Products of Positionality: How Tech Workers Shape Identity Concepts in Computer Vision

Title of the Paper

Products of Positionality: How Tech Workers Shape Identity Concepts in Computer Vision

Paper Information

• Research Domain: Artificial Intelligence, Computer Vision, Identity Concepts and Bias in Human-Computer Interaction
• Keywords: Tech Workers, Identity, Positionality, Computer Vision, Work Studies, Machine Learning

Research Background and Issues

• Identified Problems or Challenges:

  • Computer vision technology exhibits significant identity-related bias in classifying and recognizing human identities, such as gender classification and racial definitions.
  • Existing research primarily focuses on the role of data workers in contributing to bias, with limited attention on the role of tech workers who design and define identity concepts.
  • The "positionality" of tech workers—their values, experiences, and social contexts—has profound impacts on product design but remains underexplored.

• Significance:

  • Identity classification is a core component of computer vision, yet inaccuracies or biases in identity classification can lead to social discrimination and implicit inequities.
  • Understanding the role of positionality in product development is crucial for improving the fairness and accuracy of computer vision products.

• Research Motivation and Related Work:

  • Investigate how tech workers embed their positionality into the design of computer vision products.
  • Analyze how positionality is shaped and constructed through team interactions and corporate environments.
  • Complement existing research on machine learning bias by examining the influence of tech workers on the definition of identity concepts.

Solution

• Proposed Method or Solution:

  • Conduct semi-structured interviews with computer vision practitioners in the tech industry, including researchers, engineers, and project managers.
  • Focus on analyzing how participants define identity concepts based on their personal positionality and how they collaborate or conflict with colleagues within corporate environments.

• Innovative Aspects of the Solution:

  • Introduce positionality theory into the field of computer vision research, emphasizing the impact of tech workers' social, cultural, and economic backgrounds on product design.
  • Propose a multi-layered contextual research framework (macro-social, development environment, corporate level) to dissect how positionality influences identity-related decisions.

• Implementation Steps and Key Techniques:

  1. Participant Recruitment: Recruit 24 computer vision practitioners through direct outreach and snowball sampling methods.
  2. Interview Design: Develop flexible interview questions to investigate identity concepts, organizational environments, and specific product challenges.
  3. Data Analysis: Apply framework coding and memo strategies to extract themes, interpreting the data through the authors' research perspective.

Research Findings

• Specific Findings:

  • Tech workers' personal positionality plays a dominant role in defining identity classifications (e.g., gender, race), with decisions influenced by contextual factors such as economic constraints, corporate policies, and values.
  • Positionality differences among colleagues can lead to conflicts and complex negotiation processes within teams.
  • Failure to consider alternative perspectives or positionalities may result in product failures, such as embedded biases or poor user experiences.

• Advantages Compared to Existing Solutions:

  • Analyze the root causes of bias by examining how tech workers' social environments and personal experiences influence product design, rather than focusing solely on bias mitigation outcomes.
  • Propose a comprehensive framework for analyzing positionality throughout the design process.

• Experimental or Evaluation Results:

  • Resource disparities between companies significantly affect tech workers' ability to achieve fairness and responsible design.
  • Larger companies offer more economic resources to support ethical design, while smaller companies are more driven by market demands.
  • Promoting team diversity during product development helps mitigate individual positionality limitations.

• Limitations and Future Directions:

  • The richness of tech workers' positionality is challenging to capture fully, and the study may lean toward observable explicit characteristics.
  • Future research should explore more effective ways to document and utilize positionality information to establish flexible, culturally and socially adaptive technology design processes.

Conclusion and Implications

• Tech workers' positionality profoundly influences the design of identity classification in computer vision products, and current technical practices need to proactively address positionality.
• Two approaches are proposed to address positionality issues: (1) Contextual level, including social background, corporate values, and industry regulations; (2) Individual level, emphasizing the unique perspectives of tech workers, data workers, and users.
• Developers and researchers are encouraged to critically examine positionality issues to avoid design blind spots that lead to bias, fostering the creation of more equitable and responsible AI products.