Stakeholder-Centered AI Design: Co-Designing Worker Tools with Gig Workers through Data Probes

Title of the Paper

Stakeholder-Centered AI Design: Co-Designing Worker Tools with Gig Workers through Data Probes

Paper Information

• Subject Area: AI design, participatory design, user-centered design, and data visualization in the gig economy
• Keywords: gig economy, AI design, co-design, data probes, worker well-being, worker-driven tools

Research Background and Problems

• What issues or challenges did the authors identify?

  • Current algorithmic management tools tend to prioritize platform interests over the well-being of gig workers.
  • AI design processes often lack attention to key stakeholders, particularly the contexts of gig workers (e.g., driving habits, earning rates, and physical and mental health).
  • There is limited research on how to leverage gig workers' actual data (e.g., driving data) in designing AI tools.

• Why is this problem important?

  • The gig economy has become a significant part of the modern economy, with services like ride-sharing heavily relying on AI algorithms for management and optimization. However, these algorithms may unintentionally result in unfair management practices or decisions based on incomplete data.
  • Focusing on workers' perspectives can improve their well-being and lead to fairer and more efficient management tools.

• Research motivation and related work

  • Related studies suggest that design probes have potential in exploring user behavior patterns and supporting AI design.
  • By co-designing tools with gig workers, the authors aim to redefine worker-centered AI tools, focusing on worker well-being and algorithmic fairness.

Solution

• What methods or solutions did the authors propose?

  • Employing data probes as a method to co-design AI tools with gig workers, exploring ways to help workers understand their work data, optimize strategies, and protect their well-being.
  • Developing five types of data probes based on individual and city-level worker data, including personal driving animations, maps, calendars, time-distribution bar charts, and a city-level work planning tool.

• What is innovative about this solution?

  • Using data probes as boundary objects, enabling workers to reflect on and describe their work patterns, as well as uncover personal contextual factors influencing their decisions.
  • Introducing worker well-being and contextual constraints as key considerations in AI design, rather than focusing solely on productivity or platform benefits.
  • Combining individual worker data with collective data to enhance worker engagement and lay the foundation for fairness and transparency in tools.

• What are the implementation steps and key technologies used?

  1. Collecting historical data at both worker and city levels, including individual driving records and Chicago taxi data.
  2. Creating five types of data probes:
     • Animation: Visualizing driving trajectories for a single day.
     • Map: Displaying workers' starting points, destinations, and earning rates across different areas.
     • Calendar: Highlighting daily earnings with color coding for easy comparison.
     • Time-distribution bar chart: Showing hourly income rate distributions.
     • Work planning tool: Predicting workers' income and expenses by adjusting parameters such as driving days, times, and areas.
  3. Conducting two-hour design sessions with 12 Chicago-based gig drivers, guiding them to explore the data probes, analyze comparisons with city average data, and attempt to optimize their work plans.
  4. Collecting data through Zoom-recorded sessions and performing qualitative analysis.

Research Findings

• What specific findings were achieved?

  1. Workers identified key factors influencing their decisions, such as the trade-offs between high-income times/areas and personal well-being, through the data probes.
  2. The study revealed the complexity of multi-platform work and the gaps in existing data (e.g., specific patterns of drivers switching between platforms).
  3. Workers used the probes to uncover and validate unfairness in platform algorithms, such as system preferences for random location and time assignments.

• What advantages does this have compared to existing solutions?

  • Provides a fully worker-centered approach to AI tool design, rather than one centered on platforms or developers.
  • The versatility of data probes not only helps workers understand their situations but also provides researchers with insights to improve tool design.
  • Introduces visualizations that lower the barrier to understanding for workers, significantly increasing their engagement.

• What were the experimental or evaluation results?

  • Data probes demonstrated their effectiveness as boundary objects, enabling workers to reflect on and communicate their patterns, as well as uncover contextual factors beyond the data.
  • For example, through the map probe, different workers quickly identified high-income but high-risk areas and explained how these risks influenced their choices.

• Limitations and future directions

  • Limitations:
    • The participant sample was limited to the Chicago area and relied heavily on drivers who voluntarily provided data, potentially leading to a sample bias toward those with higher data literacy.
    • Some key variables could not be fully modeled due to data access limitations (e.g., cross-platform data).
    • The research method was qualitative, requiring quantitative studies with larger sample sizes for validation.
  • Future directions:
    • Developing cross-platform tools to integrate drivers' data from multiple platforms and analyze their comprehensive work patterns.
    • Expanding to other cities or occupations to study whether similar tools can enhance the well-being of general gig economy workers.
    • Exploring the application of data probes in policy-making or collective worker actions, such as auditing algorithmic fairness or advocating for greater transparency.