There’s no doubt that job automation will disrupt the labor market. But while most of us talk about when, Iyad Rahwan is focused on a different question: Where?
Rahwan, associate professor of media arts and sciences at the MIT Media Lab, started digging into that question by constructing a geographic map of how automation will impact the job market. He used data from a controversial study from Oxford University on current jobs that are susceptible to technological advances. “We don’t take those numbers as a kind of absolute prediction, but we use them as a relative comparison,” Rahwan said in his talk on AI and the future of work at the recent EmTech Conference in Cambridge, Mass.
The data enabled him to look at how the job market, say, on the East Coast will be impacted compared to the job market in the Midwest. Perhaps more importantly to CIOs, it provided a launching off point for Rahwan to construct a different kind of map — a “skill-skill network” where the specific skills and tasks making up job titles are compared for similarities. The map is a window into where the labor market is and where it’s going.
Big cities vs. small cities
In general, Rahwan’s findings indicate bigger cities will be more resilient to job automation than smaller ones. That may seem a little like a no-brainer, but he said the conclusion “wasn’t a priori” at the outset. Bigger cities tend to be better than smaller cities at attracting talent with a range of skills, but bigger cities are also better at breaking down complex jobs into explicit tasks, making them ripe for automation, he said. The findings gave credence to what many might assume — and it provided a jumping off point to ask another question: What is it that makes bigger cities more resilient to job automation?
Rahwan used bottom-up clustering to categorize job titles, creating five buckets. One bucket contains of network administrators, engineers and so on. Another bucket includes telemarketers, cashiers and so on. He compared job types in the largest 50 cities to the smallest 50 cities in the United States and found that bigger cities have more positions such as graphic designers and financial analysts than smaller cities and fewer retail or cashier-type positions.
Indeed, jobs like these that require creativity or the ability to manage people or machines grow disproportionately to city size unlike other job types, a finding Rahwan called “striking.”
“That was the first sort of finding,” he said. “Now we want to understand, within a city, what is the structure of this problem.”
Job skills vs. job titles
Specifically, Rahwan was interested in digging into the “hallowing out of middle class jobs,” which are stagnant.
Rather than talk about high-skill jobs versus low-skill jobs, Rahwan wanted to pinpoint specific job skills and find the overlap between job titles. He mapped jobs into a skill-skill network, where jobs are linked based on similarities — the more similar the skillset, the stronger the link.
“For instance an accountant needs to know math and also needs to know a bit of programming and a whole bunch of other things,” he said. “An insurance underwriter probably also needs both. So accountants and insurance underwriters are probably more similar because they rely on similar skills.”
His findings suggest that job skills can be mapped into two core groups connected by a bridge. One core group includes skills such as physical skills, sensory skills, multi-limb coordination, finger dexterity. Skills in this grouping are more susceptible to automation, are less cognitive in nature and tend to correlate with a lower education level and lower wages.
A second core group includes skills such as social skills, managerial skills, negotiation and persuasion. Skills in this category are less susceptible to automation, are more cognitive in nature and are correlated with higher education levels and higher wages. The bridge is a set of common skills such as basic memorization and basic numeracy.
If employees want to move from one grouping to another, they may struggle because the job title they have and the job title they’re pursuing may lack complimentary skills. Indeed, based on his findings, there appears to be a group of people who are “stuck,” who have gained some cognitive skills but not enough to make the transition from one group to the other or to see a median wage increase.
Helping employees make the transition “is where the real challenge is going to be,” Rahwan said. Next, he’s planning to look at how the skill-skill network changes over time, which he’s hoping will help determine what skills come into demand and what skills will be automated out of the network completely.