Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note3 min read

teLLMe Why (Ain't Nothing but a Jam): Exploratory Causal Analysis of Urban Driving Data

Source

arXiv

arxiv.org/abs/2607.15254v1

Read the source

Source

Plain-Language Brief

Researchers have developed teLLMe, a system that helps traffic agencies analyze large volumes of video-derived data for studying safety and congestion by answering causal questions about urban driving, such as 'How does rain affect traffic density?' The system uses structured event tables from dashcam annotations, combines various algorithms, and allows users to specify treatments, outcomes, and subpopulations. It provides a summary of the results in a 'Causal Card' and explains the findings in natural language.

What the Document Says

  • The paper presents teLLMe, a system for exploratory causal analysis of urban driving datasets.
  • teLLMe starts from a structured event table built from dashcam annotations.
  • It 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.
  • teLLMe returns a 'Causal Card' that summarizes effect estimates, adjustment sets, DAG support, and assumptions.
  • The system provides 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.
  • teLLMe is designed as a tool for hypothesis generation and expert reasoning rather than a source of definitive causal claims.

Key Concepts Mentioned

  • traffic agencies
  • video-derived data
  • urban driving datasets
  • structured event table
  • dashcam annotations
  • causal structure learning
  • PC algorithm
  • bootstrap-based stability checks
  • query-specific effect estimation
  • linear regression
  • DoWhy
  • natural-language questions
  • schema-aware LLM
  • treatments
  • outcomes
  • subpopulations
  • Causal Card
  • effect estimates
  • adjustment sets
  • DAG support
  • assumptions
  • hypothesis generation
  • expert reasoning

Claims Made by the Document

  • teLLMe can help answer causal questions about urban driving.
  • teLLMe can surface plausible relationships involving weather, peak hours, and traffic density.

Evidence Used

  • The text mentions that teLLMe is designed to help study safety and congestion in urban driving data.
  • The case studies on BDD-derived traffic events show the system's ability to surface plausible relationships.

Methods / Approach

  • causal structure learning
  • PC algorithm
  • bootstrap-based stability checks
  • query-specific effect estimation using linear regression and DoWhy
  • natural-language mapping through a schema-aware LLM

Limitations Stated by the Authors

The system is designed as a tool for hypothesis generation and expert reasoning rather than a source of definitive causal claims.

Limitations Noted by the Agent

Agent interpretation — not from the source.

The data used is observational, collected without interventions. This makes it difficult to answer causal questions definitively.

Open Questions

  • What is the accuracy and reliability of teLLMe's results?
  • How does teLLMe handle missing or ambiguous data points?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The paper emphasizes that teLLMe is a tool for hypothesis generation and expert reasoning rather than making definitive causal claims.

Raw Abstract / Excerpt

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.