An International Labour Organization (ILO) Senior Economist, Janine Berg, has raised concerns over the growing use of artificial intelligence (AI) in recruitment and workplace management, warning that many organisations are deploying automated systems without fully understanding how they function or whether they deliver reliable outcomes.
Speaking in a recent ILO working paper co-authored with Hannah Johnston, Berg said employers are increasingly turning to AI tools to manage rising volumes of job applications and workplace processes. However, she cautioned that these systems may simply amplify existing flaws in decision-making rather than improve them, particularly as the adoption of AI in human resource management continues to outpace understanding of the technology’s effectiveness.
Berg explained that hiring and workforce management have always been complex tasks, but the expansion of online applications and generative AI has significantly increased the number of candidates employers must assess. This development, she said, has prompted many organisations to rely heavily on technology to screen and evaluate applicants.
The study examined the use of AI across four major HR functions — recruitment, compensation, scheduling, and performance management. Researchers assessed the systems using three key criteria: the objective of the system, the quality of the data used, and the way the technology was programmed.
According to Berg, while defining the objective of an AI system may appear straightforward, it is often difficult in practice because human behaviour is complex and cannot always be easily measured. She contrasted this with simpler technical challenges, such as calculating the shortest route between two points, which depend on clear and measurable variables.
She cited an example of an organisation that developed a recruitment AI system to identify candidates with a “growth mindset”. The programme was designed to score applicants based on how frequently they used words such as “growth”, “development”, and “learning” during interviews. Berg argued that this reduced a complex human characteristic to basic word counts, raising doubts about whether the system was measuring anything meaningful.
The study also highlighted concerns surrounding data quality and relevance. Berg said AI systems rely heavily on the data on which they are trained, warning that poor-quality data inevitably produces poor-quality outcomes. She added that some organisations rely on external datasets that may not accurately reflect their workforce realities or organisational context.
She further noted that some companies are gathering large volumes of questionable data in an attempt to improve recruitment and performance systems. This includes the use of digital games, such as a virtual balloon task, in which applicants inflate a balloon to assess risk-taking behaviour and reaction speed.
Berg also expressed concern about employee-monitoring technologies that track online activity, including keystrokes, screenshots, and screen time. She warned that time spent online does not necessarily correspond with actual productivity or job performance.
The ILO expert stated that AI systems do not operate according to human theory, but instead identify patterns from data. As a result, she stressed the importance of establishing a clear HR framework to guide the collection and interpretation of workplace data. Without such a framework, she warned, HR professionals may struggle to interpret system outputs effectively.
She further explained that algorithms can reproduce hidden biases, even when they are not intentionally designed to discriminate. Berg referenced a study in which a gender-neutral STEM job advertisement was shown predominantly to men because an advertising algorithm determined that targeting male users was cheaper.
Berg added that many AI-powered HR tools are marketed as user-friendly systems requiring minimal technical expertise. However, she cautioned that this can be misleading, as many HR practitioners lack a full understanding of how such systems operate or how to interpret their findings responsibly.
To address these concerns, she called for stronger involvement of HR professionals and employees in the design, implementation, and oversight of AI systems. She said this was especially important in areas such as scheduling and performance management, where decisions directly affect workers’ wellbeing and productivity.
Berg pointed to examples of successful collaboration, including a multinational company that spent two years developing an AI recruitment system with HR input before adopting a hybrid model that combined human judgement with explainable AI outcomes. In another case, a telecommunications company collaborated with field technicians to develop a scheduling system that increased productivity by 10 per cent and reduced mental health-related absences by more than one-third.
She concluded that involving all stakeholders in the development process helps ensure AI systems are transparent, fair, and aligned with real workplace conditions, even if such collaboration requires more time and resources.
