Skip to main content

What Past Waves of Automation Can Teach Us About AI

Policymakers and the public are grappling with the rise of artificial intelligence and its potential impacts on workers. Some fear massive job losses from automation while others argue these concerns are overblown, instead highlighting the potential for new job creation. Debates and anxiety around automation and jobs are not new and have occurred throughout history. In an essay from 1930, famed economist John Maynard Keynes stated, “We are being afflicted with a new disease of which some readers may not yet have heard the name, but of which they will hear a great deal in the years to come—namely, technological unemployment.” History can provide some guidance about how automation has impacted workers in the past and help us understand what is similar and different about this new wave of automation.

What Makes AI Different from Past Waves of Automation?

Two past waves of automation can provide useful insights: the Industrial Revolution and the Information Technology Revolution. The Industrial Revolution saw innovations like steam and electric engines automate many tasks that required physical strength and dexterity and followed an explicit set of rules, such as drilling and sewing. The Information Technology Revolution later allowed computers to automate many analytical tasks that also followed an explicit set of rules, such as sorting files by date and calculating numbers in spreadsheets.

AI is different in that it can automate many tasks that do not follow an explicit set of rules and are learned through experience and intuition, such as recognizing images, drafting emails, and drawing illustrations. Modern AI mimics human intuition and learning using statistical techniques that effectively make predictions using data. For instance, an AI system can use a database of images to predict what a new image represents.

These distinctions between modern AI and technologies developed during the Industrial and IT Revolutions are important for two reasons. First, if we define routine tasks as those that follow an explicit set of instruction, AI tools focus on non-routine tasks unlike the previous waves of automation. Drafting an email is a common task that some might consider routine; however, it is non-routine in that there is no explicit set of instructions that someone can follow to draft an original email. Further, some tasks that AI automates are not routine by any meaningful standard. For instance, drawing an elaborate image is a non-routine creative task that requires considerable skill and experience which was not possible to automate until the rise of generative AI.

Second, while today’s AI systems can substitute for intuition and human learning in some non-routine tasks, they are not good at many others. For instance, today’s AI systems can suffer from “hallucination” problems that make them less reliable when giving factual advice. These shortcomings warrant a case-by-case approach to determining the efficacy and accuracy of AI technologies.

What Impact Did Past Waves of Automation Have on Jobs?

The past also provides precedent for the impacts of automation on the number and type of jobs across the labor market. If history is a guide, we can expect to see a mix of job loss, creation, and modification from AI. Historically, it is less common for technology to fully automate an entire occupation. The elevator operator is a rare example of a major occupation that was effectively fully automated.

Partial automation of an occupation tends to be more common as new technologies emerge. For instance, the introduction of medical devices to monitor patients’ vital signs automated some of the work done by doctors and nurses, which gave them more time to focus on other tasks. Partial automation can make workers more productive and augment their capabilities, but its impact on the number and types of jobs in an occupation can be ambiguous.

The agricultural and financial industries provide reference points for this. In the 1800s, two-thirds of the United States labor force was employed in agriculture. A wave of automation over the next several decades augmented farmers’ productivity, but corresponded with a dramatic decline in the share of agricultural jobs in the economy. Fortunately, new jobs were created in the emerging manufacturing industry to make up for job losses in farming. More recently, the financial industry underwent a wave of automation from the 1970s through the 2000s due to the rise of the automated teller machine (ATM). Counterintuitively, bank teller jobs increased as a share of the labor force as the role was modified to focus on other tasks—such as relationship management rather than handling cash—and the cost of operating banks decreased, leading to more employment opportunities.

Final Thoughts

History does not always repeat itself but can provide some guidance, and understanding the differences between past waves of automation and AI is helpful in shaping our expectations for the future. AI is different in large part because it can automate certain tasks that rely on intuition and learning through experience that do not follow an explicit set of rules. This suggests it will impact different types of work and workers. History also suggests the impact AI will have on jobs will be complicated, but it will likely result in a combination of job creation, loss, and modification as AI becomes more commonplace across industries.

Read Next

Support Research Like This

With your support, BPC can continue to fund important research like this by combining the best ideas from both parties to promote health, security, and opportunity for all Americans.

Give Now