Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Sim911: Towards Effective and Equitable 9-1-1 Dispatcher Training with an LLM-Enabled Simulation

Zirong Chen, Elizabeth Chason, Noah Mladenovski, Erin Wilson, Kristin Mullen, Stephen Martini, Meiyi Ma

TL;DR

Sim911 addresses the labor-intensive and inequitable nature of traditional 9-1-1 dispatcher training by introducing an LLM-powered simulation platform that grounds conversations in local context. The system integrates knowledge construction, context-aware controlled generation, and looped validation to produce realistic, diverse, and equitable call simulations, including 57 incident types and 14 caller profiles. Real-world deployment with DEC demonstrates strong realism and authenticity, with substantial time savings (26.55 hours) and high trainee endorsement (~90%), while component ablations confirm the necessity of KC, CaCG, and VLC for maintaining quality. This work offers a scalable framework for augmenting training in centers with limited staffing and lays groundwork for extending AI-driven dialogue simulations to other high-stakes domains.

Abstract

Emergency response services are vital for enhancing public safety by safeguarding the environment, property, and human lives. As frontline members of these services, 9-1-1 dispatchers have a direct impact on response times and the overall effectiveness of emergency operations. However, traditional dispatcher training methods, which rely on role-playing by experienced personnel, are labor-intensive, time-consuming, and often neglect the specific needs of underserved communities. To address these challenges, we introduce Sim911, the first training simulation for 9-1-1 dispatchers powered by Large Language Models (LLMs). Sim911 enhances training through three key technical innovations: (1) knowledge construction, which utilizes archived 9-1-1 call data to generate simulations that closely mirror real-world scenarios; (2) context-aware controlled generation, which employs dynamic prompts and vector bases to ensure that LLM behavior aligns with training objectives; and (3) validation with looped correction, which filters out low-quality responses and refines the system performance.

Sim911: Towards Effective and Equitable 9-1-1 Dispatcher Training with an LLM-Enabled Simulation

TL;DR

Sim911 addresses the labor-intensive and inequitable nature of traditional 9-1-1 dispatcher training by introducing an LLM-powered simulation platform that grounds conversations in local context. The system integrates knowledge construction, context-aware controlled generation, and looped validation to produce realistic, diverse, and equitable call simulations, including 57 incident types and 14 caller profiles. Real-world deployment with DEC demonstrates strong realism and authenticity, with substantial time savings (26.55 hours) and high trainee endorsement (~90%), while component ablations confirm the necessity of KC, CaCG, and VLC for maintaining quality. This work offers a scalable framework for augmenting training in centers with limited staffing and lays groundwork for extending AI-driven dialogue simulations to other high-stakes domains.

Abstract

Emergency response services are vital for enhancing public safety by safeguarding the environment, property, and human lives. As frontline members of these services, 9-1-1 dispatchers have a direct impact on response times and the overall effectiveness of emergency operations. However, traditional dispatcher training methods, which rely on role-playing by experienced personnel, are labor-intensive, time-consuming, and often neglect the specific needs of underserved communities. To address these challenges, we introduce Sim911, the first training simulation for 9-1-1 dispatchers powered by Large Language Models (LLMs). Sim911 enhances training through three key technical innovations: (1) knowledge construction, which utilizes archived 9-1-1 call data to generate simulations that closely mirror real-world scenarios; (2) context-aware controlled generation, which employs dynamic prompts and vector bases to ensure that LLM behavior aligns with training objectives; and (3) validation with looped correction, which filters out low-quality responses and refines the system performance.

Paper Structure

This paper contains 33 sections, 19 equations, 5 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Motivating Study
  3. Methodology
  4. Dynamic Knowledge Construction
  5. Detailing Two Components in 9-1-1 Calls
  6. Specializing Knowledge Bases for Contextual Control
  7. Context-aware Controlled Generation
  8. Validating with Looped Correction
  9. Evaluation of Sim911
  10. Effectiveness of Sim911
  11. Equity of Sim911
  12. Insights from User Study
  13. Related Work
  14. Summary
  15. Discussion and Future Work
  16. ...and 18 more sections

Figures (5)

  • Figure 1: Year-round Distribution of 9-1-1 Calls and Vulnerability Index in Nashville, 2023.
  • Figure 2: An Overview of Sim911's Key Components: knowledge construction (KC), context-aware controlled generation (CaCG), and validation with looped correction (VLC). KC integrates real-world data into knowledge bases before runtime. CaCG tailors LLM behaviors during runtime. VLC involves in-context validation during runtime and periodic correction after runtime.
  • Figure 3: An Example of the 3-step Context-aware Controlled Generation with Vector Base Incorporation, Detailed Task Explanation, and Caller Image Deciphering. The incident type specification (IS) comes with tags crash report (incident type), medical emergency(special request), and severe weather (scenario contexts), and the caller image (CI) is set to be an unhousednon-native speakeradult.
  • Figure 4: Examples on Knowledge Construction process -- introducing tags to existing calls.
  • Figure 5: Conventional Training Pieces.