Staying a Step Ahead of the Robots

“Robots Will Take your Job!” the scare headlines tell us. For many, they already have, but robots are only the latest in a long line of job stealers. Early “thieves” were horses for schlepping our stuff and dogs to hunt or herd our livestock.

The theft went on to harnessing water and wind, then steam and electricity, all to make machines turn. Electricity then powered the computers that in 1964 were predicted to cause mass unemployment.^[1] In fact, mass unemployment caused by workplace innovation has never materialized. Why not?

The most obvious reason is that the demand for products and services has always exceeded our productive capacity despite workers becoming infinitely more productive with the help of those job-stealing innovations. Another reason is the shrinking number of hours people work. In the early industrial revolution, people spent about 4000 hours a year toiling in those “dark Satanic mills.” Now in the developed world, it’s only about a third to a half of that. Yes, there were temporary dislocations that hurt many individuals and families, but in the big picture, we did well.

Could this time be different, with its onslaught of ever-smarter robots doing more sophisticated physical tasks ever? What about robotic process automation taking on more and more office tasks? And machine learning doing better than well-paid professionals at interpreting x-rays?

Over the years, people who bet against “this time is different” arguments got a lot richer than those who bet on the difference, but that’s no guarantee. Complicating the current employment picture is the massive export of lower-skill (and increasingly higher-skill) manufacturing jobs to China and other low-wage countries. So while this time may not be as different as the scary headlines suggest, wise people will take action just in case to stay a few steps ahead of the robots.

This article will not try to predict the good jobs to prepare for thirty years from now. (How many forecasters thirty years ago would have identified such currently in-demand jobs as web developers, big data scientists, and image classifiers for training machines?) Instead, it will focus on what makes tasks—not whole jobs—susceptible to being done by robots, robotic process automation, or trained machines.

Obviously, a job that is largely made up of automatable tasks will be vulnerable. The following does not claim to be an exhaustive list of such tasks; more will surely emerge as technology advances.

• The task is repetitive. There may be variations, but the steps can be described unambiguously in detail, like a computer program. It should be no surprise that welding and painting on an assembly line were early and successful targets, along with millions of centralized data-entry jobs, when online real-time automation moved data capture to the source. While there are still assembly lines, their days are numbered; robots will improve their dexterity much faster than human fingers.

• The task is recognizing patterns. An excellent example is recognizing when a skin lesion has a serious chance of being malignant. An experienced dermatologist will have had exhaustive training and developed helpful heuristics from having seen legions of lesions over the years, but human memory is imperfect.
  
  People are also subject to confirmation bias, i.e., the tendency to see what they expected to see instead of what is actually there. By contrast, the machine consistently remembers every detail about hundreds of thousands of lesions and their diagnoses.
  
  Unfortunately, both dermatologists and the machines that help them have been trained on light rather than dark skin^[2] with subsequent less predictive value. Unlike repetitive tasks, pattern recognition should supplement people, making them more effective but not replace them.

• The task is searching. Search is related to pattern recognition but has a different objective. Pattern recognition looks for similarities in vast databases to draw inferences about something, e.g., is the lesion likely malignant? Search tries to match text or images—the search argument—as precisely as possible to one or more items in a database.
  
  Text search also differs from pattern recognition in its need for human skill in fashioning sophisticated Boolean search arguments that maximize relevant hits while minimizing the chaff.
  
  Take a legal search for relevant case law as an example. When done by humans, the process is sufficiently tiresome to cause occasional mental drift—naturally at the wrong moment. Likewise, facial recognition has some success in helping law enforcement locate suspects but has become highly controversial because accuracy varies widely depending on race.
  
  Although search automation suggests we will need fewer law clerks or detectives, we must maintain a perspective that automation’s role is to supplement people, making them both more effective and efficient.

