Melding machine learning and participatory foresight
Last year SAMI colleagues assisted in beta-testing a workshop process design which was used in a project looking at the skills that may be required in the future and is the subject of the following article.
Occupations, Tasks, Skills: what might we need most in the future?
What do you want to be when you grow up? Are you thinking about choosing a new career? Worried about the jobs available for your children, and what skills they might need to succeed in the future? Pearson, the world’s largest education company, has sponsored Nesta UK and the University of Oxford in forecasting future jobs availability and skills requirements.
The resulting project, “Employment in 2030: Skills, Competencies, and the Implications for Learning,” explored which occupations may disappear in the future, and which may be in greater demand. Robots and smart things everywhere are supplanting human labour —how many occupations are actually in jeopardy?
Official government occupational lists from the UK and the USA provided the starting point for the project. Each occupation’s description includes a list of typical tasks and the skills they require. Consequently, forecasting the potential for growth or decline of a specific occupation also indicated the potential for growth or decline in demand for specific skills: will we still need chefs in the future?
Or will nimble robots assemble raw ingredients into gourmet meals in gleaming restaurant kitchens? Does that only mean that restaurants no longer require line cooks, and that the chef’s key tasks—of designing the dishes and curating the meals, calculating the ingredients required and their most efficient use—and related skills will still be in high demand?
Understanding what skills future occupations will require helps educators understand what the critical educational needs of the future might be. What do people really need to know to thrive in the transformed economy of the future?
How can experts work together to help train an algorithm about occupational futures?
Pearson, Nesta and the University of Oxford set out to explore these questions with a unique futures approach: a dialogue between human experts and an active learning algorithm. The array of methods available to foresight researchers and forecasters splits across Snow’s two cultures: the highly data-driven and quantitative vs the data-informed, intuitive and qualitative. This project sought consciously to bridge the two.
It’s a tricky bridge to build. Nesta invited world class thinkers to discuss emerging changes and their potential effects on occupations to one workshop in Boston, and another in London. I should also declare an interest at this point: I was commissioned to design and facilitate the workshops that elicited the expert opinion that informed the algorithms.
The workshop attendees were looking forward to a stimulating conversation and discussion about transformations in jobs and labour. The active learning algorithm that we needed to feed, on the other hand, simply wanted those experts to rate possible growth or decline in 30 occupations—not a very interactive task. Our goal was to design a process that supported wide-ranging thinking about change, stimulated discussions across the expert perspectives in the room—and paused periodically to inform the algorithm, via a simple web-based input form.
How did we do it? The most critical part of the process design was explaining clearly what we were asking participants to do, and why: the introductions to the project goals (understanding what skills the future will require), the overview of machine learning (introducing them to the algorithm they’d be training), and their role in sharing data and contributing insights. Second, while many of the experts had deep understanding of one sector of change, we wanted everyone to have a shared baseline of changes across multiple sectors of changes that might affect the future of occupations. Thus a key component of the project was extensive research into critical trends, and summarising them as a trends deck for discussion.
The workshop live
What, then, did we actually do on the day? Each workshop began with a round of introductions that included—as an icebreaker and a way to frame discussions —the question, “When you were ten, what did you want to be when you grew up?” It turns out that even economists, social scientists, and technology experts wanted to be astronauts or soccer stars when they were ten!
The project team then galloped rather rapidly through the extensive map of trends and emerging issues affecting economic activity and potentially transforming occupations over the next twenty years. After listening to a data-dense presentation for thirty minutes, participants deserved an opportunity to process the information and its implications, and to discuss the trends amongst themselves. We wanted them to think about how those impacts might affect all the different activities in the economic landscape. We also wanted to switch thinking modes from listening (verbal) to mapping impacts and interconnections (visual).
To spur that conversation and encourage that thinking mode, we printed out a table-sized cartoon of a city landscape (office buildings, retail space, government buildings, arts and leisure, manufacturing facilities, transport infrastructure, agriculture, suburbs, etc.), and a deck featuring each of the key trends and emerging changes, summarized in a phrase. Participants worked in pairs to cluster the trends they thought reinforced and amplified each other, and then placed each change or change cluster on the map where they thought it would have the greatest impacts.
If they thought the change affected more than one economic activity, they could use coloured tape to connect their chosen change to other activities. We then asked them to summarise by suggesting two occupations that, as a result, would increase in future, and two that would decrease.
After stretching their mental muscles with that exercise, we moved on to labelling specific occupations with directional forecasts—“will this occupation increase or decrease in the next twenty years?”—an activity that was assisted by fact sheets that included a definition of the occupation, its specified tasks and required skills, and historical trend data summarizing its growth in the economy. We instantly displayed the forecast output to the group, so they could discuss the range of answers, explore the different assumptions each of them used to arrive at their forecasts, and potentially amend their initial judgments. Participants could refer to the trend deck for change ‘evidence’ to support their assumptions and forecasts. To provide the algorithm with its required base of thirty occupation evaluations, we repeated this process two more times.
Giving the participants discussion space during the occupation forecast labelling gave them an opportunity for unconstrained and exploratory thinking, in contrast to the very constrained mental process of assigning ‘increase or decrease’ labels to 30 different occupations. But it didn’t provide space for another obvious potential output from combining emerging changes, forecasts of impacts on current occupations, and human creativity: brainstorming entirely novel occupations, or radical transformations in current occupations. To capture those creative insights, participants ended the day by pairing up again and scrawling their wild and divergent extrapolations of the occupations of the future on a wall mural with the key change clusters already posted.
Emergent forecasting… watch this space
The research team has mapped the interesting changes, the experts have discussed the possible impacts, and the algorithm is computing. We are looking forward to the output: forecasts of potential future growth—and decline—in specific occupations in the UK and the USA and, more importantly, in the range of critical tasks and skills that may be required of the next generation of workers.
Web pundits trumpet the erosion of employment due to increasing roboticisation and the emerging ecology of ‘smart everything’ (cities, factories, cars, toasters, you name it). Nesta has just demonstrated at least one instance where human intuition and machine learning can work hand in hand. Maybe that’s the best future we could hope for.
Written by Wendy Schultz, SAMI Principal. A version of this article was previously published in the January 2017 edition of Compass, the newsletter of the Association of Professional Futurists, and is republished with their permission under a Creative Commons licence. There is more information, and project reports, at Nesta UK.
The views expressed are those of the author and not necessarily of SAMI Consulting.