Not Just a Preference: Reducing Biased Decision-making on Dating Websites

Title of the Paper

Not Just a Preference: Reducing Biased Decision-making on Dating Websites

Publication Information

• Authors: Zilin Ma, Krzysztof Z. Gajos
• Conference: CHI Conference on Human Factors in Computing Systems (CHI ’22)
• Date of Publication: April 29–May 5, 2022
• DOI: 10.1145/3491102.3517587
• Research Areas: Human-Computer Interaction, Social Psychology, Technology Ethics
• Keywords: Online dating, biased decision-making, implicit bias, HCI design, racial discrimination, user experience, design interventions, bias issues

Research Background and Problem Statement

1. Issues and Challenges:

   • Online dating websites serve as an important platform for many individuals seeking intimate partners, yet current designs may exacerbate or reinforce users' racial biases.
   • Although some users claim not to consider race as a factor, experiments show that implicit biases still influence their decision-making processes.
   • Popular interaction methods (e.g., swipe interfaces and match scores) may amplify such implicit biases.

2. Significance:

   • Racial bias hinders connections between different social groups and harms marginalized communities.
   • Online dating services have the potential to foster relationships and promote social equity, but poor design choices can amplify discriminatory tendencies.

3. Motivation and Related Work:

   • The research explores how technology influences social psychology and the manifestation of biases.
   • Previous studies indicate that first impressions based on facial photos profoundly impact decision-making, highlighting flaws in current interface designs.
   • This paper aims to design and test interfaces that reduce bias and validate their effectiveness through experiments.

Solution

1. Method or Solution:

   • Proposed a Reversed Sequence Interface design, which presents users with personal descriptive information related to selection criteria first, followed by photos and names, to reduce intuitive decisions based on visual cues.
   • Tested an Update Design interface, allowing users to view match scores after their initial decision and adjust their choices based on new information.

2. Innovations:

   • Introduced new interaction models addressing users' implicit racial biases, rather than relying solely on algorithmic optimization or user education.
   • Focused on the relationship between implicit bias and information presentation order, offering actionable design improvement suggestions.

3. Implementation Steps and Key Technologies:

   • Developed a simulated dating website to test the effectiveness of different interface designs.
   • Experimental design included control groups (swipe interface and match score interface) and experimental groups (Reversed Sequence Interface and Update Design interface).
   • Presented virtual dating profiles (including photos, names, and textual descriptions) to 1,907 participants and analyzed their selection behaviors.

Research Findings

1. Key Results:

   • Users of the swipe interface exhibited significant anti-Black racial bias, even when they claimed race was not a deciding factor.
   • The Reversed Sequence Interface significantly reduced bias, particularly among users who stated race was not a deciding factor.
   • The Update Design interface and the combined Reversed Sequence and Update interface were ineffective in mitigating bias.

2. Innovations and Advantages:

   • The Reversed Sequence Interface encouraged users to spend more time focusing on textual descriptions, thereby reducing the influence of racial information in photos on decision-making.
   • Compared to existing designs, this approach significantly reduced implicit bias without altering users' explicitly stated preferences.

3. Experimental Data and Evaluation:

   • Participants spent more time reviewing each profile in the Reversed Sequence Interface (median time increased from 13.9 seconds to 15.04 seconds).
   • Experimental results showed bias scores significantly lower than those in the control group (reduced from 0.043 to 0.012).

4. Limitations and Future Directions:

   • The gender and racial distribution of participants may not fully represent real-world dating website users.
   • The study primarily focused on anti-Black bias; future research could expand to biases against other minority groups.
   • Non-binary gender users were not included, necessitating further research on comprehensive gender impacts.
   • Future studies should explore the long-term effects of design interventions on the entire dating process, beyond initial decision-making.

Conclusion

The study demonstrates that swipe interface designs on online dating websites may exacerbate racial biases, while the Reversed Sequence Interface design effectively reduces implicit bias. The political and social consequences of design decisions in social environments warrant deeper discussion and improvement. This research provides new insights for designing fairer online dating platforms and highlights directions for future studies.