AI driven job matching has a bias problem, new ITU research tackles the issue

Algorithms are far from objective. In a new three-year industry partnership with Jobindex, ITU Associate Professor Toine Bogers and Postdoc Mesut Kaya will work to develop job and candidate recommendation algorithms that take multi-sided fairness into account.

Toine BogersCollaborationsComputer Science DepartmentResearchalgorithmsartificial intelligencediversityethics

Written 7 May, 2024 08:27 by Theis Duelund Jensen

With the recent passing of the EU’s AI Act – the world's first comprehensive horizontal legal framework for AI – among other things, the problem of bias in job recruitment has become a major topic. With the advent of sophisticated AI solutions that save job recruiters time by screening applicants on various baseline parameters, securing fairness in job recruiting has become a technological issue. Unfortunately, biases tend to manifest in AI technology, because the algorithms are trained on data produced by humans. So how is fair and equal representation possible in job recruitment?

That is the question, Jobindex – the largest online portal for job listings in Denmark – has put to ITU Associate Professor Toine Bogers and Postdoc Mesut Kaya. The two are in the starting phase of a three-year collaborative research project that seeks to make AI recruitment solutions more fair and less biased by conducting qualitative surveys among job seekers and recruiters and examining the metrics the current recruitment process is based on.

"Bias in recruiting is a huge problem,” says Kaare Danielsen from Jobindex. “But with AI technology and the research cooperation with ITU we hope to be able to detect and reduce biases from both sides of the recruitment process.”

The project objectives are to develop a so-called fairness dashboard for tracking fairness metrics relevant to Jobindex, to develop job and candidate recommendation algorithms that take multi-sided fairness into account, and to integrate these solutions into the Jobindex platform, ready to roll out at the end of the project.

“We want to advance state-of-the-art research in ‘fair AI’ in human resource management by being the first to balance fairness concerns across multiple stakeholders in the development of job and candidate recommendation algorithms,” says Mesut Kaya whose doctoral work was in the field of recommender systems.

The ultimate goals for the researchers include developing new methods and tools for detecting and mitigating biases and increasing the quality and fairness of AI-driven job matching. But there are many concerns to address before a commercially viable tool can be put into action.

“We want to further the research in fair AI, but we have to ensure that we do not compromise on accuracy when we focus on fairness as a parameter in the technology. We have to strike a good balance in order for Jobindex to use the technology. Hopefully we will be able to increase fairness without harming accuracy,” says Toine Bogers.

With the EU’s passing of the AI Act, job recruitment where AI is already used extensively is categorised as a high-risk area, and businesses and organisations will be responsible for ensuring that their platforms and tools meet the adopted standards.

“There is a huge potential for Jobindex’ business,” says Mesut Kaya who will be dividing his time between the company and ITU. “But we also hope to publish our findings and push development in fair AI as a research field forward."

Further information