Abstract
This paper presents Cooperative Adaptive Ride-Sharing (CARS), a sociotechnical system designed to address the inefficiencies and inequities of microtransit in underserved communities. Unlike commercial ride-sharing platforms that rely on surge pricing, CARS utilizes empathy-building AI agents to foster prosocial behavior. By nudging flexible users to adjust their schedules or walk to pick-up points, the system prioritizes seats for riders with critical, inflexible needs, such as medical appointments or employment shifts.
The Microtransit Paradox
In many rural and small-town American communities, traditional fixed-route bus systems are often inefficient and underfunded. On-demand "microtransit" services have emerged as a promising alternative. However, these services face a significant operational paradox: while vehicles are frequently underutilized, averaging fewer than four trips per hour, approximately 25% of ride requests are denied during peak demand periods.
Standard market-based solutions, such as Uber’s surge pricing, are ethically unsuitable for public transit serving low-income, carless, or disabled populations. Current AI transit research often treats riders as passive variables. The challenge addressed here is to improve system efficiency and equity without imposing monetary penalties that would further marginalize vulnerable populations.
Methodology: The CARS Framework
The authors frame microtransit as a Sociotechnical System (STS), where AI agents and human participants collaborate toward collective efficiency. The CARS framework integrates four primary components:
- Fleet Scheduling & Optimization: Utilizing high-capacity ridesharing algorithms to solve real-time routing and assignment problems.
- Empathy-Building AI Agents: Rather than a binary "No Rides Available" response, the system employs persuasive messaging. These messages leverage psychological principles like reciprocity and social proof to encourage flexible users to make small adjustments—such as shifting a trip by 15 minutes—to accommodate a neighbor's critical needs.
- Model-Based Reinforcement Learning (RL): The AI agent learns individual user preferences, mobility constraints, and values. This ensures that suggestions are context-aware, avoiding unsafe or overly burdensome requests (e.g., suggesting a long walk in inclement weather).
- Non-Monetary Incentives: A "Karma Points" system provides social rewards, recognizing cooperative behavior without the need for cash incentives.
Data and Implementation
The project is centered in Wilson, North Carolina, a city that replaced its entire bus network with a microtransit service (RIDE) in 2020. The research draws on:
- Trip Microdata: Analysis of approximately 18,000 monthly trip requests from the RIDE service.
- Qualitative Insights: Focus groups with 32 participants, primarily female and African American, revealed user struggles and a high baseline willingness to be flexible for the benefit of the community.
- Simulation Testbed: A custom environment using historical data to model the impact of prosocial interventions on system-wide wait times and denial rates.
Key Insights and Preliminary Results
Preliminary findings suggest that community empathy is a significant, untapped resource for transit optimization. Focus group participants expressed a strong desire to help others, noting that "kindness does not cost anything."
The research demonstrates that small behavioral shifts—such as a 15-minute time shift or a 400-meter walk—can drastically improve vehicle pooling efficiency. In Wilson, where 86% of riders are carless and 57% earn less than $25,000 annually, these efficiency gains directly translate to improved social equity. Initial surveys indicate that users strongly favor features like a "commuter program" for pre-booking essential work and school trips.
Ethical Considerations and Future Work
While promising, the CARS system introduces several challenges:
- Privacy: Prioritizing rides based on trip purpose (e.g., "medical") requires data disclosure. The authors propose allowing users to choose their level of disclosure to mitigate privacy risks.
- Manipulation vs. Persuasion: There is a critical ethical boundary between encouraging prosocial behavior and exerting undue pressure on vulnerable users.
- Honesty: The potential for users to misrepresent trip purposes to gain priority is a concern. The authors suggest that "social mechanisms," such as minimal information sharing among riders, could foster accountability.
- Scalability: Future research will investigate whether empathy-based nudges, successful in close-knit communities like Wilson, can be effectively scaled to larger, more anonymous urban environments.
Summary updated