Jobs of the Future: Clear Pathways

At KIMO, we strongly believe the 2020–2030 period will be a critical phase for reskilling towards the Fourth Industrial Revolution. We are not alone in this, warnings from experts in strategy boutiques and governments alike have repeatedly indicated that in many organizations people are not re-skilling fast enough to keep up with the acceleration of technological change we see today (to dive deeper: BCG, our blog. WEF, Oliver Wyman). Not keeping up with the digital world can have dire consequences, as (partial) automation predictions for jobs range anywhere from 9% to 45% of jobs (more below). In this article, we will present our best thinking on the way forward. In particular, we will focus on how to design the right learning paths at scale for individuals in societies, based on how similar two jobs are. We will do this based on an internal project where we clustered all 2,942 European jobs based on their underlying 13,485 skills. If you prefer to jump right to the demo, you can do so here.

Two jobs (ICT Consultant and ICT Analyst) that are similar (0,88 proximity).

What is the status of reskilling efforts today? The reality on the ground is that most people are not learning fast enough. Lifelong learning (LLL) is a popular term in boardrooms and HR teams, but the term is ~50 years old by now (UNESCO/OECD, the 1970s) and exists more in theory than in practice. In addition, catching up with a digital reality might become harder over time. An example might clarify this: say that you decide to become an Artificial Intelligence Specialist (a job in high demand today, up 74% since last year, average salary 140k USD!). This sounds like a good choice, but you may find that the train has already left the station when you arrive. Why? Because the skillset you’ve selected builds on numerous underlying skills and mental models, e.g. you will probably need to code in Python, R, C/C++, understand matrix factorization, derivatives and probability distributions in the world of mathematics, learn some Cloud Deployment in GCP/AWS/Azure, update your knowledge of databases, understand privacy concerns/laws (GCPR) that apply, and learn how to use GPUs/TPUs to train models. The insight here is that knowledge builds like compound interest. New ‘information’ needs to fit into an existing framework to make sense, to become ‘knowledge’.

Two jobs (ICT Consultant and Zookeeper) that are far apart (proximity 0,17).

Which jobs will be automated? This is an area of heavy debate. The key insights here in our view are that there is a huge technological potential to automate jobs on a large scale, although other factors (labor laws, regulations etc.) will play a role as well. Recently, McKinsey mentioned that 50% of jobs are technically automatable, PwC calculated that 38% of jobs could be automated, Bain believes 20–25% fewer workers will be needed.

What learning paths are worth pursuing? This is the key question. We believe, that once the taxonomy for jobs and skills is more standardized (with ESCO in the EU and O*net in the US as good efforts) and more data online, learning paths don’t have to be guesswork. Instead, it will be possible to make decisions in a more fact-based way. The distance from job A to B can theoretically be calculated, and thus the (economic) feasibility of a journey. E.g. a learning journey from ICT Consultant to ICT Manager should be a more feasible journey than from ICT Consultant to Zookeeper.

To aid in this solution, we have designed a demo. This demo shows the logic in action. Here’s what we have done to create the demo:

  1. We mapped the 13,485 skills to the 2,942 jobs (ESCO), as this was not the case in the original datasets. We have also mapped the automation probability from the Oxford Martin School. This mapping is done based on word/text similarity (similar to work done by MIT, see here). The demo allows you to find any job as known in the EU database with its underlying skills (as described in the EU database), as well
  2. We clustered jobs based on similarities. The 3D cloud shows clusters of jobs that are more similar in the model. Note that these clusters are based on the job title, or job descriptions provided (which often include the skills). Where these descriptions are poor, the model may lack accuracy.
  3. Proximity calculation from A to B. The demo further allows you to click on multiple jobs, and see what the ‘proximity’ is (0–1). 1 means it’s the same job, 0 means it the opposite job in terms of underlying skills (Note: this is just a first attempt to model such distances.


‹  back to blog
  • Udemy
  • Coursera
  • Youtube
  • Udacity
  • Medium
  • Datacamp
  • edX

Kimo: Chat with me!