Jobs and employment after AI: reasons for optimism
Richard Zuroff Richard Zuroff
January 16 13 min

Jobs and employment after AI: reasons for optimism

“Will robots and artificial intelligence take all our jobs?”

Or more precisely, “how will the continued automation of physical and cognitive tasks disrupt the labor market?” This is a question we are often asked, and the short answer is that nobody knows for sure, and any forecasts at this stage are likely to be inaccurate. However, we believe there are good reasons to be optimistic about the overall impact of AI, so long as the technology is developed, deployed, and regulated in a human-centric way.

This is because insights about prior waves of automation and careful analysis of current technology can help clarify what might (or might not) be different about the impact that AI will likely have on employment compared to other transformative technologies. In particular, what is missing from pessimistic analyses that focus only on what tasks AI can (or will) be able to do instead of people is a consideration of AI’s secondary impacts. We should consider whether AI might increase job quality and demand for complementary skills that are more widely accessible; deliver real improvements to products and services which increase the purchasing power of workers’ wages; and strengthen the need for policies that incentivize companies to invest in human capital and human involvement in decision-making.

The different ways automation impacts jobs

Most literature on the impact of AI on the labor market focuses first on the question of whether (or when) AI will be able to do certain tasks. Recent scholarship recognizes that jobs are collections of tasks, so the impact of automation will be much broader than the number of jobs that might be fully eliminated. Automating just some of the tasks that compose a job can decrease demand for that job, or create new tasks for that job-holder or others. It can also change the skills that employers will look for in filling a position. For example, while self-checkout machines are meant to eliminate the need for store employees to process transactions, cashiers often spend time helping customers using these machines, doing the emotional labor of soothing angry customers when machines don’t work properly and fixing the machines when they break down.1 Architects and designers who have access to computer-aided design software that use AI to generate plans are unlikely to lose their jobs. Instead, they will be able to design more complex or more imaginative buildings or products more quickly.

These examples reflect the broader pattern that introducing new technologies rarely just eliminates demand for human workers; it also increases demand for workers with complementary skills. For example, electrification created new demand for telegraph workers and electrical engineers to generate and transport power,2 factory workers to use industrial machinery, as well as office clerks and managers3 to process the information generated by more complex supply chains and business models.

In fact, most workplace technologies both substitute for one set of tasks while simultaneously complementing others: medical imaging tools substitute for technicians but complement doctors; in advanced warehouses, robots have taken over the human work of walking with packages, but human workers continue to do the manual “pick and pack” tasks that are very difficult for robots.

Since the beginning of the era of computers and digital technologies, new jobs (titles that didn’t exist in a previous decade) such as “software engineer” have accounted for over half the 50 million or so jobs added since 1980 in the United States.4 Therefore, the overall impact of a technology like AI on the labor market will be the aggregate of job creation, job destruction, and job change. What can history tell us about these different impacts?

Automation, then and now

Historically, the negative effects of substituting human labor with automated technology have been outweighed by the positive benefits. For example, while industrial textile machinery displaced artisanal weavers in 19th century England and caused widespread rural unemployment,5 it also reduced the cost of clothing and other goods and raised living standards all around. Similarly, electric lighting allowed industrial plants to operate in shifts around-the-clock, increasing demand for factory workers while reducing employees’ exposure to the risk of smoke and fire. More recently, McKinsey estimated that the introduction of the personal computer enabled the creation of 15.8 million net new jobs in the United States since 1980, including the jobs displaced.6

In looking back to previous eras, it is also important to distinguish the positive overall impact of automation on employment from its impact on wages and different cohorts of workers, which shows more of a mix of positive and negative effects. The Industrial Revolution caused average wages (adjusted for inflation) to stagnate for decades in England even as productivity rose.7 Eventually, wage growth caught up, but the transition period was difficult and required policy reforms.8 In the U.S., thanks in part to the success of collective bargaining in tying wage increases to rising economic fortunes,9 the period from 1940 to 1980 saw sustained productivity growth that was closely matched by increases in typical workers’ compensation.10 However, since 1980, wage and productivity growth seem to have diverged.11

U.S. Labor Productivity and Compensation Growth, 1948-2016

There are two theories (among many) about why wages have not grown as strongly in the last 20 years that are important to consider when thinking about the potential future impact of AI. One theory points out that the first generation of digital technologies could not replicate the physical dexterity, visual recognition, and face-to-face communications skills associated with manual and service jobs.12 As a result, the previous wave of automation tended to displace middle-skill workers and drag down overall wage growth because the impact was unevenly distributed in the workforce.

