Can Voice Assistants Be Microaggressors? Cross-Race Psychological Responses to Failures of Automatic Speech Recognition

Title of the Paper

Can Voice Assistants Be Microaggressors? Cross-Race Psychological Responses to Failures of Automatic Speech Recognition

Paper Information

• Subject Area: Human-Computer Interaction (HCI), Language Technology, Psychology
• Keywords: Language Technology, Voice Assistants, Automatic Speech Recognition, Race, Microaggressions, Psychological Impact, Quantitative Analysis, Technology Evaluation

Research Background and Problem

• Identified Problem or Challenge: Automatic speech recognition technology exhibits differential error rates for users of different racial backgrounds, with Black users experiencing higher rates of recognition errors. This phenomenon may not only affect the functionality of the technology but could also be perceived by Black users as microaggressions (e.g., subtle racial discrimination in social interactions).
• Significance: Voice assistants have become widely adopted globally, playing a critical role in daily life and essential tasks (e.g., emergency calls). However, the poorer performance of these technologies for certain user groups not only diminishes user experience but may also negatively impact mental health and social belonging.
• Research Motivation and Related Work:
  • Previous studies have revealed higher error rates in automatic speech recognition for Black users, attributing this to the underrepresentation of Black voice samples in training datasets.
  • Psychological research indicates that racial microaggressions significantly affect the mental state of minority groups. However, methods to measure these impacts within technological systems remain underexplored. Existing studies primarily focus on the "behavioral" bias of technology rather than the "harmful" consequences.

Solution

• Methods and Solutions:
  • Propose a framework centered on "harm" rather than "behavior" to study the psychological impact of automatic speech recognition systems on Black users.
  • Design and conduct a controlled experiment in which Black and White participants interact with voice assistants preset to high or low error rates, quantifying the potential psychological effects of speech recognition errors.
• Innovations:
  • Apply the concept of racial microaggressions from psychology to the field of human-computer interaction.
  • Provide empirical quantification of the potential psychological harm caused by voice assistants.
• Implementation Steps and Key Techniques:
  • The experiment employed a 2x2 between-subjects design (racial identity and error rate conditions), with participants randomly assigned to high or low error rate groups.
  • Used the "Wizard of Oz" method, where researchers simulated voice assistant interactions using text-to-speech tools.
  • Validated psychological measurement tools (e.g., emotional response scales, self-awareness scales, self-esteem scales) were used to assess participants' subjective experiences.

Research Findings

• Specific Findings:
  • The proposed harm-based framework successfully revealed the psychological impact on Black participants under high error rate conditions, while White participants showed no significant response to error rate conditions.
  • Black participants exhibited significant changes in metrics, including heightened self-awareness, reduced positive emotions, and lower personal and group self-esteem.
  • Black participants rated the voice assistant's performance significantly lower under high error rate conditions, whereas White participants did not exhibit similar subjective evaluation differences.
• Advantages Compared to Existing Solutions:
  • Combines psychological theories with HCI experimental methods to deeply explore the non-functional impacts of technology on racial users.
  • Provides reliable quantitative data to support the design of technological solutions aimed at reducing potential psychological harm.
• Experimental or Evaluation Results:
  • High error rate conditions significantly reduced the psychological state and technology acceptance of Black participants, confirming the applicability of the racial microaggressions theory in technological contexts.
  • White participants did not exhibit similar psychological harm differences, highlighting the influence of racial privilege on technology experiences.
• Limitations and Future Directions:
  • The experimental environment may not fully reflect real-life interactions with voice assistants. Additionally, the study focused on short-term psychological effects and did not examine long-term cumulative harm.
  • Future research could expand the sample population to include more racial, linguistic, and social identity backgrounds to explore intersectional impacts.
  • Combining real-world user interaction logs with voice assistants and long-term longitudinal studies could evaluate the health and psychological consequences of technological bias over time.
  • Future designs should aim to mitigate microaggressions, such as incorporating feedback mechanisms in voice assistants to address negative experiences among minority users.

Design Recommendations

• Technical Improvement Directions:
  • Develop feedback mechanisms in voice assistants that explicitly recognize and address microaggressions, such as acknowledging errors and providing social affirmation.
  • Offer culturally sensitive and identity-affirming design solutions while increasing the diversity of voice data used for training.
• Community Involvement and Co-Design:
  • Adopt participatory design methods, involving affected communities in the technology development process to ensure design strategies genuinely reflect their needs.

Conclusion

This study demonstrates the importance of the racial microaggressions theory in researching technological bias and reveals the psychological harm caused by high-error-rate voice assistants to Black users. The research not only advances the understanding of technological racial bias and its consequences in the HCI field but also provides practical guidance for designing more inclusive voice technologies.