In recent recruitment processes, a novel challenge has emerged: artificial intelligence (AI) systems used for candidate screening may exhibit bias towards resumes generated or enhanced by AI tools. This phenomenon adds a complex layer to an already competitive hiring landscape where candidates increasingly rely on AI to streamline and optimize their applications.
Early this year, Addleshaw Goddard, a UK-based law firm, encountered a surge of 178 applications for a single IT role, a significant increase from previous years. To efficiently manage this volume and improve the selection process, the firm implemented an AI agent powered by its internal Microsoft Copilot system to evaluate candidate resumes alongside human recruiters.
The AI quickly assessed the shortlisted applications but produced a ranking that diverged notably from human evaluators. The AI favored polished CVs boasting exceptional accomplishments despite limited experience—applications that appeared to bear the hallmark of AI enhancement. This discrepancy raised critical questions about the AI’s evaluative criteria.
Katherine Rathbone, head of HR at Addleshaw Goddard, speculated that because AI competency was a specified job criterion, the system might have interpreted signs of AI-assisted resume creation as proof of the candidate’s AI skills. This raises concerns that AI-driven recruitment tools might inadvertently prioritize AI-crafted resumes, creating a feedback loop that favors certain candidates solely on the basis of their AI usage.
Supporting this concern, recent research analyzing AI hiring tools found that candidates employing the same AI language models as recruiters were 23% to 60% more likely to be shortlisted, compared with those submitting human-generated CVs. This ‘self-preferencing’ effect was most pronounced in business-related roles like sales and accountancy.
Such findings are particularly troubling in the current hiring climate, where candidates increasingly utilize AI to craft applications in response to automated screening systems. Bias towards AI-generated documents risks disadvantaging applicants who either do not have access to or choose not to use such technology.
Encouragingly, research indicates that this bias can be mitigated by instructing AI models to disregard resume origin and focus solely on content quality. Additionally, incorporating multiple AI assessment models can reduce the influence of any individual tool’s bias.
At Addleshaw Goddard, the final hiring decision remained a human prerogative, with the successful candidate being one who did not top the AI-generated rankings. This highlights the ongoing necessity of human judgment in recruitment to balance AI-driven insights with nuanced understanding.
As AI integration in hiring evolves, organizations should remain cautious of potential biases and proactively apply strategies to ensure equitable evaluation. Embracing AI’s potential in recruitment requires mindful implementation, combining automated tools with human oversight to foster fair and effective hiring practices.
Source: Channel News Asia