Algorithmic Power or Punishment: Information Worker Perspectives on Passive Sensing Enabled AI Phenotyping of Performance and Wellbeing

Title of the Paper

Algorithmic Power or Punishment: Information Worker Perspectives on Passive Sensing Enabled AI Phenotyping of Performance and Wellbeing

Paper Information

• Subject Area: Human-Computer Interaction, Passive Sensing in Information Workplaces, and the Impact of AI Algorithms
• Keywords: Passive Sensing, Algorithmic Phenotyping, Information Workers, Human Resource Management, Employee Wellbeing, AI Ethics, Privacy, Data Flow Dynamics, Workplace Surveillance

Research Background and Issues

• Identified Issues:

  1. Passive Sensing AI (PSAI) technologies are driving the evaluation of information workers' "digital phenotypes" in terms of performance and wellbeing, raising concerns about privacy violations and power imbalances.
  2. Within organizations, these systems carry risks of misuse (e.g., privacy breaches or data being used for negative purposes) and often provide little direct benefit to employees.
  3. The adoption of PSAI by companies to facilitate remote and hybrid work may force employees to compromise between privacy and convenience.

• Research Significance: Passive Sensing AI has the potential to balance improving employee work experiences with organizational efficiency. However, understanding the expectations of data subjects (workers) is crucial to guide technology design and policy regulations, preventing further exacerbation of power imbalances.

• Research Motivation and Related Work:

  1. Engages with existing literature on organizational surveillance, privacy, and employee monitoring, drawing on Foucault's theories of control to explore how technology seeks to balance enhancing wellbeing with reinforcing power structures.
  2. Emphasizes the Contextual Integrity framework to analyze the appropriateness and distribution norms of information flows in social and workplace contexts.

Solution

• Proposed Methods and Approach: The authors conducted scenario-based interviews with 28 information workers to explore their perspectives on PSAI, particularly the appropriateness of algorithmic predictions of performance and wellbeing ("normative appropriateness") and the reasonableness of sharing related data with various stakeholders ("distributional appropriateness").

• Innovations:

  1. Introduced the Contextual Integrity framework to study the dynamics of information flows in specific workplace contexts.
  2. Used scenario comparison methods to encourage participants to reflect and reimagine the possibilities of data flows and AI applications.
  3. Centered workers' perspectives, proposing a "worker-centered PSAI" design direction that benefits data subjects.

• Implementation Steps:

  1. Designed and presented seven workplace scenarios based on real AI applications, covering different types of sensing technologies and data distribution methods.
  2. Discussed with participants the applicability, privacy, security, and sharing conditions of these scenarios.
  3. Conducted thematic analysis of interview data using the Contextual Integrity framework to organize participants' perspectives.

• Key Technologies:

  1. Passive sensing technologies: such as CCTV cameras, PC activity tracking, communication data logs, etc.
  2. Human-computer interaction frameworks: scenario evaluation and user reflection methods.
  3. Algorithm evaluation: exploring the potential of PSAI for human resource management and employee self-optimization.

Research Findings

• Specific Findings:

  1. Insights on "Normative Appropriateness":
     • The applicability of PSAI depends primarily on its impact on work outcomes, work-life boundary management, and flexibility.
     • PSAI is seen as potentially beneficial if it helps employees improve performance feedback, develop better work habits, and support self-development.
     • However, employees are generally concerned about potential misuse of their data, leading to career limitations or additional stress.
  2. Insights on "Distributional Appropriateness":
     • Workers desire greater control over the flow of AI-generated data, particularly regarding the interpretation and timing of shared content.
     • Employees prefer sharing data with direct managers rather than HR departments and advocate for clear data-sharing policies.

• Advantages Over Existing Solutions:

  • Emphasizes the worker's perspective, focusing on protecting privacy and autonomy.
  • Proposes new visions for "human reappraisal" and distributed data sharing from the worker's standpoint.

• Experimental or Evaluation Results:

  1. Scenario-based interviews revealed that PSAI must address diverse individual work preferences and privacy needs.
  2. The study suggests balancing algorithmic accuracy with interactive transparency to make results more understandable and acceptable.

• Limitations and Future Directions:

  1. This study focuses on the perspectives of information workers and does not sufficiently address the viewpoints of other stakeholders (e.g., employers or developers).
  2. While the interview and scenario simulation methods are effective, they may not capture the complexity of PSAI in real-world environments.
  3. Future Directions:
     • Expand investigations to include more types of work (e.g., freelancers or logistics personnel).
     • Conduct field studies based on actual PSAI technologies to complement the depth of scenario-based interviews.
     • Advocate for multi-stakeholder participation in algorithm design to explore the comprehensiveness and sustainability of PSAI.