SEATTLE — In 2015, Stanford University Ph.D. student Michael Webb watched as his classmates in the computer science program became increasingly interested in the advancements of artificial intelligence (AI). An outlier, he was more focused on the economic implications of the technologies.
But there was a dearth of reports that solely focused on the impact AI has on different sectors and occupations. Instead, most analyses generalized technology to include a wide range of automation and software.
Although often used interchangeably, AI and automation are not one in the same: AI technologies are designed to mimic human thinking and actions, while automation performs repetitive tasks.
Then it struck Webb to analyze the overlap between the language in patents — which offer insight into the commercial use of technologies — and the text of job descriptions, which provide a wide-ranging view of the labor market.
Equipped with a blueprint, Webb set about to create a machine learning system that would accurately shed light on the economic impacts of a rapidly changing field. His results were the basis of a new report by the D.C.-based research group Brookings Institution focused on the potential impact of AI on the workforce.
Webb’s findings revealed that workers with higher education and wages will experience the greatest changes in their jobs due to AI, for better or worse.
“While earlier waves of automation have led to disruption across the lower half of the wage distribution, AI appears likely to have different impacts, with its own windfalls and challenges,” wrote the authors of the Brookings Institution report.
Washington state stood out in the Brookings Institution’s state-by-state comparison of the role of AI, which the researchers attributed to the Puget Sound region’s focus on technology and manufacturing.
In the report titled “What Jobs are Affected by AI?,” released last week, Brookings researchers extrapolated on Webb’s statistics and analyzed the impact of AI on various industries, demographics and geographies.
Using over 16,000 patents that contained words describing AI technologies, Webb trained a natural language processing algorithm — a mathematical formula that amounts to a set of processing instructions — to identify thousands of verb-object pairs to quantify the usage of AI in nearly 800 job applications from a U.S. Department of Labor database.
For example, his system would extract the words “diagnose disease” in the description of an AI patent, and then find a similar phrase in job descriptions to determine if, say, a doctor’s task will be affected by the technology.
Webb’s analysis revealed that 740 out of the 769 job posts he analyzed matched with AI patent language, showing that the occupations may be impacted by AI technologies.
Higher paid and educated workers, as well as some agriculture and manufacturing positions will be the most affected by AI, the report noted. The motor vehicle manufacturing and textile industries will experience the most changes due to AI deployment, which is already seen in production lines where AI systems are used to identify defective clothes.
Demographic groups including men, white and Asian Americans, as well as workers ages 25 to 54, will be disproportionately involved with or impacted by AI, the study found.
Webb believes the difference in impact between automation and AI on different demographics is because AI is particularly adept at performing tasks that involve optimization, judgment, and learning from experience. Meanwhile, AI is less skilled at roles that involve human interaction — tasks typically fulfilled by less-educated workers in service jobs.
Will interacting with AI be helpful or detrimental to highly paid workers? That’s hard to say, says Webb: “On the one hand, AI could make them more productive, and increase their wages. On the other, it could reduce employment.”
Mark Muro, one of the report’s authors, stressed that involvement with AI doesn’t equate to job replacement. In fact, higher-wage workers may be the best equipped to weather the changes brought about by AI.
“White-collar workers with higher education levels may have better resources to roll with it and adjust more than, say lower-educated manufacturing workers in the U.S. with fewer skills,” said Muro.