NSF Award
2427711
9-1-1 dispatchers are frontline members of emergency services who receive calls, advise callers, and direct emergency units to specific locations. Dispatchers' performance directly impacts response times, the efficiency and effectiveness of emergency services, and, ultimately, lives. Nationwide, the training of public safety dispatchers is a continual effort. Currently, conducting these training exercises is laborious and time-consuming. It involves multiple experienced employees acting as role-players to simulate service calls, allowing trainees and trainers to practice call-processing skills and assess performance. Additionally, the quality of training can vary based on available resources and the trainers' experience. Many dispatching centers, particularly in areas with limited resources, struggle with heavy workloads and insufficient training. To address this need, this project aims to transform the 9-1-1 dispatcher training by developing a Generative Artificial Intelligence (GenAI) enabled training system with three essential functions: playing the role of callers and simulating 9-1-1 calls from a broad and diverse spectrum of emergency scenarios; effectively evaluating the dispatcher’s performance in the simulation; and providing strategies for training enhancement. The proposed system can provide dispatchers with more high-quality training with fewer resources. It can help emergency communications centers with limited staffing by allowing trainees to interact individually with the training program. Nearly 6,000 emergency communications centers could benefit from this training opportunity. Moreover, the GenAI-enabled solution can be extended to other training spaces, such as teachers and medical students. This work can also provide guidance for governments on developing policies for utilizing AI/GenAI solutions.<br/><br/>The project's solution is to adopt a responsible design approach to construct a training system to develop a realistic simulation-based training environment. It will adopt an iterative design approach to construct a training agent and debriefing dashboard, using a repository of past emergency calls and the latest Large Language Model (LLM) and Generative AI technologies. A responsible design approach will adopt four important principles: (1) develop caller images and call types from a repository of emergency phone call recordings; (2) adopt a user-centered design approach and a co-design methodology with dispatchers and instructors in the design of the training agent; (3) a combination of controlled prompt generation and chain of thought reasoning to elicit LLM conversational segments that are customized to the caller and call types and a validation phase that evaluates the LLM-generated content against benchmarks and actual data. and (4) evaluation metrics and test cases to ensure inclusion, responsibility, and trust.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
