The robot revolution has long been thought of as apocalyptic for blue-collar workers whose tasks are manual and repetitive. A widely cited 2017 McKinsey study said 50 percent of work activities were already automatable using current technology and those activities were most prevalent in manufacturing. New data suggests white-collar workers — even those whose work presumes more analytic thinking, higher paychecks, and relative job security — may not be safe from the relentless drumbeat of automation.
That’s because artificial intelligence — powerful computer tech like machine learning that can make human-like decisions and use real-time data to learn and improve — has white-collar work in its sights, according to a new study by Stanford University economist Michael Webb and published by Brookings Institution. The scope of jobs potentially impacted by AI reaches far beyond white-collar jobs like telemarketing, a field that has already been decimated by bots, into jobs previously thought to be squarely in the province of humans: knowledge workers like chemical engineers, physicists, and market-research analysts.
The new research looks at the overlap between the subject-noun pairs in AI patents and job descriptions to see which jobs are most likely to be affected by AI technology. So for example, job descriptions for market-research analyst — a relatively common position with a high rate of AI exposure — share numerous terms in common with existing patents, which similarly seek to “analyze data,” “track marketing,” and “identify markets.”
It’s more forward-looking than other studies in that it analyzes patents for technology that might not yet be fully developed or deployed.
Typically, estimates of automation effects on the workforce, which vary widely depending on the study, have focused on what jobs could be automated using existing technologies. The findings have generally been most damning for lower-wage, lower-education workers, where robotics and software have often eliminated part or all of certain jobs.
The specter of increased automation has raised concerns about how large swaths of Americans will be able to support themselves when their jobs become mechanized and whether the loss of low-income jobs will increase wealth inequality. This new patent research suggests automation’s impact could be much broader and affect high-paying white-collar jobs as well.
A caveat: Some AI patents might never be used, and they might not be used for their initial intentions. Also, one’s actual job is not wholly defined by the text of the original job description. But this study does provide a framework with which to view general exposure to automation.
As Adam Ozimek, chief economist at freelancing platform Upwork, put it, “Just because someone patented a device, for example, that used artificial intelligence to do market research does not mean that AI will in fact be successful at this for practical business use.”
The Stanford study also doesn’t say whether these workers will actually lose their jobs, only that their work could be impacted. So it’s perfectly possible these technologies will be used to augment jobs rather than supplant them.
This is not the first time white-collar jobs have been in jeopardy from technology. Many white-collar jobs that were supposed to be sent overseas — actuaries, technical writers, and customer service reps — didn’t see the cuts that were predicted more than a decade ago.
And for what it’s worth, market-research analysts seem to think many elements of their jobs can’t be done by using artificial intelligence.
“It can tell us what’s going on, but it can’t tell us why it’s happening,” Gina Woodall, president of marketing research firm Rockbridge Associates, told Recode. “It’s getting better and better at telling us what consumers are doing, but it won’t be able to tell us what’s driving them.”
What market research analysts say
Nearly 700,000 people in the US work as market-research analysts, people who study market conditions to sell products or services, and their job growth outlook for the next decade is much higher than average, according to the Bureau of Labor Statistics. It’s a job that, as its title implies, brings analytic thinking to bear on disparate data and thus is usually safe from talk of automation.
But according to Webb’s research, market-research analysts are much more likely than average to have overlap with AI patents.
Market-research analysts have already begun to contend with AI in their jobs, but so far it’s been used to assist their work or free them to do other tasks.
“We’ve deliberately moved up-market, doing less cookie-cutter simple research because there are a lot of automation and tools that can do that easily,” Woodall, who has been working in market research for 20 years, said. “Where I see our place in the world is focusing on more complex business problems using higher-level analytics.”
Steve King, partner at the small business consulting firm Emergent Research, put it this way: “Our value-add is not data collection or even the first layer of analysis (we mostly outsource these steps). Our value-add is being trusted advisers and an outside source of business insights.”
George Augustaitis, director of industry analytics at CarGurus, previously worked at a company that worked directly with an AI tool called Lucy, which would be able to answer analyst questions based on the data they uploaded to it.
“My team embraced it because we all had the idea that it would be less of the grunt work for us building the charts,” Augustaitis told Recode. “We would spend more time analyzing the data, connecting the dots, being in meetings presenting to clients.”
Webb’s data shows that not every aspect of market research analysts’ jobs fall under AI patents. Those aspects include conducting research on consumer opinions, collaborating with other professionals, and attending staff conferences to brief management on their findings.
Augustaitis said he thinks someday AI could be a “great junior analyst” but thinks such tools would fail when situations were abnormal, like the 2008 recession.
“I don’t know how they would perform when something happens that no one is expecting,” he said.
Jim Loretta, a qualitative-research consultant based in Miami, Florida, emphasized the human element of his job: the importance of conducting surveys and other market research in person.
“How people feel — a lot of that is captured face to face,” he said, referring to subtle human gestures and reactions he sees when, for example, asking a focus group about a new marketing campaign. “I don’t know how [AI] would capture any of those qualitative buttons you would get in a face-to-face meeting.” That may be true now, but the field of AI emotion recognition is improving rapidly.
And for now, artificial intelligence tools are not especially accurate when it comes to making more nuanced decisions.
“Even today you’ll see headlines very clearly written by a computer on Bloomberg and they are wrong a lot,” a hedge fund investment analyst told Recode, referring to how Bloomberg uses automation to help with a third of its content, usually for split-second reports on company earnings. While such technology is good at quickly writing up an earnings miss or a basketball score, it can miss the larger context.
“It’s amazing what an edge you can get just from actually reading every word of a document.”
He also points out that if more jobs could be automated, they would be.
“There’s a reason there’s not more unemployed hedge fund analysts: Because you do still need that human hand, at least for the time being.”