A decade ago, the retailer built an internal screening tool to rank job applications and surface promising candidates for human recruiters. The system trained on historical hiring data, most of which came from successful male applicants, and it used those patterns to judge new resumes.
Over several years, the model downgraded indicators associated with women, making gender a hidden filter in the process. By 2018, the company shut the project down after seeing how the algorithm’s learning had gone off course.
Recruitment has since become one of the busiest testing grounds for AI inside workplaces. Employers now deploy automated tools to sift large volumes of resumes, triage applicants and highlight those who most closely match a role’s criteria.
Job seekers use generative systems to draft cover letters, tweak resumes and mirror key words they believe the bots scan for. The cycle turns a traditionally human conversation into one where algorithms talk to other algorithms before any person steps in.
Automation promises to strip out some subjectivity and speed up hiring, yet the early Amazon experience shows how training data can quietly tilt decisions. As both sides in the job market lean on AI to manage repetitive tasks, questions grow about transparency, accountability and who gets left out of the shortlists.
