Algorithms, predictive recruitment…. And the human in all that?

Algorithms, predictive recruitment…. And the human in all that?

FacebooktwitterlinkedinmailFacebooktwitterlinkedinmail

I’m always really excited to see the public attending conferences on innovation, new technologies and how they’ll transform our way of working. The latest presentations I’ve led on new methods of assessment (predictive recruitment) have prompted the same reactions. I can put them into 3 phrases: “The whole world is so attached to their decision-making power that they forget why we recruit…”

 

When we present predictive recruitment

#1 “I already use the test for my short list and it’s enough for me”, “Yes… in the end, that’s only recruitment tests…”

#2 “Wow… we can really anticipate everything???”, “It’s incredible to be so accurate”, “This is exactly what we lack today”, “This is very powerful”, “I didn’t think that was possible”, “It largely exceeds the tests I’ve seen thus far…”

#3 “It’s a very powerful tool… but a little scary aswell”, “Finally, it shouldn’t replace us either”, “I’m afraid to rely on this tool to make decisions”, “I might be influenced in my interpretation”, “AND THE HUMAN IN ALL THAT???”

The whole world is so attached to their decision-making power that they forget why they recruit: to make sure that everyone succeeds and embodies the job that recruiters entrusted to them. And if an algorithm – which takes into account hundreds of data points on a person and analyzes thousands of positions in a few tenths of a second – makes the best decision before a recruiter even sees the resume and motivation letter, what’s the problem?

Most people are astonished by what the algorithm can produce… until they project what it could mean for their personal situation. Either,  they are brilliant people who can capture their potential to grow and take advantage of that to better their position (example: Fathallah Charef, HRD of BHV Marais). Or they fall back to the dogma –  everything that touches humans can only be dealt with by humans.

Algorithms are better than the intuition of experts in 46% of situations.

The science vs the intuition of experts

Let’s get away from beliefs and look at scientific articles that address this topic. The meta-analysis of Grove and Al. (2000) reported 136 studies that compare the quality of “clinical” (intuitive) decisions to  “mechanical” (following a model). Here are the conclusions:

  • Algorithms are better than the intuition of experts in 46% of situations.
  • Algorithms are equivalent to the intuition of experts in 48% of situations.
  • The intuition of experts is better than algorithms’ in 6% of situations.

Here is a specific example of one of the studies conducted in the meta-analysis, which is focused on the prediction of managerial success: the correlation between the predictions by experts to algorithms amounts to .19 compared to .46.

A few remarks from the authors:

  • The level of practice and experience of the experts doesn’t influence the difference in accuracy between the two methods.
  • The fact that the experts have more information than the algorithms doesn’t influence the difference in accuracy between the two methods.

The fact that experts have data from an interview even lowers the quality of their predictions, which reinforces the superiority of the algorithms! (Reread this sentence 3 times before posting a comment regarding the importance of meeting people before making a decision about them…)

In another meta-analysis conducted by Kuncel (2013) published in the Journal of Applied Psychology, also conducted in the Harvard Business Review in May (2013), the quality of prediction conducted by statistical methods versus experts concluded the same results:

  • Prediction correlation by algorithms on (for 1392 people): .46
  • Prediction correlation by experts on (for 1156 people): .27

 

Comments by the authors:

The algorithms improve the prediction by more than 50%.

An important quantity and quality of information is lost when the data is reinterpreted by experts, even when they know the position and company in question.

 

Science vs popular beliefs

And yet, recently featured in Forbes, is the opinion of experts who comment on predictive models: “I believe that prediction is a decoy. We can uncover potential, but we can’t predict the future.”

Taking a recruitment decision by prediction anticipates the ability of a person to succeed in the future in a given position. What our current level of knowledge shows is that algorithms are two times better than the experts on the subject. Playing on fears is dishonest and will not last long.

Another classic error found in the same article is: “We have to see algorithms as an observation, but there is still a need for an observer to make a decision and exercise their intuition.” It’s a popular belief that everyone wants to hear and that scientific studies converge on the same conclusion. However, just a reminder: an important quantity and quality information is lost when data is re-interpreted by experts, even when they know the position and company in question. (Kuncel, 2013)

In France, we’re still too proud to accept these conclusions. But we are pleased to say our intuition always prevails in decisions.

 

The new role of experts

Experts will retain an important role in the future, but not the one we think. They’re not the best to analyze the algorithms’ results at the end of the process. Instead, they have an essential role in the creation of algorithms and their ideal expectations. The statistical models only have the value and credibility they are given at the get go. The role of experts is to supply quality data to the learning machine to ensure predictions coincide with expectations. The Algorithms don’t automatically know what to do with all of the data they receive. It’s we, the experts, that have the information necessary to configure the algorithms and that’s where we must position.

Today, we tolerate 2 times more the errors when humans make decisions than algorithms

With such conclusions, why not use the recommendations of algorithms to help make our decisions? I understand that it’s difficult for people to accept the idea that an algorithm could be wrong. “It’s a machine and it has all the information – but if it’s wrong, it’s not reliable and we can’t trust it.” While at the same time, when a person makes a mistake, we tolerate it and allow for 2 times more the errors than algorithms!

This problem is not specific to the HR sector, there’s the same debate going on in the automobile industry with automatic driving cars. Are we willing to accept auto-pilot driving cars that would reduce fatal accidents by half on the roads?

The future will be algorithms. You must decide whether you are in or not.  

 

In conclusion

Access to this intelligence will soon be given to the whole world. It’s found that tools we work with today (shared calendars, email, instant messaging, transport applications…) are equivalent to a personal assistant in the 1970s. We have a way to personally organize our affairs today with the ease only leaders had access to 40 years ago. Soon, we will be able to make better decisions than any expert in their field. It’s the most beautiful contrast of inequalities, knowledge, and the IQ we’ve been given.

What’s worrisome is that people are too distracted by the slight imperfections of this tool, instead of participating in its’ development.  It’s now that we need to take a step forward and use these tools at our disposal to build the future. You must decide whether you will be part of it.

 

 

Simon BARON

Chief Science & Innovation 🚀🚀

FacebooktwitterlinkedinmailFacebooktwitterlinkedinmail
Matthieu Gaudichau

Leave a Reply

Your email address will not be published. Required fields are marked *