AI news
AI model reviews
AI tools
How-to guides
Our staffAboutContact

911 centers are adding AI to help dispatchers, not replace them

Dispatch centers from O'Fallon to Fairfax are using AI for transcription, translation and non-emergency calls. The people running them say a human still answers emergencies.

man in black and white checkered dress shirt using computer
Photo: CDC on Unsplash

Police dispatchers in O'Fallon are now using AI software to help them handle 911 calls, and officials there say a person will always answer. The story, from InvestigateTV and carried by local stations on November 17, also points to similar tools in Fairfax, Virginia, and New Orleans.

It matters because 911 centers are short of staff and overloaded. Going by USA Today's reporting, operators nationwide answer about 240 million calls a year, and non-emergency calls significantly outnumber real emergencies.

What the tool does in O'Fallon

The system suggests questions for the dispatcher to ask, based on the call. It also listens in, transcribes the call in real time and flags details the dispatcher might have missed.

Paul Jokerst, emergency communications administrator for O'Fallon Police, said the aim isn't to cut staff. "It's not here to replace anybody," he said.

Dispatcher Mike Conte said he supports it, mainly because it helps when a caller is hard to understand. His promise to the public was short: "You will always talk to a human."

Other centers, other uses

The same report says AI in Fairfax, Virginia, translates calls into more than 170 languages instantly. In New Orleans, AI automatically answers calls about accidents officials already know about, then lets anyone involved stay on the line to reach a dispatcher.

Sumter County, Florida, handles about 80,000 calls a year and uses AI to transcribe and summarise every incoming 911 call. Stephen Kennedy, the county's 911 coordinator, told USA Today that other dispatchers can then pass details to responders almost in real time. A Motorola system there can also flag keywords such as "cardiac arrest" or "shooting", so help can start moving while the call-taker's still gathering information.

Translation is a big draw. Kennedy said finding a human interpreter can take five or even 10 minutes, depending on the language, while AI translates as the caller speaks.

The push to hand off non-emergency calls

The most aggressive use is on non-emergency lines. Aurelian, one of the vendors, says 64% of calls reaching dispatch centers aren't emergencies and that AI can handle 70% of those without a person. That's the company's own figure, not an independent measurement.

The staffing numbers explain the interest. A 2023 study by the International Academies of Emergency Dispatch found a 25% vacancy rate in dispatch centers nationwide. Aurelian's CEO, Max Keenan, says 80% of new dispatchers quit within two years; that's his claim, and I couldn't find an independent check of it.

Aurelian says its systems now serve about 12 million Americans daily, including Hamilton County, Tennessee, and Kalamazoo, Michigan. TechCrunch has reported that the company launched its assistant in May 2024 and that it's built to transfer real emergencies to a human dispatcher. In one example Keenan gave USA Today, about 600 people called a non-emergency line during a power outage, and the system pulled out one caller whose life-sustaining medical device had stopped working.

The limits and risks

Not everyone's rushing. April Heinze, vice president for operations and standards at the National Emergency Number Association (NENA), said callers in an emergency still want a human voice. NENA is watching how the public reacts, given the publicity around AI systems that invent facts, and cost is another factor.

At the Denver 911 center, USA Today reports, workers are looking at several options but have questions about cost and reliability. I think that caution is sensible: a wrong answer on a medical call costs far more than a slow one on a parking complaint.

AI is also making life harder for dispatchers from the other side. A June survey of nearly 770 public safety communications professionals found about 14% of dispatchers had met swatting attempts involving AI, according to NPR as summarised by Newser. NENA and APCO, the two groups representing dispatchers, say they're working on training, though they haven't shared details.

What happens next

Heinze said some centers are testing AI to triage surges, such as a major interstate crash or a wildfire, when operators get hundreds of identical calls. A computer would ask whether a caller is only reporting the known incident, has information responders need, or has a different emergency.

Kennedy predicts that in 10 years "everyone is using AI to make things more efficient." I'd guess that's an optimistic forecast, since cost and reliability concerns haven't gone away.

Honestly, the human-answers-first promise is the part to watch. Transcription and translation probably carry less risk. But the more calls AI handles on its own, the more the line between non-emergency and emergency depends on software getting that call right, and I'd like to see centers publish how often it gets it wrong.

Keep reading