Back to AI Research

AI Research

teLLMe Why (Ain't Nothing but a Jam): Explorato... | AI Research

Key Takeaways

  • teLLMe Why (Ain't Nothing but a Jam): Exploratory Causal Analysis of Urban Driving Data Traffic agencies increasingly rely on large volumes of video data fro...
  • Traffic agencies now have access to large volumes of video-derived data for studying safety and congestion.
  • Most of these data are observational and collected without interventions, which makes causal questions such as "How would rain change traffic density?" difficult to answer.
  • We present teLLMe, a system for exploratory causal analysis of urban driving datasets.
  • Natural-language questions are mapped to structured causal queries through a schema-aware LLM, enabling users to specify treatments, outcomes, and subpopulations.
Paper AbstractExpand

Traffic agencies now have access to large volumes of video-derived data for studying safety and congestion. Most of these data are observational and collected without interventions, which makes causal questions such as "How would rain change traffic density?" difficult to answer. We present teLLMe, a system for exploratory causal analysis of urban driving datasets. The system starts from a structured event table built from dashcam annotations and combines causal structure learning with the PC algorithm, bootstrap-based stability checks, and query-specific effect estimation using linear regression and DoWhy. Natural-language questions are mapped to structured causal queries through a schema-aware LLM, enabling users to specify treatments, outcomes, and subpopulations. teLLMe returns a "Causal Card" that summarizes effect estimates, adjustment sets, DAG support, and assumptions, followed by a short natural-language explanation. Case studies on BDD-derived traffic events show that the system can surface plausible relationships involving weather, peak hours, and traffic density, while making uncertainty and modeling choices explicit. The system is designed as a tool for hypothesis generation and expert reasoning rather than a source of definitive causal claims.

teLLMe Why (Ain't Nothing but a Jam): Exploratory Causal Analysis of Urban Driving Data
Traffic agencies increasingly rely on large volumes of video data from dashcams and CCTV to study urban safety and congestion. However, because this data is observational—meaning it is collected without controlled experiments—it is difficult to determine true cause-and-effect relationships, such as how specific weather conditions directly impact traffic density. The teLLMe system addresses this by providing a structured, query-driven workflow that allows researchers to perform exploratory causal analysis on urban driving datasets, helping them generate and inspect plausible hypotheses while keeping modeling assumptions transparent.

How the System Works

The teLLMe workflow is divided into two main stages. In the offline discovery phase, the system processes raw dashcam annotations into a structured event table. It then uses the PC algorithm, combined with domain constraints and bootstrap resampling, to learn a causal graph (a Directed Acyclic Graph, or DAG) that represents relationships between variables like weather, time of day, and traffic density.
In the online query phase, users ask questions in natural language. A schema-aware Large Language Model (LLM) translates these questions into formal causal queries. The system then uses the learned causal graph to select the appropriate variables to adjust for, ensuring that the resulting effect estimates are grounded in the data.

Generating Causal Cards

When a user submits a query, teLLMe returns a "Causal Card." This card serves as a summary of the analysis, providing the estimated effect, the adjustment set used to calculate it, evidence from the causal graph, and key assumptions. The system also provides a short, plain-language explanation of the results. By using both linear regression and the DoWhy library, the system offers a consistent way to estimate the average treatment effect while making the underlying modeling choices visible to the user.

Insights and Limitations

Case studies using the BDD100K dataset demonstrate that teLLMe can successfully surface relationships, such as the impact of rain on traffic density at intersections or the effect of peak hours on highway congestion. However, the authors emphasize that the system is a tool for hypothesis generation rather than a source of definitive causal claims.
Several limitations remain: the system cannot account for unobservable factors like driver intent or road surface conditions, and it treats event windows as independent, ignoring broader temporal or spatial patterns. Because the system relies on observational data, the reported effects are intended to guide expert reasoning and deliberation rather than provide final, absolute answers for urban planning.

Comments (0)

No comments yet

Be the first to share your thoughts!