Lay Perceptions of Algorithmic Discrimination in the Context of Systemic Injustice

Research Background and Issues

• Issues and Challenges: The use of algorithms in high-risk decision-making scenarios can lead to unfair treatment of certain groups, such as racial and gender minorities. Although research on algorithmic fairness aims to make algorithms more "computationally fair," it often overlooks the context of unjust social structures in which algorithms are embedded. This neglect of context may result in algorithms exacerbating rather than addressing social injustices.
• Significance: The unfairness of algorithms not only directly impacts social minority groups but may also further entrench unjust social structures in decision-making processes.
• Research Motivation: Existing studies on algorithmic fairness often fail to consider the systemic injustice context in which algorithms operate, leading to recommendations that may not effectively achieve fairness goals. The authors aim to fill this research gap by exploring how informatization influences public perceptions of algorithmic gender or racial discrimination.

Solution

• Proposed Approach: The authors conducted a 2x3 experimental design (involving 716 participants) to investigate whether embedding algorithmic decisions within a context of systemic injustice affects public perceptions of algorithmic discrimination. They also explored whether explaining how algorithms learn from human decisions and perpetuate biases changes people's attitudes toward algorithms.
• Innovativeness: This study is the first to directly compare the effects of two intervention approaches (historical context and algorithmic learning of prior biases) on participants' perceptions.
• Key Techniques and Implementation Steps:
  1. Use racial discrimination in hiring scenarios as a case study.
  2. Conduct experimental manipulations: participants read content or were exposed to different contextual settings (whether or not they included historical injustice context, and whether or not they explained how algorithms learn from human decisions).
  3. Measure key variables, including perceptions of algorithmic fairness, trust, and attitudes toward improving or banning algorithms.

Research Findings

• Specific Findings:
  • Embedding algorithms in a systemic injustice context had heterogeneous effects: participants from beneficiary groups had more negative attitudes toward algorithms, while participants from disadvantaged groups evaluated algorithms more positively.
  • Explaining how algorithms learn and perpetuate biases did not significantly influence participants' perceptions.
• Strengths: The study demonstrates that linking existing algorithm research to systemic social issues can reveal the complexities behind public perceptions. It also highlights how descriptions of algorithms can, in turn, influence attitudes among different groups.
• Experimental Results:
  • Participants generally rated the distributive fairness, procedural fairness, and trustworthiness of algorithmic decisions negatively.
  • Explaining systemic injustice led to significant differences in attitudes among racial groups.
  • Beneficiary groups (e.g., White participants) exhibited more negative perceptions after reading about the historical context, while minority groups (e.g., Black participants) tended to have more positive views of the algorithms.
• Limitations and Future Directions:
  • Limitations:
    • The experimental design may have a "floor effect," where the explicitness of algorithmic discrimination suppressed more significant experimental effects.
    • The textual manipulation was limited and may not have fully prompted participants to deeply understand systemic issues.
    • The study was confined to the U.S. context and focused on racial issues; future research should expand to other countries and groups.
  • Future Directions:
    • Explore more diverse intervention formats, such as interactive tools, data visualizations, or simulation experiments.
    • Examine how different interventions influence public understanding of algorithmic mechanisms and their social impacts.
    • Introduce other contexts, such as gender or economic injustice, to observe whether interventions effectively transform public concerns.

Conclusion

This study highlights the critical influence of the social context of algorithm evaluation on public perceptions and suggests that future research should further explore effective ways to explain algorithmic unfairness and learning mechanisms to the public. Additionally, the study reminds us that the design and deployment of algorithms must not ignore the systemic social issues they are embedded in, to avoid exacerbating injustices.