🔮 Role vs Skills, AI Recommendations and The Power of Weak Links

Hello! 👋

For this week’s edition, I had the pleasure of talking with Marian-Andrei Rizoiu, a Senior Lecturer leading the Behavioral Data Science lab at the University of Technology Sydney.

He’s done very interesting research on: online popularity prediction, real-time tracking and countering disinformation campaigns, and understanding shortages and mismatches in labour markets.

🤹Transitioning from Role based to Skill based jobs

This is the most important topic we covered. In a nutshell: the job market is shifting from a role-based to a skill-based system, and automation and digitization are causing some skills to become obsolete while creating a shortage of new skills. This is having an effect on the economy, with some roles staying stable but a need for new skills arising.

He bases this on two main factors that drive labor ‘market shocks’:

• Crises (i.e. Covid-19)

• Structural shifts in demand (i.e. technological advances)

Transitioning workers at scale requires maximizing the similarity between current skills and target jobs. Also, detecting AI disruption that could accelerate job transitions is another interesting task. Joining the two together, you can understand which roles at risk of automation can transition to others that are at lower risk (an example in the study is of a ‘Sheetmetal Worker’ that is at high risk, but shares most skills of an ‘Industrial Designer’ which is at very low risk - 4%).

You can find an example on Linkedin’s page ‘Future of Skills’ (though it doesn’t include the automation/AI displacement part):

The system also needs to consider that job transitions take into account educational requirements.

At this level, Universities are also moving to micro-credentialing, and intensive short term courses that deliver a well defined set of skills:

Bridging the gap between Education and Work - We see more and more how GenZ are starting to work earlier on in their lives (60% claim to have a more entrepreneurial spirit and look to have side-hustles vs Millennials) but missing key skills when switching from the academic world to their first jobs. Micro-credentialing can help by being surgical and personalized.

Up-skilling/Re-skilling - More and more jobs change inside of the same companies at a faster pace. Micro-credentialing can help the use case of: ‘I want to apply to that more senior role, what do I need to focus on learning next?’ and actually having a way to certify this while applying it on the job.

🤖 AI-Job Recommendation

Together with other researchers, Marian has built an AI-job recommender, which aims to solve these problems.

We know that a lot of AI is being used by HRs/Companies (i.e. ATS, Applicant Tracking Systems) and both third party SaaS and proprietary models that try to do candidate scoring automatically. Some companies, like Amazon, are saying it’s almost as accurate as human evaluation, and have even started reducing their talent acquisition teams.

But currently, as also Marian-Andrei argues, there are two untapped opportunities:

💼 Company-level - All the tools we’ve seen out there mainly focus on evaluating single profiles. What about the collective population? Currently, there are some ways to do that through manual, individual testing, but even if you manage to get a snapshot of the collective skill mapping, you are then left with the task to optimize for a) what you need b) what you currently have.

However, current solutions give out something like this - which is not very useful:

There’s also a catch: As an individual, it is best to focus on getting the best skills and courses to reach your goals. However, when it comes to a company, it is not always best to focus on the best thing for each individual. It is important to look at the bigger picture and make sure that the collective best does not lead to a global failure. For example, if everyone in a job is a financial analyst, it may not be the best use of resources. It may be more beneficial to hire data scientists. The challenge is to figure out how to get to the desired outcome in the most efficient way with the resources available and their individual needs. Plus, you need to figure out how to match those skills when you’re forming out a team.

👨‍💼 Candidate-level - Individual candidates need to do skill-matching and career planning. So recommenders like the one we discussed above, could be very helpful for people to figure out where they are in their career, spot trends and plan for specific training.

Linkedin has launched a specific section, called ‘Career Explorer’, which aims to do just that:

Lightcast, a labor analytics company, has also launched ‘Career Pathways’ that does a similar job:

I think an interesting question is: how do you apply these findings and skill-driven transitions to real-life?

OK - I’ve followed the suggestions, taken the trainings. But are recruiters, or even worse, AI-software, going to be able to evaluate whether that is enough? What if experience is weighted more in this scoring, and what other biases may there be?

💾 Data Ownership and Validation

Who owns the data? Who stores it? Who validates it?

These three questions are on top of my mind when I think about the way forward.

Think about it: should the company you’re working for own data on your skills, or should you? What happens when you move to another company (and that happens a lot faster than ever, with tenure shortening down to almost two years on average)?

When touching upon credentialing, Marian-Andrei talks about the European Skills Passport as being an example of how central institutions are moving to this model too. More recently, the Council of the European Union has redacted a document on this (Recommendation on a European approach to micro-credentials for lifelong learning and employability) outlining how it can help establish common ground, transparency and facilitate this transition.

His argument is that the data, and the validation, will need to be centralized by institutions, who will also help direct people towards the right education. There’s two takes on this, from my point of view:

Privacy-first - Private storage, can solve privacy related problems like GDPR where tight control and regulation can instruct where data is stored and how it’s moved from one party to the other.

Community-first - De-centralized data storing won’t cut it for institutions, at least not yet. On validation: should peers and experts in your field evaluate whether you have one particular skillset? Or colleagues at your current job? This approach has a big problem on accountability and trust, but there could be useful elements to use in a broader model.

🍸 Online Popularity and Human Behaviour

Marian-Andrei works on mathematical models to understand how information flows within digital social systems - The models are used to understand the occurrence of events, such as when people share and re-share content, which are known as atomic level actions.

These are then taken to collective level, where clicks, views, shares, likes, upvotes, and downloads are often used to measure the success of a product or idea outside of the platform it was created on. He works on how social and psychological factors are accounted for in a model and how virality and popularity occur, as well as such as the half life of human memory and the fact that popular people attract more attention, no matter what they say.

I find it really interesting to see how there is a strong connection between research on human behavior online, and everything we’ve discussed on the job market.

If you think about it, the key question that this research is trying to solve, ‘Why did X become popular, but not Y?’ can definitely be applied to understand why there are particular trends in certain technologies, tools, skills and so on. 

One particular research he reported on, was around testing the power of weak links in the job search, proving causality for the first time of Granovetter’s (1973) concept:

• ‘People you May Know’ Recommender engine significantly shapes link formation - Users formed more weak/strong links as recommended.

• Evidence shows moderately weak ties (defined as on average 10 mutual friends, who rarely interact) are more than twice as effective than strong ties for job-seeking.

• Strength of weak ties varied by industry - weak ties increased job mobility in digital, strong ties in less digital.

So the bottom line, for me, is that understanding and modeling human behavior is at the heart of everything, especially online where you have so much available information that you can connect together.

And yes, it’s true that we are all unique and unpredictable (or ‘stochastic’ as Marian-Andrei technically puts it) - but when I asked him what surprised him the most in his years of research, he was sure:

Thank you for reading Work3 - The Future of Work. This post is public so feel free to share it.