• Opinion
  • October 22, 2019
  • 10 minutes
  • 0

Can you trust an algorithm to map your career path?

Opinion: The future of work is changing, but no-one can predict how

This article is written by Sarah Doyle, Director of Policy and Research at the Brookfield Institute for Innovation and Entrepreneurship and Hasan Bakhshi, Director of the Creative Industries Policy and Evidence Center (PEC); and the Executive Director of Creative Economy and Data Analytics at Nesta.

We’re told that automation means that many jobs and skills that are important now will not be so ten years from now. 

In fact, many forces, from AI to climate change and the aging of the population, are reshaping the jobs that will be available down the road, but in ways that are difficult for us to predict. Governments everywhere acknowledge the need to invest in new skills for workers in vulnerable jobs, but which skills do they need? 

While it’s clear that no one has the tea leaves or crystal balls required to paint an accurate picture of the future, forecasting can be a helpful tool for considering the range of possibilities, thereby enabling more robust decisions on reskilling to be made. 

Many forecasts, however, assume that historic trends will continue. 

This fails to take into account the possible impacts of extraordinary disruptions, resulting from, say, breakthrough technologies, changes in the natural environment, or shifts in policy. 

The Brookfield Institute has teamed up with Nesta —  a leading innovation foundation based in the UK — to develop long-term skills forecasts for Canada (forthcoming in early 2020), combining predictive modeling with expert insights, which take into account how a wide range of changing trends might interact to shape the future of work. In doing so, we aim to identify the skills that will better equip us to navigate uncertainty. Our goal is to help employees, employers, educators and governments in Canada make more future-proof skills decisions.

Who knows what tomorrow holds?

Our approach starts from three insights. 

First, there are clear trends in the workforce composition of developed economies, such as a rising share of management occupations and declining share of production jobs in the US, but such changes are gradual. This suggests that looking back at the history of employment is a good starting point for making predictions about the future. 

Second, the workforce in all countries is exposed to new sources of disruption, such as the effects of climate change and the AI revolution, meaning that simple extrapolation of the past paints a distorted picture of the future. 

And lastly, occupations are complex. Even seemingly straightforward jobs, on reflection, require a subtle configuration of knowledge, skills and attitudes, suggesting that the models we use to generate quantitative forecasts must look beyond simple occupational categories.

we’ve gone beyond traditional sources of labour market intelligence

In recognition of these insights, we combine a detailed look at trends and structural change, the judgement of different experts from across the country — such as economists, technologists, industry experts, academics, and entrepreneurs — and the use of machine learning techniques to generate our forecasts. 

Specifically, we first scan the horizon for past and emerging evidence on the drivers of change in the labour market; we then invite experts to consider the implications of these changes for a set of occupations, and lastly, we use the experts’ judgments on the future prospects for these occupations to train a machine learning classifier to generate probabilistic predictions for all occupations.

Sign of the times

By doing this we can produce forecasts that are grounded in rich, history-informed judgments from experts, and in predictive models that recognise the complex relationship between a job’s features and its future prospects. This makes them compelling, in our view, and gives them an advantage relative to existing approaches. 

Approaches, for example, that rely mostly on extrapolating from what has happened in the past, are well known to under-predict the growth of fast-growing occupations and the decline of fast-declining ones. And newer approaches that use machine learning methods alone to make predictions can suffer from having a “black box” quality (in other words, while the inputs and outputs are known, the method remains mysterious). This can limit their value to policymakers, who must base policies on assessments not just about what might happen, but why.

An additional advantage of our mixed method approach is that it allows us to stretch our understanding of what the future might hold. To do this, we’ve gone beyond traditional sources of labour market intelligence.  

Interestingly, the trends that experts highlighted varied —  to some extent — across the country

We have scanned for signals of change, identifying a wide range of trends in our report Turn and Face the Strange —  some mature, and others just emerging or speculative — that might impact future skill demand. 

We’ve also partnered with organisations across Canada when convening our experts at a series of six workshops. At each one of these workshops, we have asked participants to reflect on how this wide range of trends might interact to impact job and skill demand over the next decade.

As well as collecting the experts’ individual judgments on the likelihood that specific occupations will experience growth or shrinkage in the future, we record their qualitative insights on which trends are most significant and the potential interactions between them. We also record the experts’ thoughts on prospective new jobs in our report, Signs of the Times

Shaping the future

Interestingly, the trends that experts highlighted varied —  to some extent — across the country. 

Technological advancements, particularly related to AI, and the potential for growing resource scarcity, were seen across the country as forces likely to shape the future of work; however, the experts we invited to participate stressed that the speed of tech adoption was likely to be much slower than some estimates have suggested, and uneven across the country and across sectors. 

At our Calgary and Whitehorse workshops, the emergence of alternative energy sources was seen as an important driver of change. The potential for blockchain technology to impact employment emerged as a key driver uniquely in Toronto, while the potential for a shrinking middle class to affect employment in some industries, for example through changes in consumer demand, was highlighted in Toronto and Whitehorse. In St. John’s, Newfoundland, demographic trends seemed to carry more weight. 

A number of participants also stressed that future employment for many occupations would be at least partially dependent on policy decisions made by governments, for example related to investments in the oil and gas or health sectors, or to regulations that could inhibit or enable tech adoption —  a helpful reminder that governments aren’t exogenous to the labour market. They (and through them, we) have the power to shape the future, not just respond to it.  

We’re excited to find out what changes in skill demand our model will predict, and how these will differ from or mirror those predicted using existing approaches, such as the Canadian Occupational Projection System (COPS). Based on early insights from this project, we expect the results to reflect the potential for a number of different trends to interact, magnifying the uncertainties and shaping the future of work in ways that extend well beyond the (likely over-anticipated) forces of automation. 

No forecast is certain, but we hope this one will be useful. Watch this space. — Sarah Doyle and Hasan Bakhshi

(Picture credit: Unsplash)


Leave a Reply

to leave a comment.
Master the skills you need for the public service.

Discover inspiring resources, tools and policies.