This post is also available in:
עברית (Hebrew)
Researchers at Johns Hopkins University have developed a new artificial intelligence system designed to better understand and predict traffic accidents across the United States. The tool, named SafeTraffic Copilot, combines multiple forms of data to identify risk patterns behind crashes and to support preventative planning by transportation authorities.
Unlike traditional crash analysis models that rely on broad statistical data, this system uses large language models (LLMs) to process and analyze complex, real-world information. These include written descriptions of incidents, numerical data like blood alcohol levels, as well as satellite imagery and on-site photographs, according to TechXplore.
What sets SafeTraffic Copilot apart is its ability to evaluate both single and combined risk factors—such as poor weather conditions, traffic volume, or road design—to determine how they interact in causing a crash. The model also quantifies its confidence in predictions, giving users an estimate of how likely a forecast is to match real-world outcomes.
The system is designed to improve over time. As more data is fed into it, the model’s predictive performance becomes increasingly accurate, enabling continuous learning and refinement. This positions the tool as a dynamic support system for traffic planners and policymakers, rather than a one-time assessment method. It offers detailed insights into the underlying causes of incidents and helps identify where targeted interventions—such as road redesign or policy changes—might reduce risks.
One of the key objectives of the project is to ensure that AI systems like this can operate transparently and responsibly in sensitive, high-stakes environments such as public safety. The researchers emphasize that the tool is meant to assist, not replace, human judgment.
The project appears in Nature Communications and continues to explore how AI and human expertise can be integrated in complex decision-making scenarios involving health and safety.