A second theory argues that productivity has actually not grown significantly, because many digital technologies performed tasks previously done by workers without necessarily improving the quality of the product or service.13 For every positive example of effective automation (such as automated turnpike tolls substituting for toll collectors) there are also many negative examples to support this theory, like computerized telephone systems that forced users to navigate complex branching menus and were unable to conclusively resolve anything but the most basic customer issues.14 AI and machine learning systems are expected to do better, but it is unclear whether this means we should expect the impact on jobs and labor to feel more like a third industrial revolution, 1940s shared prosperity, or something else entirely.

The likely impacts of AI tomorrow

How much will AI substitute for existing jobs and tasks? Estimates vary, but representative figures for developed economies suggest that approximately a quarter to a third of jobs have a high proportion of tasks that could be automated, a bit more than a third have medium exposure, and around 40 percent have a low proportion of tasks that can be accomplished with technology that is or will soon be commercially available.15

Routine, repetitive tasks are those that are at highest risk of being automated, so these overall numbers can mask significant differences between regions that have different concentrations of job types. For example, cities and companies with high proportions of back-office administration jobs are likely to be impacted more by automation in the near term than centers of research and development. However, as history and theory shows, substitution is only part of the story.

Given the large number of tasks and jobs that could be partially automated, what reasons do we have to be optimistic that the overall effect of AI - including new job creation, job change, and productivity growth - will eventually be positive for the economy and labor market?

First, while workers with lower levels of education have been dropping out of the labor force in developed economies for the last several decades,16 the increasing sophistication of AI will not necessarily exacerbate this trend. The skills that are complementary to modern AI - such as sociability, empathy, and judgment17 - are less correlated with educational attainment than the complements to early computer technologies (such as quantitative reasoning). This suggests that a broader cross-section of the population might have the innate talents that will be valuable in a job market that expects people to work alongside machines. More evenly-distributed workforce demand might reduce some of the labor market polarization that has concentrated wage growth among the most skilled workers while devaluing the majority of work in the recent past.18 Increasing demand for skills and traits like empathy could also increase the quality of jobs, which might be more important than purely growing the quantity of jobs since demographic trends point toward rising labor scarcity in developed economies by the end of this decade.19

Second, modern AI that does displace some human work has the potential to actually improve the quality of the products and services. For example, the AI image processing in camera software allows ordinary people to take nearly professional-looking photos with just their phones; diagnostic medicine systems (like AI-powered skin and lung cancer detectors) can decrease cost and expand patient access; and collaborative robots can safely work alongside people to dramatically accelerate manufacturing. If AI improves the quality of goods and services without significantly increasing their cost, overall living standards will increase and workers (who are also consumers) will get more purchasing power from their wages.20 Also, if policies help ensure companies recognize employees as important stakeholders (not just another kind of intangible asset21), then workers will benefit alongside shareholders when AI allows companies to profit from building and delivering products and services more efficiently.

Third, the actual volume of tasks that will use AI to fully substitute for human activity will likely be significantly lower than the total that technically could be substituted in the short and medium-term. Social acceptance of using AI for tasks that have important legal or economic consequences will require mechanisms to ensure that basic principles of ethical decision-making — such as treating people consistently, fairly, and providing effective remedies when decisions or actions are incorrect or unfair22 — are respected by automated systems. This will require technical progress in areas like AI explainability, fairness and human-centred design23 that might advance more slowly than AI skills such as perception, categorization, forecasting, and optimization, this, in turn, would keep more “people in the loop” of business processes for longer.

Even assuming all the legal and ethical prerequisites are in place, organizations will only substitute AI for human labor when it is economically advantageous. Today, there are many incentives in place that favor investing in capital goods like machines and software.24 But in an era when financial capital is abundant but people’s time, talent, energy, and ideas are especially important to companies’ performance,25 incentives might be rebalanced toward using more human capital. This would reduce the impact of AI displacing workers below the theoretical potential for automation.

Action-oriented optimism

We believe neither large-scale unemployment nor shared prosperity and growth are guaranteed outcomes from the widespread adoption of AI. The anxiety about the changes to jobs and employment associated with transformative technologies like AI is natural, and it can be a positive force if it heightens the focus of technologists, business leaders and policy-makers on the choices that make positive outcomes more likely. The power to choose remains in human hands.

