A Scoping Review of Gender Stereotypes in Artificial Intelligence

Research Background and Issues

• What problems or challenges have the authors identified?

  • People often attribute gender stereotypes to artificial intelligence (AI) applications, and AI design frequently reinforces these stereotypes, perpetuating traditional gender norms in emerging technologies.
  • Current research on AI gender stereotypes suffers from conceptual ambiguity and inconsistency, lacking systematic reviews to clarify their actual impact.

• Why is this issue important?

  • Embedding gender stereotypes into AI design may exacerbate gender inequality and reflect outdated societal values and power structures, hindering technological and social progress.
  • As AI technologies (e.g., generative AI) become more widely adopted, these technologies may further reinforce stereotypes through biases inherent in their training data.

• Research Motivation and Related Work

  • This study aims to fill the gap in existing reviews by providing a more comprehensive perspective on the issue of AI gender stereotypes.
  • The authors emphasize that the goal of the research is to identify the specific manifestations of gender stereotypes in AI and to promote responsible AI design to mitigate these harmful effects.

Solutions

• What methods or solutions have the authors proposed?

  • The authors adopted a normative literature review approach to summarize and categorize research on gender stereotypes in human-computer interaction (HCI), human-robot interaction (HRI), and related social science fields over the past 20 years.
  • The study developed a trait-based and domain-based classification of gender stereotypes to provide a structured framework for future research.

• What are the innovative aspects of this solution?

  • This research is the first systematic literature review on AI gender stereotypes, identifying the operational methods for most AI-related gender stereotypes.
  • It offers two core classifications: trait-based stereotypes (e.g., communication style, competence, and agency) and domain-based stereotypes (e.g., context, task allocation, or professional fields).

• What are the implementation steps? What key techniques were used?

  • The PRISMA literature screening and systematic data extraction methods were used to screen 445 studies, ultimately identifying 73 relevant papers.
  • A framework was designed to explain the multidimensional mechanisms of gender stereotypes, focusing on their manifestations and operationalization.

Research Findings

• What specific findings were obtained?

  • The study identified two major types of AI gender stereotypes:
    1. Trait-based stereotypes: For example, male-associated traits like agency and competence, and female-associated traits like warmth and care.
    2. Domain-based stereotypes: For instance, female AIs are perceived as more suitable for domestic, service, and caregiving roles, while male AIs are seen as better suited for mechanical, technical, and STEM (science, technology, engineering, mathematics) fields.
  • The research found that removing gender information from AI does not completely eliminate gender stereotypes, as implicit gender cues may still trigger biases.

• What advantages does this solution have compared to existing ones?

  • It provides a comprehensive framework that systematically reveals how gender stereotypes manifest in different contexts, enabling future research and design to address these issues more precisely.
  • The literature review not only examines experimental studies but also explores the unique manifestations and dynamics of gender stereotypes in real-world scenarios.

• What were the experimental or evaluation results?

  • Multiple experiments found that user experiences were more positive when AI aligned with gender stereotypes, but some users expressed discomfort and skepticism when AI challenged these stereotypes.
  • Experiments showed that certain design elements, such as gendered voices or names, could trigger specific stereotypes.
  • Explicitly challenging gender stereotypes (e.g., through voice content) effectively reduced biases against certain types of AI, but the effectiveness depended on users' gender and cultural backgrounds.

• Limitations and Future Directions

  • Limitations:

    • Almost all existing studies are conducted in Global North societies, lacking insights into Global South cultural contexts.
    • Experiments primarily focus on individual or dyadic interactions, with limited exploration of stereotypes in team or group human-AI collaboration scenarios.
    • Research on non-binary gender AI design and stereotypes is severely lacking.

  • Future Directions:

    • Advocate for more field studies and qualitative methods to capture the real and complex characteristics of gender stereotypes.
    • Explore gender stereotypes in team interaction scenarios involving AI, as well as their further application in workplaces and other domains.
    • Design more non-binary and gender-ambiguous AI systems to promote diversity and inclusivity in technology development, while testing user reactions and stereotypes.
    • Develop specialized measurement tools to systematically evaluate gender-related stereotypes in AI.

The above provides a comprehensive analysis and summary of the paper, adhering to academic standards and cutting-edge developments in the field.