Commentary: What if the HR department's AI prefers resumes written by AI?

A recent experiment at the law firm Addleshaw Goddard demonstrated how AI systems may inadvertently favor applicants who have used AI to craft their resumes. As part of the experiment, the firm's HR department utilized Microsoft Copilot to evaluate job applications, comparing the agent's selections against those of human recruiters.

Surprisingly, the AI identified candidates whose resumes appeared more polished and impressive, often reflecting achievements and projects that seemed too good to be true. The discrepancy between the AI-selected resumes and those chosen by humans raised concerns about the system's bias towards AI-generated content. Researchers have found that AI hiring tools can lead to a significant increase in the chances of candidates whose resumes were generated using AI, ranging from 23% to 60% more likely to be shortlisted compared to those with human-written resumes.

The phenomenon, known as self-preferencing behavior, has the potential to exacerbate an already competitive hiring landscape. Experts suggest that to mitigate such bias, AI models should be instructed to disregard the source of resumes and focus solely on their content, and multiple assessment models should be employed to reduce the influence of any single tool.