AI Bias: The Unequal Treatment of Gender and Age in Employment Resumes
A recent study reveals troubling biases embedded in AI systems regarding age and gender, particularly when assessing resumes. This research, published in the journal Nature, indicates that AI-generated resumes for female candidates are, on average, 1.6 years younger than those for male candidates. Furthermore, the AI assigns lower qualifications to female applicants compared to their male counterparts, highlighting an alarming trend of age and gender discrimination in recruitment.
Impact of AI on Workforce Perception
Contrary to real-world demographics, where male and female employees in the United States are approximately the same age according to U.S. Census data, AI models tend to favor younger women and older men. This bias persisted even in sectors typically dominated by older women, such as sales and service industries.
Research Methodology
The study utilized an array of analytical methods, encompassing the evaluation of around 1.4 million online images and videos, as well as text analysis and controlled experiments. Researchers sought to establish how these skewed data inputs distorted AI outputs, specifically favoring particular demographic groups.
Douglas Guilbeault, a computational social scientist and co-author of the study, emphasized that many organizations aiming for gender diversity often still favor younger female hires, ultimately affecting their promotion opportunities. This further demonstrates the complexity of the “glass ceiling” phenomenon.
Visual Assessment of Age in Media
To illustrate the age perceptions influenced by media portrayals, over 6,000 individuals assessed images from various professions found on platforms like Google and Wikipedia. The findings showed a consistent trend: women were perceived as younger than their male counterparts, particularly in high-status roles like doctors and CEOs. This indicates a societal bias where older women struggle to be seen as authoritative compared to older men.
Textual Analysis and Its Findings
Alongside visual assessments, researchers analyzed job-related text across nine language models to rule out biases stemming from visual factors such as image enhancement tools or cosmetics. Their analysis found a link between job prestige and perceptions of age, with less prestigious roles often associated with younger women and more prestigious roles linked with older men.
Influence of Image Uploads on Age Perception
A subsequent experiment involving over 450 participants tested if uploading images impacted beliefs about age in careers. Participants who uploaded images of women estimated the average age of others in the occupation as younger compared to those who uploaded images of men, who were rated as older on average. This highlights a significant influence of AI on public perception.
Consequences of AI Biases
The study showcased how biases built into AI systems affected evaluations of resumes submitted by candidates across 54 occupations, yielding nearly 17,300 resumes per gender category. The AI consistently generated younger and less experienced resumes for women and assigned them lower scores compared to the resumes of their male counterparts, reflecting a systemic bias.
These biases have broader implications, not just for women but across the board; even young men received lower scores in certain scenarios when compared to young women. This suggests a pervasive issue within AI systems that could further entrench societal biases.
The Path Forward
As the usage of AI continues to grow, there’s an urgent need for a more nuanced approach to address overlapping biases of age, gender, and race in digital environments. Guilbeault calls for a comprehensive strategy that recognizes the intersectionality of different inequalities, underscoring that real discrimination arises from these interconnected biases.
This research underlines the importance of addressing AI biases proactively to create a fairer workplace and challenges developers to refine their models to reflect diverse and accurate perspectives on age and gender in professional settings.
