Engineers at Johns Hopkins University have developed an artificial intelligence tool called SafeTraffic Copilot that can predict how changes to traffic signals or road conditions affect crash rates. The system uses generative AI and large language models to analyze vast amounts of data, including descriptions of over 66,000 accidents, weather conditions, traffic patterns, roadway designs, blood alcohol levels, satellite images, and on-site photos. The AI tool helps transportation planners and policymakers evaluate how modifications—such as extending a traffic light’s cycle from 20 to 30 seconds—could influence safety outcomes. Unlike traditional machine learning models, which struggle to make predictions for scenarios unlike those in their training data, SafeTraffic Copilot offers “what-if” capabilities. This allows users to simulate changes and estimate their impact on accident likelihood. Senior author Hao “Frank” Yang, an assistant professor of civil and systems engineering at Johns Hopkins, emphasized the importance of transparency in AI decisions. To address the “black box” nature of AI, the model generates confidence scores—indicating how reliable its predictions are. This feature builds trust, especially in high-stakes areas like traffic safety. The model identifies alcohol use and aggressive driving as the top contributors to crashes, responsible for three times more accidents than other factors. This insight could guide targeted interventions and public safety campaigns. Currently, states like Maryland rely on older machine learning systems that only predict safety based on historical data similar to what they’ve seen before. SafeTraffic Copilot overcomes this limitation by understanding context and adapting to new scenarios. The tool is designed to be customizable for different regions. Yang hopes it will benefit communities in Baltimore City, Baltimore County, and across Maryland. He also aims to expand its use internationally, noting that in countries like Taiwan and the Philippines, motorcycle-related crashes are common and driving behaviors differ significantly. Previous models could not easily account for such cultural and behavioral differences, but SafeTraffic Copilot can adjust by processing descriptive inputs about local conditions. The research, published in Nature Communications, represents a significant step forward in using AI to improve road safety. With traffic fatalities in Maryland rising from 466 in 2013 to 621 in 2023, the need for smarter, more responsive safety tools is urgent. SafeTraffic Copilot offers a data-driven, adaptable solution to help reduce crashes and save lives.