Mitigating Age Biases in Resume Screening AI Models

Abstract

As populations age, an increasing number of workers beyond the traditional retirement age are opting to continue working. Nevertheless, discrimination against older job seekers seeking new employment opportunities remains widespread. To address this issue, we enlisted a pool of crowdworkers to assess the resumes of IT job candidates and guess each candidate's age, race, and gender. Using this crowdsourced data, we trained an AI model and applied bias correction techniques from IBM's AI 360 and Microsoft's Fairlearn toolkits to correct for biases based on race, gender, and age. We analyzed the effectiveness of these tools in mitigating different types of bias in job hiring algorithms, explored why age may be more challenging to eliminate than other forms of bias, and discussed additional approaches to enhance fairness. Our results indicate that implicit age bias, or ageism, is prevalent in hiring decisions and more pervasive than other well-documented forms of bias, such as race and gender biases.