Intelligent risk management: natural language processing real-time triage of police calls for service

Can the intelligent call center concept improve a “public safety answering points” (“PSAPs”) ability in identifying opportunities for a safe and effective diversified response (e.g., differential police response, co-response and Alternate First Responders)? This article examines the development of a proof-of-concept intelligent call center for enhanced 911 call processing, at the City of Seattle (Washington, USA). A “naïve, multi-label classifier” was trained using historic 911 call audio, to recognize the four response tiers proposed by Risk Managed Demand. This study employed common commercial technology used to 1) transcribe incoming 911 call audio, 2) render a real-time forecast of call risk and 3) visualize the results for personnel handling the call as “intelligent decision support.” This project proves a “human-in-the-loop” application of Machine Learning (ML) can support the professional judgement of experienced human operators with a precise, low-latency forecast of call risk. Further, the demonstrated system is designed to learn. As a diversified response system evolves, statistical feedback is incorporated using the Risk Managed Demand framework. Implications for risk management, the opportunity for diversified response, and the ethics of ML are discussed.