Research Summary: Do job seekers value procedure in AI hiring only for error correction? Evidence from a conjoint experiment

Research from arXiv explores job seekers' perceptions of AI-driven hiring systems, specifically investigating whether procedural fairness elements are valued independently or primarily for their role in error correction. A conjoint experiment with 1,919 US job seekers found that certain procedural safeguards like appeal, opt-out, and independent bias audits were not valued as expected, suggesting their perceived worth might be tied more to performance and error mitigation rather than inherent procedural value.

Why it matters

This research is strategically important because it informs the design and deployment of AI in human resources, particularly in talent acquisition. Understanding job seeker perceptions is crucial for maintaining public trust, ensuring fair hiring practices, and optimizing the effectiveness and acceptance of automated decision-making systems across various industries.

Key insights

• Employers are increasingly using automated systems for initial job applicant screening, often rejecting candidates without human review.
• Applicant acceptance of these AI systems likely depends on both system performance and the procedural fairness of the decision-making process.
• Prior studies have not sufficiently isolated the impact of procedure from performance, making it unclear if procedure is valued intrinsically or for error correction.
• A preregistered conjoint experiment with 1,919 US job seekers evaluated preferences for AI hiring systems based on decision authority, error rate, explanation, opt-out, appeal, and independent bias audit.
• The study observed that job seekers did not significantly value the inclusion of an appeal mechanism, an opt-out option, or an independent bias audit.
• The source text is an abstract, and the full findings regarding the value of these procedures (e.g., whether they were valued only for error correction) are not fully disclosed, as the sentence ends mid-thought.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.16390