AmCham Slovakia

AI bias in the hiring process

More than half a century ago, conductor Erich Leinsdorf, who directed the Boston Symphony Orchestra, was seeking fresh performers to join his orchestra. At the time, discrimination was prevalent in the music industry, with a pronounced bias towards men over women. These barriers were frequently caused by societal expectations, or the notion that specific instruments or roles were inherently masculine and hence not acceptable for women. 

To prevent this bias, Leinsdorf devised a method for conducting the audition where the candidates were positioned behind a sheet, preventing the judges from discerning their identity, age, or gender. Consequently, all candidates were evaluated exclusively based on their musical performance. The unique approach, known as “blind auditions,” led to a significant increase in the selection of women based simply on the criterion of “musical talent” compared to the standard selection processes.

This straightforward approach, which is now widely used, not only revealed significant discrimination, but was also offered a method to address it.

What is the parallel with AI?

Fast forward to today -- AI algorithms have become prevalent across all aspects of business operations, including the hiring processes.

AI algorithms used in the hiring process of employees might be influenced by a multitude of biases and discriminatory factors. This occurrence is referred to as AI bias and carries significant legal and ethical consequences that society cannot accept.

connection2024_02.png The primary issue is that, since AI is trained on historical data, AI algorithms have a tendency to mirror and reinforce preexisting social imbalances, prejudices and biases. For instance, if the AI is trained on data that exhibits discriminatory biases towards specific demographic groups, these algorithms may inadvertently perpetuate such disparities in employment judgments.

It is widely known that Amazon abandoned an AI-based recruitment tool that was employed to assess job application resumes due to concerns over bias. These algorithms tended to favor male candidates over female candidates, as a result of the historical imbalance in the number of male applicants for positions. Biased AI systems have the potential to result in discriminatory practices depending on gender, age, nationality, race, or any other characteristic captured in the data.

Another example of targeted discrimination involves a corporation that employed artificial intelligence to assess video interviews with job applicants. In this instance, the AI exhibited a tendency to display bias towards applicants whose accents were not included in the AI’s training data.

Similar to the blind auditions, it is evident that there is a pressing need for a “sheet” to mitigate the biases that AI has acquired during the learning process. The issue at hand, as well as similar variations of it, can be resolved through the implementation of legal regulations governing artificial intelligence.

As an illustration - Article 5 of the EU Artificial Intelligence Act sets forth artificial intelligence practices which are prohibited. These include for example:

  • The use of biometric categorization systems that classify natural persons based on their biometric data to deduce or infer their race, political opinions, trade union membership, religious or philosophical beliefs, sexual preferences, or orientation
  • The use of AI systems for evaluation or classification of natural persons or groups thereof over a certain period of time based on their social behavior or known, inferred or predicted personal or personality characteristics, with the social score leading to detrimental or unfavorable treatment of certain natural persons or whole groups thereof
  1. in social contexts that are unrelated to the contexts in which the data was originally generated or collected, or
  2. that is unjustified or disproportionate to their social behavior or its gravity.

While the adoption of new EU legislation is a huge step forward, will it be sufficient? It will take a considerable amount of time before this regulation can be lawfully enforced. As always, there is uncertainty over the degree to which this regulation will be efficiently implemented, and its infringement penalized.

What can businesses do to avoid AI bias? 

First, it is imperative to implement proactive measures that will consistently monitor and assess the use of AI in hiring procedures to detect and subsequently rectify occurrences of AI bias. It is also imperative to regularly audit algorithms, data, and system implementation in order to identify and remove biased trends. Additionally, it is crucial to prioritize diversity and inclusion in the development and implementation of AI technologies for hiring processes.

It is advisable to bring together professionals from various disciplines to oversee your company’s AI governance and participate in the creation of these procedures, with the aim of ensuring that the algorithms employed are considerate of diversity, inclusion, and other ethical dimensions. Furthermore, such standards and governance might potentially be integrated into ESG reporting.

Katarína Pecnová, Managing Counsel, Dentons