If you’d like to learn more about how AI changes your people or talent strategy, or want help in your AI adoption journey, contact our Advisory team.

Footnotes

  1. Data and Society, “AI in Context: The Labor of Integrating New Technologies” (2019), notes that “[f]illing the gap between shoppers and checkout machines requires a different skill set than that of simply operating a check stand, more akin to that of a traffic officer coordinating vehicles at a convoluted intersection.”
  2. Daron Acemoglu Pascual Restrepo, “Automation and New Tasks: How Technology Displaces and Reinstates LaborJournal Of Economic Perspectives (2019).
  3. Guy Michaels, “The Division of Labour, Coordination, and the Demand for Information ProcessingCentre for Economic Policy Research Discussion Paper (2008).
  4. Daron Acemoglu and Pascual Restrepo, “The Race between Man and Machine: Implications of Technology for Growth, Factor Shares, and EmploymentAmerican Economic Review (2018).
  5. Joel Mokyr, “A review essay of Carl Benedikt Frey, The Technology Trap: Capital Labor, and Power in the Age of AutomationJournal Of Economic History (2019).
  6. McKinsey Global Institute, “Jobs Lost, Jobs Gained: Workforce Transitions In A Time Of Automation” (2018).
  7. Robert C.Allen, “Engels’ Pause: Technical Change, Capital Accumulation, And Inequality In The British Industrial RevolutionExplorations in Economic History (2009).
  8. McKinsey supra note 6.
  9. Frank Levy and Thomas Kochan, “Addressing the Problem of Stagnant WagesComparative Economic Studies (2012).
  10. MIT Task Force on the Work of the Future, “The Work of the Future: Shaping Technology and Institutions” (2019).
  11. Ibid
  12. Ibid
  13. Ibid
  14. Ibid
  15. For example, see estimates for the U.S. in Mark Muro, Robert Maxim, and Jacob Whiton, “How machines are affecting people and placesBrookings (2019). Canadian figures (using a job-based rather than task-based approach) are available from Brookfield Institute “The Talented Mr Robot” (2016).
  16. See for example White House Council of Economic Advisors, “The Long-term Decline In Prime-age Male Labor Force Participation” (2016).
  17. See for example Ajay Agrawal, Joshua Gans, and Avi Goldfarb, “Prediction, Judgment, and Complexity: A Theory of Decision Making and Artificial IntelligenceNational Bureau of Economic Research (2019).
  18. David H. Autor, “Skills, Education, and the Rise of Earnings Inequality Among the ‘Other 99 Percent,’” Science (2014).
  19. Bain & Company, “Labor 2030: The Collision of Demographics, Automation and Inequality” (2018) expects the working age population (25-54) to shrink by 0.4% in advanced economies such as the U.S., Western Europe, Canada, Australia and Japan. See also Daron Acemoglu and Pascual Restrepo, “Demographics and Automation,” NBER (2019).
  20. “Labor productivity increases generally translate into increases in average wages, giving workers the opportunity to cut back on work hours and to afford more goods and services.” See Executive Office of the President of the United States, “Artificial Intelligence, Automation, and the Economy” (2016).
  21. Although the notion that companies should focus exclusively on maximizing shareholder value is popular in the U.S., most other market-oriented economies “recognize employees and communities as legitimate stakeholders to whom a firm must be responsive.” MIT Task Force on the Work of the Future supra note 10.
  22. Ben Green and Yiling Chen, “The Principles and Limits of Algorithm-in-the-Loop Decision MakingProc. ACM Hum.-Comput. Interact 2019
  23. See Lilian Edwards and Michael Veale, “Slave to the Algorithm? Why a 'Right to an Explanation' Is Probably Not the Remedy You Are Looking ForDuke Law & Technology Review (2017) for an argument about the challenges of creating truly effective remedies.
  24. “[Because] the effective tax rate on human capital investments—in the form of labor income taxes— greatly exceeds the tax rate on capital investments… firms [have] an incentive to replace workers with tax-subsidized machinery where possible.” MIT Task Force on the Work of the Future supra note 10.
  25. Michael Mankins, Karen Harris, and David Harding, “Strategy in the Age of Superabundant Capital,” Harvard Business Review (2017).