There is intrigue over what the future of work will look like. In 2017, a well-known study by Frey and Osborne of Oxford University employed – what else? – a machine-learning algorithm to calculate how susceptible 702 different kinds of jobs in the United States could be to computerisation. The study was intended to be more thought exercise than forecast: while the authors conclude that up to 47% of all jobs in America could be susceptible to automation using current technologies, they are candid about the study's limitations[1] and note that it is not intended to be taken as a prediction of the future. However, it helps to give an idea of what computer-controlled technology could feasibly do in the workplace.

A new working paper from researchers at the OECD builds on the previous work of Frey and Osborne and applies it across 32 OECD countries. The OECD paper evaluates the automatability of each task within a certain job, based on the 2015 Survey of Adult Skills (PIAAC). The results of the OECD study hold some important takeaways for policy makers, namely that the future of work may depend on where a worker lives, the stage of their career, and the level of skills they have.

Takeaway #1: Risk of Automatability Varies Across Country and Occupational Mix

Overall, the study finds that 14% of jobs across the 32 OECD countries are highly vulnerable to automation, defined as having at least a 70% chance of being automated based on current technological capabilities. An additional 32% of jobs are “quite” vulnerable to automation (probability between 50% to 70%). At current employment rates, that puts some 210 million jobs at risk across the 32 countries in the study.

But the variation in the likelihood of automation by country is large: one-third of all jobs in Slovakia are highly automatable, compared to only 6% of jobs in Norway. More generally, jobs in Northern European countries (Norway, Finland, Denmark, UK, the Netherlands), North America (US, Canada) and New Zealand are less automatable than jobs in Eastern and Southern European countries (Germany, Slovakia, Italy), and Japan.

Figure 1: Cross-Country Variation in Job Automatability,
Percentage of Jobs at Risk by Degree of Risk

Why is this the case? Differences in job mix and in industrial structure play a role, but the first matters more. A higher risk of automatability stems from the fact that some countries have relatively larger shares of jobs in manufacturing, but also from differences in the job content within nearly similar industries and occupations. The OECD study finds that over two-thirds of the variation across countries is related to differences in how national economies organize work within the same industry sectors (i.e. occupational mix within sectors and the mix of tasks within occupations), with the remaining one-third explained by differences in the diversity of industries across countries.

For example, in South Korea about 30% of jobs are in manufacturing, compared to 9% in Canada. Nonetheless, on average, the paper finds Korean jobs harder to automate than Canadian jobs. This may be because Korean employers have found more productive ways to combine routine tasks (which can be automated) and social/creative ones (which, so far, computers have difficulty producing) within the same job.

Takeaway #2: The Skill-Bias

The OECD study concludes that the risk of automation declines with higher levels of education, skills (as measured by PIAAC’s numeracy and literacy scores) and wages in most countries. The report cites evidence to support the hypothesis that artificial intelligence and automation affect lower-skilled jobs more significantly than previous waves of automation, suggesting there is a skill-bias. In the same vein, the authors do not find that AI or automation have had a measurable negative impact on jobs characterised by high levels of education or skills, or that relied heavily on non-routine cognitive or dexterous tasks.

Figure 2: Mean Probability of Automation by Occupation

Source: OECD, Survey of Adult Skills (PIACC), 2012, 2015.

Takeaway #3: The “U” Shape and It’s Impact on Youth Employment

Another notable finding from the report is automation’s “risky” relationship with age: it’s U-shaped. More precisely, jobs at the highest risk of being automated are entry level positions often held by young adults. The risk of jobs being automated declines and then bottoms out around middle-age, and climbs back up again in the years before normal retirement, as many older workers downshift into part-time, less demanding roles. Although this U-shape can mostly be explained by the sorting of youth into entry-level positions in sectors like retail and food services, the findings also underscore that a younger person is more at risk of losing a job from automation than a more mature adult worker, even if they both hold the same occupational title.

The wider implication is that automation may disproportionately hurt younger workers. The combined impact of increased immigration, rising minimum wages, and more older workers staying in the workforce may put downward pressure on the number of entry-level jobs. The OECD warns that policymakers should pay close attention to falling youth employment rates in the context of ongoing automation.

Takeaway #4: The Importance of Work-Integrated and Life-Long Learning

The OECD reports that across the 32 countries, there is strong evidence that upskilling and continuous learning can help workers transition from more to less automatable jobs. The paper stresses the importance of work-integrated learning within post-secondary studies, and calls for more policies and supports aimed at encouraging life-long learning in adults.

Takeaway #5: Don’t Forget About the 14%

Overall, the OECD working paper indicates that job “destruction” as a result of automation may not be as extensive as some analysts suggest. And while the estimate of automation risk varies across the 32 countries, the picture that emerges in terms of who is most affected is broadly similar. From a policy perspective, it is crucial not to ignore the retraining and upskilling needs of the 14% of workers who are deemed to be highly vulnerable to automation. Most people in this group currently receive little retraining and may face barriers to upskilling – including lower basic numeracy and literacy skills. In the same vein, the large share of workers (32%) whose jobs are likely to experience significant shifts in their content/tasks because of automation sends a signal to policy makers to be alert to the unequal distribution of automation risk across industries and organizations.

---