Designing Fair AI in Human Resource Management: Understanding Tensions Surrounding Algorithmic Evaluation and Envisioning Stakeholder-Centered Solutions

Title of the Paper

Designing Fair AI in Human Resource Management: Understanding Tensions Surrounding Algorithmic Evaluation and Envisioning Stakeholder-Centered Solutions

Paper Information

• Subject Area: AI design and ethical issues in human resource management
• Keywords: Artificial Intelligence (AI), Fair and Responsible AI, Explainable AI (XAI), Future of Work, Algorithmic Management, Human Resource Management (HRM), Stakeholder-Centered Design, Explainability, Transparency, Human Intervention

Research Background and Issues

• Problem Identification: As companies introduce AI into HRM, it is expected to address biases in traditional performance evaluations. However, the application of AI has triggered conflicts among stakeholders, such as privacy concerns, negative emotions, and differences in perceptions of fairness.
• Importance of the Problem: Human resource management not only impacts organizational performance but also directly affects employee well-being, job satisfaction, and livelihoods. Traditional performance management systems face issues such as politics, biases, and human errors, further eroding employee trust.
• Research Motivation and Related Work: Although AI theoretically reduces subjective biases, its practical adoption often leans towards supervision and monitoring rather than addressing traditional issues. Existing studies predominantly explore AI from the employee perspective, neglecting the complex relationships involving multiple stakeholders.

Solutions

• Proposed Methods/Solutions:
  • The authors adopt a stakeholder-centered design approach, organizing participatory design workshops that include diverse perspectives (e.g., employees, employers, AI experts) to capture core conflicts and tensions in AI applications within HRM.
  • To address these tensions, specific AI design strategies are proposed, such as partial transparency, explainability, and human oversight of algorithmic decisions.
• Innovative Contributions:
  • Defines AI design in HRM as a "wicked problem" and introduces the concept of tensions among stakeholders.
  • Combines Roberts' strategies (authorization, competition, collaboration) with iterative design thinking to identify and refine tension points.
• Implementation Steps:
  1. Conducting workshops and in-depth interviews based on real-world scenarios.
  2. Gradually identifying stakeholder needs and conflicts through phased approaches of authoritative design, competitive design, and collaborative design.
  3. Employing iterative design methods to continuously review and optimize the design process.
  • Key Technologies: Focus on balancing core features such as transparency, explainability, and accuracy, alongside external audit mechanisms as socialized solutions.

Research Outcomes

• Specific Findings:
  • Identified five major tensions:
    1. Differences in fairness perceptions among stakeholders.
    2. Trade-offs between AI accuracy, implementation costs, and privacy concerns.
    3. Challenges in designing transparency for algorithms and decision-making processes.
    4. Pros and cons of algorithmic decision explainability.
    5. Balancing work efficiency with the risk of dehumanization.
  • Proposed design insights to address each tension, such as achieving process credibility through partial transparency, enhancing human intervention by HR teams, and using explainability as a tool to help employees improve rather than merely judge them.
• Advantages Compared to Existing Solutions:
  • While most current solutions focus on specific stakeholder needs (e.g., employee perspective), this paper emphasizes comprehensive stakeholder-centered design.
  • Places AI design within a broader socio-technical context, marking the first instance of viewing HRM as a high-risk socio-technical environment.
• Experimental or Evaluation Results: Results from participatory design workshops indicate that stakeholders can reach consensus on fairness issues under certain conditions, while identifying specific feasible design directions.
• Limitations and Future Directions:
  • The study is based on scenario assumptions and lacks direct research on the real-world effects of AI in HRM.
  • Future research should explore real-world corporate cases to validate the theoretical framework.
  • Further investigation is needed to precisely balance different types of fairness and optimize the trade-off between protecting sensitive attributes and enhancing accuracy.

Conclusion

This paper systematically analyzes potential conflicts in AI applications within human resource management and proposes a novel stakeholder-centered approach. Its design concepts and methods offer valuable insights for future exploration of fairness and transparency in high-risk AI systems, while advancing research that integrates AI's social applications into broader ethical and managerial perspectives.