FairMatch project for fair recruiting
Many companies are starting to use AI to match jobs and companies. But this can lead to unfair biases in recruiting, if the algorithms are biased or there is bias in the training set. And with the new EU AI act, companies have to take precautions before they use new AI algorithms for selecting people.
This is actually a two-sided problem, because candidates can also be biased in the jobsearch process by e.g. prefering to look for jobs in companies that they already know. This can lead to both some candidates having trouble finding jobs and some companies having trouble finding candidates.
The FairMatch project is a 3-year research project between that aim to develop fair AI matching algorithms, thereby helping both candidates that have troble finding jobs and companies that have troble finding candidates. The project is supported by Innovation Fund Denmark with 1 million Dkr.
Jobindex and ITU will develop fair matching
Bias in recruiting is a problem for both candidates and companies, and the problem can become even bigger if companies and candidates use AI algorithms that contain unfair bias. Researchers at Jobindex and IT-University of Copenhagen will take on the challenge. The project is a continuation of the JobMatch project between University of Copenhagen and Aalborg University and Jobindex.
The key to the system is data from the approx. 100,000 jobs manually matched by Jobindex’ recruitment consultants during 10 years. The researchers will use this data to provide the system with knowledge of the personal characteristics and professional skills that fit various jobs.
The first task in the project will be to determine what fair matching actually means. This is not as trivial as it sounds, because there are many different definitions for this, and researchers do not agree on what is fair. Are quotas in recruiting for instance fair or unfair?
But no matter what definitions of fair matching the project ends up using, we hope that the project will lead to better matching and help candidates find jobs and help companies find candidates, and in this way also a benefit to society.
The system will be built with the same kind of algorithms used by Tinder to recommend matches between parties. If Tinder used the same algorithms as Netflix and Amazon and Jobindex, everybody would be matched with models and nobody would get a date. The key is to avoid unfair biases and find matches that both the candidates and the workplaces find interesting, and thereby increase the number of people who finds a job using Jobindex.
Denmark’s largest recruitment portal
Jobindex is involved in about 15 % of the recruitments in Denmark. Almost 1 million people visit Jobindex every month, and according to a survey by Megafon, approx. 500,000 Danes have found their current job via Jobindex.
Project participants
Carina Emma Sørensen, Jobindex
Articles
Press
