To stay up to date with new research and commentary on issues relating to the American middle class, economic mobility, and inequality, sign up for Class Notes—a twice-monthly newsletter curated by the Future of the Middle Class Initiative.
The long-run labor market consequences of advances in artificial intelligence (AI) and other forms of automation technology remain uncertain. This uncertainty stems from a recognition that while many jobs will undoubtedly be affected, it is difficult to predict which jobs are at risk and in which sectors new jobs will be created. This is not an idle concern: One in six Americans think that robots and computers will take over many jobs now done by humans; only 25% believe that automation will bring new, better paying jobs. And the estimates of jobs at risk are quite variable.
For example, a 2017 McKinsey Global Institute study suggests that roughly 50% of work activities are automatable using current technologies. Another study by Frey and Osborne (2013) argues that 47% of U.S. employment is at high risk of automation. By contrast, a study by Arntz et al. (2017) suggests that only 9% of individuals have jobs that are at high risk. Of course, as we have noted previously, even if the share of jobs at risk is quite high, other barriers such as adoption or transition costs for firms potentially limit the speed and severity of displacement. Regardless, employers and policymakers need credible information to make good decisions, especially when seeking to mitigate harm when worker displacements occur. To understand better where the difficulties arise in predicting the future of work, it is important to revisit both the promise and peril of automation technologies and AI.
Tasks, jobs, or occupations? Difficulties in predicting automation impact
Forecasts of how exactly AI and related automation technologies are expected to affect the job market depend on understanding fundamentally what a job is. A job is just a collection of tasks. Many tasks associated with jobs can be performed by either humans or machines. Consequently, these tasks are potentially subject to partial or full automation. Thus, researchers have focused on characterizing the task content of occupations when trying to determine whether the job is at risk. For example, introducing robots to factories and ports reduced the need for human brawn as robots are stronger, need little rest, and can work around the clock. One study focusing on international investment in industrial robots suggests that one robot was equivalent to six workers. The advent of more sophisticated AI has extended these trends into white-collar settings. You’ve probably visited a bank to deposit or withdraw using a human teller to complete the transaction. Tasks performed by tellers are now almost fully automated; depositing a check often requires only taking a picture.
Indeed, Frey and Osborne (2013) study the likelihood of a job being automated and predict that tellers are among the workers in transportation, logistics, service, and office and administrative occupations—many of the occupations that constitute middle class jobs—that are at the highest risk of being replaced by technology. Overall, they conclude that 47% of total employment in the United States is at “high risk” of automation (i.e. in an occupation with at least a 70% chance of automation) in the next decade or two.
Arntz et al. (2017) dispute the high levels of automation risk predicted by Frey and Osborne (2013), appealing to the considerable variance in tasks performed within occupations. Manyinka et al. (2017) essentially conclude that while tasks are highly automatable, only roughly 5 percent of jobs are fully automatable at present. Nevertheless, all these papers agree that less educated workers in low-skill, lower-wage jobs featuring routine tasks are those most likely to be displaced by automation.
Ultimately, the biggest questions surround advances in AI. According to Frey and Osborne (2013), existing engineering bottlenecks limit the ability of computers to perform jobs that require human perception and manipulation, creativity, and social intelligence. While tellers have a near 100 percent likelihood of being displaced by computers, the likelihood that recreational therapists (e.g. music or art therapists) will be replaced in the same manner is quite remote, mainly because the job requires human perception and social intelligence skills that are not currently replicable by machines. While machine learning algorithms and other forms of AI can win games and mimic many aspects of human interaction, it remains to be seen how far these technologies will be able to go.
broad agreement that automation will have an impact
Although the job risk estimates vary, economists agree that automation and artificial intelligence technologies will continue to transform the nature of work. Some workers will lose their jobs to automation, others will get new jobs, and many will need to acquire new skills to transition across occupations. Given these changes in the structure of society, policymakers, private sector organizations, and members of the business community will need to partner to develop consensus on the appropriate path forward.
AI and automation pose a significant challenge to middle-class work and wellbeing. Our work in this space will explore several key questions including: what can we learn from industries recently affected by automation, what are the best policies to help displaced middle class workers make successful transitions across occupations, and what potential equity issues might arise from the structural changes in the labor market? In the coming months, the Future of the Middle Class Initiative, through a series of public and private conferences, research papers, and podcast conversations, will begin to address these questions and more.