying a plane without it.
Pilots have already shifted their training processes. “It was sort of a new way of teaching pilots about autopilot systems,” George Perry, who runs the Air Safety Institute at the Aircraft Owners and Pilots Association (AOPA), told Quartz in an interview last year. “The mindset has to be, it’s a pilot-relief mode, not an airplane-flying machine. You are responsible for monitoring the machine.” He said commercial airplane pilots often practice simulated situations in which they can’t rely on automation, like when autopilot fails, or during an emergency.
With the exception of maybe the trucking and taxi industries, most industries facing new waves of automation today don’t face the same safety issues as airplane pilots. But some of the strategies for adjusting training programs for a more automated world, however, could be similar.
EY, for instance, two years ago added simulated situations into its training programs and plans to increase their role in the future.
In these scenarios, teams composed of workers with different levels of expertise compete to solve a business problem like the ones they may encounter in their jobs. A computer system scores the outcome of their decisions. Workers gain experience for how react in a practice environment (“a very safe space,” as Altobello calls it), rather than a refresher in how to do the process manually, and their learning occurs more quickly than if they were doing repetitive grunt work.
Training may also use the same artificial intelligence that allows automation of repetitive entry-level work.
Consider the art of “chick sexing.” Because female chicks lay eggs, farmers find them more desirable than male chicks. But just after chicks hatch, both sexes look almost exactly the same. Chick sexers sort them. As described by David Eagelman in his book Incognito, a style of chick sexing one Japanese school started teaching in the 1930s was a bit mystical.
“The mystery was that no one could explain exactly how it was done. It was somehow based on very subtle cues, but the professional sexers could not report what those cues were. Instead, they would look at the chick’s rear (where the vent is) and simply seem to know the correct bin to throw it in.”
Students came from all over the world to learn how to sort chicks accordingly. As they practiced the skill, instructors gave a “yes” or “no” feedback until the student’s “brain was trained up to masterful—albeit unconscious—levels.”
But what if a computer could have helped understand rules for chick sexing that the brain can’t articulate? Students wouldn’t have needed to spend all that time repeating the task of sorting chickens to learn it. The learning would be, as Machine, Platform, Crowd: Harnessing Our Digital Future author Andrew McAfee describes it, “turbo-charged.”
There are plenty of more mainstream professions with similar training strategies. “Physicians like pathologists get trained by looking at lots and lots and lots of different slides of tissue,” offers Erik Brynjolfsson, McAfee’s co-author. “Can we speed up that process by bringing machines into it? I think we absolutely could.”
In other words, technology may make status quo training models obsolete, but it’s also likely to make better models possible. Few entry-level employees really enjoy sifting through endless documents or chopping their 100th onion of the day. It’s still unclear what exactly will replace these tasks as training tools, but there’s no law of nature that says dues must be paid in grunt work.