Artificial intelligence is booming, also when selecting new employees. More than a quarter of the larger companies use smart algorithms to assess applications. But does that lead to a fairer selection? ‘The system came up with a uniform sausage for applicants.’
Innovation scientist Elmira van den Broek worked for two years in the HR department of a large multinational, MultiCo, for a study conducted by the Free University of Amsterdam. There she investigated how a selection tool for commercial employees that runs on artificial intelligence (AI) worked in practice. Her verdict: ‘AI cannot do it alone. But it is a good mirror for the manager.’
What are the promises of an AI algorithm over human selection?
‘The big promise is that algorithms are more objective and efficient than humans. An algorithm does not get tired and does not suffer from a bad mood. Research has shown that judges who had to decide on parole of prisoners were more likely to reject a request before lunch than after, when their stomachs were full.
READ ALSO
How to lose weight by managing your expectations about food
If you want to lose weight, it seems logical to choose foods that are labeled healthy and low in calories, carbohydrates and fat b…
But algorithms are not neutral either. They are fed with data produced by humans. That’s how Amazon’s algorithm continuously rejected women. After all, it was trained on decisions made by HR employees who had been biased against women for years.’
How did the computer work at MultiCo?
‘The algorithm is, of course, a human product. The computer was provided with endless amounts of data from employees who used it concern were qualified as successful. The criteria were, for example: high turnover, multiple promotions and long-term employment. Based on that, an ideal profile emerged. The algorithm then calculated to what extent the measured characteristics of the applicant matched the profile of the successful employee.’
What exactly were the candidates supposed to do?
‘The trainees had to play a number of online games. Like blowing up a balloon as far as it can go. Underlying idea: measuring risk behaviour. A seller must take initiatives, but not be reckless. The balloon should not burst. Multitasking was tested by having candidates drive a car virtually and collect falling blocks at the same time. They were also shown a video clip where they had to read the emotions of faces.
In addition, the candidates had to give a video presentation. The system then pays attention to the expression in the face. If you laugh a lot, you radiate positivity, if you look straight into the camera, you come across as confident. Women scored higher than men here due to their way of talking and facial expressions. Because of this gender bias the company decided to stop using the video algorithm.”
What is the disadvantage of AI-driven application procedures?
‘The big drawback is that the algorithm looks very strictly at personal performance – how smart or risk-seeking someone is – but does not look at how someone behaves in a group or towards the customer. The system also came up with a uniform sausage of applicants: all good at sales, while in a larger organization there is also a need for people who are more analytical. Bokitos alone are no use to you.’
Could an acommercial candidate slip through the algorithm’s selection?
‘According to the managers of MultiCo that was possible. They generally found the AI-selected candidates to be mediocre. They were less streetwise, hungry and ready to run. A manager complained that the algorithm rejected an intern with whom he had been working satisfactorily for a year. The developers of the algorithm actually found that managers pay attention to characteristics that are not important for the company’s performance.’
What is your opinion on the use of AI in recruitment and selection?
‘The algorithm can never do it alone. Artificial intelligence is a great mirror for the organization. It makes visible patterns and preferences that people have shown in the past. The algorithm at MultiCo has shown that a high IQ or a degree from a top university has no predictive value whether someone will become a good salesperson, while managers assumed it did.
The algorithm has also started the discussion about the ideal employee. The algorithm learned from the entered data that hard work was a success factor. People who also find leisure time important, who would rather be on the tennis court at five o’clock than get another order, were rejected. But in the current juncture with its many burnouts young people in particular attach less importance to earning a lot of money and more to leisure time. MultiCo therefore decided to adjust the algorithm.’
What is the ideal mix?
‘As far as I am concerned, we are moving towards parallel decision-making in application procedures. The manager assesses the social aspects, the machine the individual skills. The manager decides, the machine advises.’
MULTICO
The name of the multinational has been changed for privacy reasons. It is one of the world’s largest consumer goods companies, taking on 100 commercial trainees annually from a pool of 10,000 applicants.