In recent years, there has been an explosion of research into the impacts of automation on work. This makes sense: artificial intelligence and robotics are encroaching on areas of human activity that were simply unimaginable a few years ago.
We ourselves have made contributions to this debate (here, here and here). In The Future of Skills, however, we argue that public dialogues that consider automation alone are dangerous and misleading.
They are dangerous, because popular narratives matter for economic outcomes, and a narrative of relentless technological displacement of labour markets risks chilling innovation and growth, at a time when productivity growth is flagging in developed countries.
They are misleading because there are opportunities for boosting growth – if our education and training systems are agile enough to respond appropriately. However, while there is a burgeoning field of research on the automatability of occupations, there is far less that focuses on skills, and even less that generates actionable insights for stakeholders in areas like job redesign and learning priorities.
There is also a need to recognise that parallel to automation is a set of broader technological, demographic, economic and environmental trends which will have profound implications for employment. In some cases, the trends will reinforce one another; in others, they will produce second-order effects which may be missed when viewed in isolation.
Consider the implications of an ageing population. While much of the public debate on automation has focused on the potential for mass unemployment, it overlooks the fact that robots may be required to maintain economic growth in response to lower labour force participation. The risk, in other words, may not that there will be too few jobs, but that there will be too few people to fill them, which may explain why countries undergoing more rapid population ageing tend to adopt more robots.
What do long-run trends like population aging, urbanisation, climate change and globalisation mean for the future of work?
Will the positive effects on employment of technological progress in all its forms – automation yes, but also biotechnology, the materials revolution and the Internet of Things – offset the negative? What skills, abilities and knowledge will individuals need to do the jobs of the future?
These are the questions that we have addressed in our research on the future of skills for the UK and US economies and that we’re launching today.
The research design starts from three key facts:
First, that – despite disruptive technological change – the occupational composition of the workforce changes only slowly over time, suggesting that looking back at the history of employment is a good starting point for making predictions about its future.
Second, that the US and UK economies are experiencing multiple breaks in long-run trends that theory suggests will have major consequences for employment. The implication is that extrapolation alone will paint an incomplete and potentially biased picture of the future.
Third, that occupations are complex. They deploy a complicated mix of knowledge, skills and abilities, and are performed using a variety of activities and tasks, meaning that our predictive models must be sophisticated.
These three key facts are the backdrop to our study and are what underpins our research design.
Which is why we combine historical trend analysis, expert human foresight and machine learning algorithms, to generate predictions about the future of skills.
In stage 1 of the study, we review in the form of a trends deck what the literature says about long-run trends impacting on the labour market.
In stage 2, we present this trends deck to experts at foresight workshops to debate the implications of structural breaks for future employment. During these workshops, we ask the experts to label a set of 30 occupations according to whether they expect them to rise or fall in demand by 2030. In addition, we ask them how certain or uncertain they are in making their judgments.
In the third stage of the analysis, we use these responses to train a machine learning classifier of whether an occupation will become more or less important in the future workforce.
The classifier makes use of a rich data set of the importance of 120 different skills, abilities and knowledge requirements of occupations drawn from the US Department of Labor’s O*Net database. Our classifier tells us which combinations of these requirements are most associated with occupations predicted to grow but also – crucially – which human capital investments are most potent in boosting the demand prospects of an occupation, given its existing requirements.
What do we find? Rather than making definitive proclamations about the future of some occupations, we conclude that the majority of people (around 70 per cent) are in occupations with highly uncertain prospects
Furthermore, roughly one-fifth are in occupations that are very likely to decline, while one-tenth are in occupations that are very likely to grow.
This uncertainty is a critical dimension to our findings, because it suggests that the future of most occupations is not inevitable: individuals in different occupations can improve their labour market chances if they can invest in the skills that are right for their particular occupation – read our report to find out what these are.
Any reconfiguration of skills (and knowledge) requirements entails an evolution of the occupation. Or put differently, occupations may need to be redesigned in order to make effective use of skills and knowledge complements. Our findings should be a useful guide in this exercise, especially for occupations that have challenging futures
It is also useful to think about new occupations which may emerge in the future. Technically, these occupations correspond to high-demand locations in the skills space (as identified by our classifier) that are distinct from existing occupations. The classifier allows us to identify hypothetical occupations, which were they to exist today, would ’almost certainly’ experience an increase in workforce share and the combination of skills and knowledge variables most associated with them. We hope you find these results as intriguing as we do.
Skills investment must be at the centre of any long-term strategy for adjusting to structural change. A precondition is access to good quality, transparent analysis of future skills needs, as without it, labour market participants and policymakers risk flying blind. The approach we’ve developed is a step towards improving our understanding of this vital agenda and one that invites a more pro-active reaction than the defensive one that has characterised public discussions on automation in recent years. We’d love to hear your comments.