• The task is evaluating applications. Algorithms can easily make decisions to approve or deny a loan or life insurance application. Here the danger lies in overuse when the decision is about something of potentially life-changing significance to the applicant –a small business loan, home mortgage, or college admission.
  
  When the algorithm’s result is a close call, human intervention is critical to supplement the automation to make it more fair, humane, and ethical. With proper feedback, human overrides help improve the algorithms. (Our blog post “Don’t Let Your Computer Get Above its Pay Grade” deals with this topic in more detail.)

It’s important to note that a job’s status is not entirely correlated with its ability to be automated. Dermatologists and radiologists are highly trained specialists; their expertise in pattern recognition and interpretation lets them enjoy high status and a good income.

By contrast, strawberry pickers enjoy neither. It’s stoop labor in the broiling sun on fields that seem to go on forever, at least in California’s Central Valley. Yet finding and recognizing strawberries ready for picking and picking them without damaging them has proved devilishly difficult for robots, though attempts are ongoing^[3].

The other aspect of robot-proofing one’s career is developing widely applicable personal skills, the skills that make us human, that are extremely difficult if not impossible for computers to learn and that are of value far beyond the job market. They fall into four categories: empathy, collaboration, creativity, and thinking.

Learning these skills would ideally start in primary school. None of the examples below are amenable to standardized testing, and none should be targeted only to the college-bound.

Empathy:

• Exercises to build empathy and understanding of others’ motivations and behavior
• Role-play exercises where assigned roles are not natural to the player
• Community good deeds and works of charity
• Defusing tense interpersonal situations

Collaboration:

• Working collaboratively on team projects, particularly where there is a choice of approaches
• Finding common ground among competing ideas
• Clear, persuasive communication of one’s thoughts, orally and in writing
• In the extracurricular realm, sports like basketball and soccer where the roles are fluid and the state of play unfolds by the second
• Also extracurricular, culture like drama, improv or jazz where one has to sense and react to others more than just doing one’s own part

Creativity:

• Appetite for problem-solving, particularly where there is no single correct solution, e.g., case studies
• Improvising with what you have on hand—materials or information—to achieve a goal when the usual means are unavailable, and a deadline looms

Thinking:

• Curiosity about the “why” of things, both natural phenomena and how people behave individually and in groups
• Critical thinking
• Basic logic and common fallacies
• The debating team, especially being called upon to support the side you may personally disdain.

It is hard to imagine a workplace (or any social system or situation) in which these skills would not be helpful, recognized, and rewarded.

Besides the usual academic subjects, the curriculum should include, again for all K-12 students:

• Computer-related basics—what they’re good at and why, plus elements of coding, website development, office suites
• Basic concepts from statistics—not theory—and how to interpret graphical presentations of data and recognize those designed to mislead
• Personal financial management
• Personal time management

We should not try to prognosticate in detail how the workplace will evolve, but it is safe to say that “no brainer” jobs will dwindle away. It is inevitable that when a computer or robot becomes cheaper to use than a person, it will be “hired.”

That’s potentially bad news. The good news is that most people in lower-level jobs are overqualified for them. They have cognitive and social skills that could be valuable in the future workplace – however it evolves. The challenge is to create an education system that helps young people find what they care about and are good at and develop those uniquely human skills for a promising career and work life.

We need everybody’s talents. We cannot tolerate the social consequences of massive unemployment and underemployment, as we see in parts of the country. The goal is to launch K-12 graduates, whether proceeding to higher education or not, with capabilities that will serve them well through life, not just in the workplace but also as householders, consumers, parents, community members, and, not least, responsible citizens.

[1]“Technology and the American Economy”, Report of the National Commission on Technology. Automation and Economic Progress, US Department of Health, Education and Welfare, Office of Education, February 1966 ↑

[2]“Dermatology’s Skin Color Problem”, New York Times, S3ptember 8, 2020 ↑

[3]“The Age of Robot Farmers” by John Seabrook, The New Yorker, April 15, 2019. ↑