The Orleans Parish Communication District (OPCD) in New Orleans has implemented an artificial intelligence system to assist with 911 emergency communications, specifically for repeat calls about auto crashes that have already been reported. This AI-driven triage system, developed in partnership with emergency communications company Carbyne, is designed to help manage call volume and keep human operators available for other emergencies.
The AI-V Emergency Call Triage system, launched in 2023, is activated only when all human call-takers are occupied. It identifies if an incoming call is within approximately 650 feet of a known auto crash. If the caller confirms they are reporting the known crash, the AI informs them that emergency services are already aware of the incident. Callers with additional information can opt to remain on the line for a human operator, while others are informed they may hang up. The AI itself does not end calls, leaving that decision to the caller.
OPCD emphasizes that the AI has strict limitations. It cannot dispatch emergency responders or determine call urgency. Its function is solely to identify specific types of repeat crash reports and assess if a caller needs to be placed back into the queue for a human operator. The system is not a large language model like ChatGPT and performs a much narrower, specialized task within the emergency network. It has been trained using local 911 recordings to better understand New Orleans accents and street names, and it supports multiple languages.
Safeguards are in place to ensure callers are not lost in the system. If the AI cannot understand a caller, or if the call falls outside the designated crash scenario, the call is returned to the queue for a human operator. Even if a caller hangs up while waiting in the queue, the OPCD's normal callback procedures remain active. All AI interactions are logged with metadata and audio recordings for record-keeping and potential investigations. Data from these calls are not uploaded to external large language models.
OPCD reported that during an initial review of 100% of calls handled by the system in its first three months, no false positives or negatives were found. The agency continues to audit the technology closely. Vendor-reported data from Carbyne suggests that during a 90-day period, the system triaged over 3,500 events, resulting in a more than 30% drop in redundant calls and an average faster answer time of up to 40 seconds for distressed callers.
OPCD stated that the AI is intended to support, not replace, human call-takers, addressing persistent staffing shortages in the 911 industry. The current live application of the AI remains focused exclusively on the specific auto-crash scenario, with no expansion to other call categories. Exploratory testing for other potential AI uses is being conducted in a separate testing environment and will not be deployed live without rigorous validation and clear performance data.