The backdrop of rising unemployment has made urgent the need to understand how workers' skills and experiences can be transferred between jobs. In our new report, Mapping Career Causeways: Supporting workers at risk, we have used machine learning to create a map that captures the similarities between over 1,600 jobs, based on the skills and work activities that make up each role.

In the map, similar occupations are placed close to each other, showing how a worker's skills and experience can be transferred to a range of nearby jobs. These jobs can then be compared along different dimensions, such as their average earnings, risk of automation and potential exposure to COVID-19.

Over the next year, these data-driven measures of occupational-similarity will be verified and tested. Ultimately, we hope that this type of map could help to broaden career horizons and tailor skill recommendations.

A map of the labour market

The map, shown below, captures the relationships between over 1,600 jobs. The closer two occupations are to each other, the greater the overlap in their skills and work activities.

Overlaid on the map is a measure that reflects potential exposure to automation risk. These estimates of risk are derived from the results of a study by Brynjolfsson, Mitchell and Rock (2018), in which the authors used crowdsourcing to rate thousands of tasks on their suitability for machine learning. We translated these estimates to a European set of occupations (by developing a crosswalk between the O*NET and ESCO frameworks).

Whilst these estimates of task automation represent the most detailed ones available, they are still experimental in nature and the translation of these findings to a European context adds additional uncertainty. The complete methodology is detailed in the report, where we also consider the role of 'bottleneck tasks' that are particularly difficult to automate. An occupation may have high overall suitability for machine learning but a handful of bottleneck tasks may slow or even prevent automation.

As shown in the map above, the sales and services sector and the business and administration sector both contain many jobs whose tasks are judged to be suitable for machine learning. The actual likelihood of automation will depend on many other factors, and there may be cultural, societal or financial barriers that prevent or slow automation. Moreover, innovation in job design may ensure that the adoption of machine learning augments these roles instead of displacing them. The estimates of suitability for machine learning will also continue to evolve as machine learning advances. For example, progress in the design and adoption of robotic technology would raise the risk of automation for a number of manually-intensive roles in the construction sector.

The cluster of higher-risk jobs across the map shows that, if workers in these jobs are displaced by machine learning systems, they may find it hard to transition into lower-risk jobs that offer at least a similar level of pay to their current job (which are called 'desirable transitions'). The reason is that many of the desirable transitions for these workers would put them back into jobs whose tasks appear suitable for machine learning. These workers face the seemingly impossible challenge of finding a job that is sufficiently similar to their current role so that it is a viable transition, but also sufficiently dissimilar such that it is less suitable for machine learning. The groups that have the smallest number of desirable transitions into lower-risk roles include sales and services and a subset of clerical business and administrative occupations.

Using the map to provide guidance

In the example below, we show how the map might be used to provide insights about the role of a shop assistant. This job contains a number of tasks that would be suitable for machine learning. It has also been severely impacted by COVID-19. Using the map, we can identify alternative jobs that require similar skills and work activities. We then compare these 'viable transitions' along dimensions such as earnings and employment as well as risk of automation and potential exposure to COVID-19.

Tailoring advice on upskilling

The map can be used to pinpoint the potential skill gaps between a worker's current job and their desired job. We also can identify the skills that would have the greatest broadening effect on a worker's options. For workers in jobs that have high suitability for machine learning, we find that there are four broad skill types that would increase their 'safe and desirable transitions'. The most effective of these core skills can unlock, on average, between two and three new options per occupation. The four groups are:

• Management skills to manage staff, budgets and projects;
• Communication skills to build and maintain business relationships, use different communication channels and liaise with managers and authorities;
• Information analysis and evaluation skills to execute feasibility studies, assess financial viability, analyse risk and perform research;
• Compliance knowledge to comply with company guidelines, work health and safety standards and environmental legislation.

Taken together, our observations underscore the role of non-routine activities requiring advanced cognitive reasoning, human judgement, and working with other people, in protecting workers against automation risk.

Partner with us

Over the next year, we aim to validate and trial the map of career transitions, and we are actively seeking partners to work with us. This work will involve seeking feedback on the transition pathways, testing different methods for delivering insights about career transitions, and enriching the information that can be overlaid on the map. Through this work, we aim to broaden the information that is available to individuals, businesses and public services, and thereby drive change that helps to connect people to good work.

If you are interested in working with us, or supporting our research, please get in touch by emailing [email protected].