Research at: Harvard School of Engineering and Applied Sciences.

Role: UX Research, HCI, UX Design.

Team of 2 with a ML researcher.

Project Overview: This research explores biases in online doctor-ranking platforms, particularly against female physicians, by examining AI-generated rankings in list-based and visualization-based interfaces. Using visual attention theories, we designed the interface to display doctors as circles sized by relevance. We then conducted user studies through which we uncovered behavioral patterns and provided insights for creating more equitable and user-friendly designs.

Problem

Search Engines May Perpetuate Gender Biases

Studies show online doctor reviews are biased against female providers, leading to inequitable search results for top healthcare providers.1.

Skewed Attention Distributions Do not Match Relevance Distributions

Current list-based ranking interfaces exacerbate these biases as visual attention trends often differ from the relevance of results. The goal of fair ranking interventions proposed in past literature is to ensure that the attention on different ranked individuals is in proportion to their relevance to the search query.

Figure: Research demonstrates how minor differences in relevance can result in significant exposure disparities 7 in list-based ranked interfaces.

Research Questions

• How might we reduce gender-based biases in ranked interfaces?
  • How can we meaningfully intervene on expected user attention distributions?
  • How might we get the user to engage more with the idea of relevance?
• How does the transition from a list-based ranking interface to a data visualization-based interface impact user attention and decision-making?