IF-City: Intelligible Fair City Planning to Measure, Explain and Mitigate Inequality

Area of Science:

• Artificial Intelligence
• Visual Analytics
• Algorithmic Fairness

Background:

• Existing AI fairness tools primarily target data scientists, neglecting domain experts.
• Fairness in allocation and planning is under-researched compared to predictive decisions.
• Domain-specific tools are crucial for inclusive AI fairness in complex planning scenarios.

Purpose of the Study:

• To propose an inclusive framework and visual tool for domain experts to address algorithmic fairness in allocation.
• To enable assessment and mitigation of unfairness in allocation problems through explainable AI.
• To support fair urban planning for equitable access to city amenities and benefits.

Main Methods:

• Developed the Intelligible Fair Allocation (IF-Alloc) Framework using causal attribution, contrastive, and counterfactual reasoning.
• Created the Intelligible Fair City Planner (IF-City), an interactive visual tool for urban planners.
• Applied IF-City to a New York City neighborhood case study with international urban planners.

Main Results:

• IF-City enables users to perceive inequality, identify its sources, and simulate mitigation strategies.
• The framework and tool facilitate iterative design and constraint satisfaction for fair allocation.
• Practicing urban planners found the tool useful for assessing and addressing fairness in urban planning.

Conclusions:

• The IF-Alloc framework and IF-City tool effectively support domain experts in tackling algorithmic fairness in allocation.
• Domain-specific visualizations and explainable AI are vital for inclusive AI fairness in planning.
• The framework shows potential for generalization to diverse fair allocation applications beyond urban planning.