Empathy and AI: Achieving Equitable Microtransit for Underserved Communities

Eleni Bardaka, Pascal Van Hentenryck, Crystal Chen Lee, Christopher B. Mayhorn

Abstract

Public microtransit in underserved rural and suburban communities often suffers from a "utilization paradox," where vehicles are underused yet systems are overextended during peak hours. This paper introduces the Cooperative Adaptive Ride-Sharing (CARS) framework, a sociotechnical system that leverages AI-driven empathy-building messaging instead of surge pricing to manage demand. By encouraging prosocial behaviors—such as shifting trip times or walking to pick-up points—the system aims to enhance transit efficiency while maintaining equity for low-income, elderly, and disabled riders.

Introduction

In many small U.S. towns, traditional fixed-route bus services are inadequate or non-existent, leaving carless residents—often the poor, elderly, or disabled—stranded. Microtransit, an on-demand, app-based van service, has emerged as a promising solution. However, these systems frequently face a "utilization paradox": vehicles average fewer than four trips per hour, yet the system becomes overextended during peak times, leading to long wait times and frequent cancellations.

Commercial ride-sharing services like Uber manage such demand through surge pricing. However, this approach is inequitable for public transit, as it unfairly penalizes low-income riders traveling for essential needs like work or medical appointments. The challenge lies in managing demand and increasing "ride-pooling" without using monetary incentives that exacerbate inequality.

Overview of the CARS project, highlighting research objectives and community partnerships.

The CARS Framework

The proposed CARS system integrates three core components to manage demand through a "sociotechnical" lens:

  1. Fleet Management Algorithms: Building on the Alonso-Mora et al. framework, the system uses high-capacity ridesharing algorithms to solve complex "Traveling Salesman Problems" in real-time, identifying feasible shared trips.
  2. AI-Based Decision Support: Instead of simple ride assignment, an AI agent maintains a model of user preferences and values. Using model-based Reinforcement Learning (RL), the agent predicts a user’s willingness to be flexible based on context, such as weather, time of day, or trip purpose.
  3. Choice Architecture & Empathy Messaging: The system employs "nudges" and psychologically validated messaging. For example, if the system is congested, it might ask a flexible user to move their grocery trip by 20 minutes so a neighbor in a wheelchair can make a doctor’s appointment. It also introduces "karma points"—a non-monetary social reward system to recognize cooperative behavior.
The CARS Agent architecture, illustrating the interaction between user decisions, context, and system measurements.

Community-Centered Insights

The project is centered on a pilot in Wilson, North Carolina, a community where 57% of riders earn less than $25,000 per year. Key insights from focus groups and historical data include:

  • Prosociality over Pricing: Focus groups revealed that many riders were willing to walk further or wait longer without financial reward if they knew it helped a more vulnerable community member.
  • Contextual Sensitivity: A "nudge" to walk 400 meters might be healthy on a sunny day but dangerous for a user with low vision during a rainstorm. The AI must learn these nuanced boundaries to remain truly "empathetic."
  • System Inefficiency: Data analysis showed that despite high demand, only one-third of current trips in Wilson are shared, highlighting a significant opportunity for AI-driven optimization.

Conclusion and Future Work

Preliminary results confirm strong community buy-in for a more cooperative transit model. Future work includes a four-month pilot with 1,000 participants in Wilson to test the effectiveness of empathy-building messages in real-world environment. Ethical considerations regarding privacy and the potential for manipulation remain central to the project's ongoing development, with strict user controls and community oversight planned for the pilot phase.

Summary updated

BibTeX
@inproceedings{bardaka2024empathy,
  author = {Eleni Bardaka and Pascal Van Hentenryck and Crystal Chen Lee and Christopher B. Mayhorn},
  title = {{Empathy and AI: Achieving Equitable Microtransit for Underserved Communities}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2024},
}
Figure