What Pronouns for Pepper? A Critical Review of Gender/ing in Research

Title of the Paper

Which Pronoun Does Pepper Use?—A Critical Review of Gendering in Research

Paper Information

• Subject Area: Human-Computer Interaction (HCI), Gendering in Social Artificial Intelligence Agents
• Keywords: Gender, Social Artificial Intelligence Agents, Humanoid Robots, Pepper, User Perception

Research Background and Issues

• Issues and Challenges:

  • Social artificial intelligence agents (e.g., voice assistants, humanoid robots) are often influenced by gendering during design, including users consciously or unconsciously attributing gender characteristics to these agents.
  • Despite existing critiques of gendered design, there is still limited understanding of how gendering issues are addressed in academic research.
  • A systematic exploration of researchers' unconscious gendering behaviors during design and research is lacking, which may affect the validity and representativeness of research outcomes.

• Significance:

  • Failure to address gendering issues in research may lead to biased studies and reinforce existing gender stereotypes, thereby impacting social equity.
  • The gendering of artificial intelligence agents not only affects user experience but also reflects researchers' perspectives and implicit assumptions, necessitating critical reflection.

• Motivation and Related Work:

  • Gendering phenomena are prevalent in the design and use of social agents, such as the default use of "female" voices in voice assistants.
  • The humanoid robot Pepper was officially designed to be gender-neutral, yet it is often gendered as male or female in practice.
  • Encouraging researchers to apply critical reflection in their practices promotes a deeper understanding and discussion of the variable "gender."

Solutions

• Research Methods:

  • This paper systematically reviewed 75 HCI research papers related to the Pepper robot.
  • Using meta-synthesis and directed content analysis, the study specifically examined how researchers and participants attributed gender to Pepper (e.g., through pronoun usage).
  • A critical framework and checklist were proposed to help future researchers address gender issues more openly and critically.

• Innovations:

  • Provided preliminary empirical evidence of how researchers' gendering of Pepper influences participants' gendering.
  • Developed a theoretical and practical framework based on critical theory and human-centered design to address the causes of gendering phenomena.
  • Systematically summarized six major challenges of gender issues in related research, including "invisibility," "variability," "decentralization," and "neutrality."

• Implementation Steps and Techniques:

  • Conducted literature search and screening to compile research data.
  • Performed qualitative and quantitative data analysis on the sampled literature, including frequency statistics of pronoun usage and categorization of gender perceptions.
  • Derived specific gendering issues and key questions within the proposed framework.

Research Findings

• Specific Findings:

  • Researchers predominantly referred to Pepper using the pronoun "it" (63%), but in some cases, it was gendered as female or male.
  • Participants exhibited significant variation in attributing gender to Pepper, including "it," "masculinization," "feminization," and "neutralization."
  • Only 8% of studies conducted manipulation checks on gender attribution, and most studies did not address gender issues.
  • Gendering was often unconsciously influenced or directly determined by researchers, reflecting either consistency or divergence in gender attribution between participants and researchers.

• Advantages Compared to Existing Solutions:

  • Highlighted how researchers' gendering behaviors could influence participant data, proposing reflexive measures to avoid bias.
  • Provided a series of easily applicable self-check questions and operational suggestions, addressing the gap in guidance on handling gendering in research practices.

• Experimental or Evaluation Results:

  • Many studies employed a binary gender model, neglecting broader gender possibilities and failing to fully capture non-binary gender individuals or outcomes.
  • Only a limited number of studies actively discussed or tested Pepper's gender attribution, with gender consistency remaining ambiguous.

• Limitations and Future Directions:

  • Limitations:
    • Data was limited to publicly available literature, making it impossible to evaluate unpublished research processes and materials.
    • Case studies focused on the Pepper robot, potentially limiting generalizability.
    • Did not comprehensively analyze the potential relationship between researchers' gender and gendering behaviors.
  • Future Directions:
    • Expand to other types of social artificial intelligence agents (e.g., other humanoid robots).
    • Use survey and interview methods to gain deeper insights into researchers' motivations and perspectives on gendering issues.
    • Explore how cultural factors influence gendering, particularly the impact of language, ethnicity, and social environment on gender perceptions.

Conclusion

This paper provides important empirical evidence and a critical framework for addressing gendering issues in social artificial intelligence agents. The study calls for researchers to maintain reflexivity regarding gender issues in design and research and to enhance the theoretical depth and methodological rigor of future studies through a clear operational framework.