Will Automation Take Your Job? What History and AI Suggest

Predictions about machines replacing human workers are as old as the industrial revolution, and they have been both right and wrong. Automation has eliminated whole categories of work, from switchboard operators to typesetters, while creating new jobs that nobody could have predicted. With artificial intelligence now capable of writing, summarizing and coding, the question has returned with new urgency. The research offers a more nuanced picture than the headlines.

Lesson from history: ATMs and bank tellers

When automated teller machines spread across the United States from the 1970s onward, many expected bank tellers to disappear. Economist James Bessen found that the opposite happened for decades: ATMs reduced the number of tellers needed per branch, which made branches cheaper to run, so banks opened more of them. Teller employment grew for years, and the job shifted from counting cash toward customer service and sales. More recently, as online and mobile banking spread, teller numbers have fallen. The story shows that automation often changes jobs before, or instead of, eliminating them, and that the effects can unfold over decades.

Jobs versus tasks

Most jobs are bundles of tasks. Automation usually takes over specific tasks rather than whole occupations. A widely cited 2013 study by Oxford researchers Carl Benedikt Frey and Michael Osborne estimated that 47 percent of US jobs were at high risk of automation. Later analyses that looked at the tasks within jobs, including work published by the OECD in 2018, produced much lower estimates: about 14 percent of jobs across member countries at high risk, with another 32 percent likely to change significantly. The difference highlights how much depends on whether you count whole jobs or the tasks inside them.

What generative AI changes

Earlier automation mainly affected routine physical and clerical work. Generative AI reaches into tasks once considered safe, such as writing, analysis, customer service and software development. A 2023 study by the International Labour Organization concluded that the most likely effect of generative AI is augmenting work, automating some tasks while leaving people to do others, rather than replacing jobs outright. It found clerical work to be the most exposed category, and noted that women are overrepresented in those roles in many countries.

Early workplace studies show productivity gains. A study of customer support agents published by researchers from Stanford and MIT found that access to an AI assistant increased the number of issues resolved per hour by about 14 percent on average, with the largest gains, around a third, among the least experienced workers. Other studies have found that AI can make work faster but also introduce errors when people rely on it for tasks beyond its capabilities.

Which work is most and least exposed

More exposed: routine, predictable tasks, whether physical, such as some assembly and warehouse work, or cognitive, such as data entry, basic bookkeeping, standard writing, scheduling and first-line customer support.
Less exposed: work requiring physical dexterity in unpredictable environments, such as plumbing, electrical work and nursing care; complex social interaction, such as teaching, counseling and negotiation; and judgment in ambiguous situations.

Exposure is not the same as job loss. Whether automation replaces workers or makes them more productive depends on costs, regulation, how employers choose to use the technology and whether demand grows as products and services become cheaper.

New jobs

Economist David Autor and colleagues have estimated that a majority of US workers today are employed in occupations that did not exist in 1940. Each wave of technology has created roles that were hard to imagine beforehand, from web developers to social media managers. Roles related to AI itself, such as building, testing, overseeing and applying it, are already growing.

What workers can do

Learn the tools. People who use AI and automation well in their field are likely to be in a stronger position than those who avoid them.
Build skills that complement machines, such as judgment, communication, problem-solving, leadership and expertise in a specific domain.
Keep learning. Short courses, certifications and employer training programs make it easier to update skills throughout a career.
Watch which tasks are changing in your own job, and move toward the ones that require human strengths.

What employers and governments can do

The transition costs fall unevenly, often on workers in specific industries and regions. Retraining programs, apprenticeships, portable benefits and support for displaced workers can soften the impact. How the gains from automation are shared, between companies, workers and consumers, is ultimately a policy choice as much as a technological one.