Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From Perceived Effectiveness to Measured Impact: Identity-Aware Evaluation of Automated Counter-Stereotypes

Svetlana Kiritchenko, Anna Kerkhof, Isar Nejadgholi, Kathleen C. Fraser

TL;DR

The paper tackles the challenge of reducing gender bias on social media through automatically generated counter-stereotypes, testing two strategies—Broadening Universals and Counterfacts—via an online, between-subjects experiment with over 1,200 participants. Using the Implicit Association Test ($D$-scores) and explicit bias measures, along with a Becker-DeGroot-Marschak mechanism to gauge perceived utility, the study finds that actual bias reduction is limited and varies by demographic subgroup, often diverging from perceived effectiveness. Notably, older participants (especially men) show some implicit-bias reductions for certain interventions, while younger participants (especially women) may exhibit increased bias under the same treatments. The results underscore the need for dynamic, identity-aware interventions and longitudinal, cross-cultural evaluations to achieve meaningful, scalable reductions in online stereotypes, highlighting the gap between perceived and real impact and the importance of human-centered, interdisciplinary design.

Abstract

We investigate the effect of automatically generated counter-stereotypes on gender bias held by users of various demographics on social media. Building on recent NLP advancements and social psychology literature, we evaluate two counter-stereotype strategies -- counter-facts and broadening universals (i.e., stating that anyone can have a trait regardless of group membership) -- which have been identified as the most potentially effective in previous studies. We assess the real-world impact of these strategies on mitigating gender bias across user demographics (gender and age), through the Implicit Association Test and the self-reported measures of explicit bias and perceived utility. Our findings reveal that actual effectiveness does not align with perceived effectiveness, and the former is a nuanced and sometimes divergent phenomenon across demographic groups. While overall bias reduction was limited, certain groups (e.g., older, male participants) exhibited measurable improvements in implicit bias in response to some interventions. Conversely, younger participants, especially women, showed increasing bias in response to the same interventions. These results highlight the complex and identity-sensitive nature of stereotype mitigation and call for dynamic and context-aware evaluation and mitigation strategies.

From Perceived Effectiveness to Measured Impact: Identity-Aware Evaluation of Automated Counter-Stereotypes

TL;DR

The paper tackles the challenge of reducing gender bias on social media through automatically generated counter-stereotypes, testing two strategies—Broadening Universals and Counterfacts—via an online, between-subjects experiment with over 1,200 participants. Using the Implicit Association Test (-scores) and explicit bias measures, along with a Becker-DeGroot-Marschak mechanism to gauge perceived utility, the study finds that actual bias reduction is limited and varies by demographic subgroup, often diverging from perceived effectiveness. Notably, older participants (especially men) show some implicit-bias reductions for certain interventions, while younger participants (especially women) may exhibit increased bias under the same treatments. The results underscore the need for dynamic, identity-aware interventions and longitudinal, cross-cultural evaluations to achieve meaningful, scalable reductions in online stereotypes, highlighting the gap between perceived and real impact and the importance of human-centered, interdisciplinary design.

Abstract

We investigate the effect of automatically generated counter-stereotypes on gender bias held by users of various demographics on social media. Building on recent NLP advancements and social psychology literature, we evaluate two counter-stereotype strategies -- counter-facts and broadening universals (i.e., stating that anyone can have a trait regardless of group membership) -- which have been identified as the most potentially effective in previous studies. We assess the real-world impact of these strategies on mitigating gender bias across user demographics (gender and age), through the Implicit Association Test and the self-reported measures of explicit bias and perceived utility. Our findings reveal that actual effectiveness does not align with perceived effectiveness, and the former is a nuanced and sometimes divergent phenomenon across demographic groups. While overall bias reduction was limited, certain groups (e.g., older, male participants) exhibited measurable improvements in implicit bias in response to some interventions. Conversely, younger participants, especially women, showed increasing bias in response to the same interventions. These results highlight the complex and identity-sensitive nature of stereotype mitigation and call for dynamic and context-aware evaluation and mitigation strategies.

Paper Structure

This paper contains 19 sections, 10 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Psychological Studies on Countering Stereotypes
  4. Identifying and Tackling Stereotypes with NLP
  5. Study Design
  6. Study Overview
  7. Conversation Generation
  8. Control Conversations:
  9. Stereotypes and Counter-stereotypes:
  10. Implicit Association Test (IAT)
  11. Measuring Explicit Gender Bias
  12. Assessing Perceived Utility of Counter-Stereotypes
  13. Data Verification
  14. Results
  15. Implicit Bias
  16. ...and 4 more sections

Figures (10)

  • Figure 1: The mean D-scores for the four treatment groups in the main and the pilot studies. Error bars indicate 95% confidence intervals.
  • Figure 2: The mean D-scores for the four treatment groups for male and female participants in the main study. Error bars indicate 95% confidence intervals.
  • Figure 3: The mean D-scores for the four treatment groups for younger ($\leq$35 y.o.) and older (>35 y.o.) participants in the main study. Error bars indicate 95% confidence intervals.
  • Figure 4: The mean D-scores for the four treatment groups for younger ($\leq$35 y.o.) and older (>35 y.o.) male and female participants in the main study. Error bars indicate 95% confidence intervals.
  • Figure 5: The mean explicit stereotype scores for the four treatment groups for male and female participants in the main study. Error bars indicate 95% confidence intervals.
  • ...and 5 more figures