It’s the question on many workers’ minds: Will I lose my job to artificial intelligence?
New research from Northeastern University, however, finds that evaluating AI’s impact on the labor market involves much more than analyzing government unemployment statistics. The researcher says measuring the impact of AI on the workforce must take into consideration that the technology is changing jobs by redefining skills and tasks, and not just eliminating positions.
As a result, researchers are developing a skill-based assessment that predicts a worker’s likelihood of being replaced by AI.
“We need to understand that the impact of AI on the job market is not just at the end of a job when you get displaced,” says Esteban Moro, a professor of physics at Northeastern who is affiliated with the university’s Network Science Institute. “You can get your job skills or job tasks redefined. You can get a totally different job. You can move from one job to another within the same company. You can stay in the same job, but you can do many more things or things much faster.”
“This is why we need to go beyond the data which is collected right now and start collecting much more data about skills within jobs, how the skills that compose a job are redefined and are changing,” Moro continues. “All of our industries are affected by AI, but within the aggregated data that we’re using right now, I think we are missing most of the changes.”
With AI cited for job losses in the last year and “shrinking workforces” in the future, it appears that the technology is changing the workforce. Indeed, predictions of AI’s impact have been dire, with historical models predicting nearly 40% of certain jobs will disappear and 50% of all U.S. jobs at risk from AI.
But in a recent study published in the journal PNAS Nexus, Moro and colleagues at the University of Pittsburgh and Indian University evaluated these historical models and compared them with unemployment data from different sectors, each state, and over a period of time.
“What we found in this paper is that none of those doomsday predictions were accurate. They didn’t happen,” Moro says.
Moro cites radiologists as an example. When AI was first used to analyze X-rays, the prognosis for future radiologists was grim. That hasn’t come to fruition.
“The number of radiologists in this country increased in the last 10 years,” Moro notes. “(Reading X-rays) was automated, but the actual job of the radiologists is not only that skill, it’s a lot more.”
But taking all the different models that predicted job loss and applying them together revealed distinct aspects of how automation affects unemployment, Moro explains.
By considering your job as a set of skills, researchers can measure an individual’s “unemployment risk” — or a measure of the potential of unemployment due to AI. The more of your skills that can or will be automated, the higher your unemployment risk.
“That doesn’t mean you’re going to be displaced,” Moro stresses. “We can adapt, we can pivot and do something else, or companies and universities can train people in new skills.”
Moro says that he and researchers at institutions including Carnegie Mellon and MIT are building the Observatory of US Job Disruption to collect more data on job skills — from resumes, job descriptions and job postings, for instance — to make measuring unemployment risk even more accurate.
He envisions a future website where someone can plug in their job, sector and location to find their unemployment risk.
“We have to go farther and farther, which means more data, more analysis and more resources,” Moro says. “The only way to understand and act on what is happening is to measure it properly.”