Research Summary: Who Bears the Risk When Generative AI Enters Transport? A Distributional Sociotechnical Audit of Algorithmic Equity, Synthetic-Data Validity, and Public Trust

New research introduces a Distributional Sociotechnical Audit (DSA) to assess the risks of generative AI in transport, addressing gaps in current governance frameworks. The audit integrates algorithmic equity, synthetic-data validity, and public attitudes, applying persona-controlled queries to large language models and testing crash-record generators to measure distributional risks across diverse populations.

Why it matters

The introduction of generative AI into critical sectors like transport necessitates robust frameworks for risk assessment, particularly concerning equitable outcomes and data integrity. This research provides a structured approach to identify and mitigate biases and validity issues, which is crucial for maintaining public trust and ensuring the responsible deployment of AI technologies.

Key insights

• Generative AI is being deployed in transportation for traveler advisories, synthetic crash-record generation, and policy decision support.
• Current governance frameworks lack transport-specific statistical tools to evaluate distributional risks across varied populations.
• A Distributional Sociotechnical Audit (DSA) has been developed to integrate algorithmic equity, synthetic-data validity, and public attitude heterogeneity into a single empirical pipeline.
• The audit involved analyzing 5,760 persona-controlled queries across four large language model families, 12 demographic cues, and four transport topics.
• The methodology includes using two cross-family judges and a Wasserstein-2 Equity Dispersion Index to measure equity.
• It also tested three FARS crash-record generators with conditional projected maximums to assess data validity.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.11611