The first thing you notice about a real 911 call is the chaos. Someone's screaming. There's a dog barking. You can't understand half of what's being said because the caller is gasping for air, or whispering so the intruder doesn't hear. That's the job. That's what a human dispatcher handles every single day.
But in New Orleans, a new player is listening in: an AI system from a company called Carbyne. The city's testing it as a triage tool — a way to sift through the incoming flood of calls and decide which ones need a human right now, and which can wait. The pitch is seductive: faster response times, fewer missed emergencies, less burnout for dispatchers who are drowning in call volume.
Here's the thing, though. When a life's on the line, do you trust a machine to hear the panic in a voice? Because panic isn't a data point. It's a sound. And that's where this test gets messy.
The Details of the Test
Carbyne's software is running in a pilot phase in New Orleans, handling a slice of the city's 911 traffic. It listens to calls, transcribes them, and uses an algorithm to flag potential emergencies — heart attacks, shootings, fires — while simultaneously identifying non-critical calls that could be routed to a lower priority queue or handled through alternative channels.
The company claims the AI can cut response times by up to 30% in some cases, by reducing the time a dispatcher spends typing and coding a call. Instead of a human manually entering the nature of the emergency, the AI does it in seconds. Dispatchers then review the AI's assessment before dispatching units.
That last part is crucial. The system is not meant to replace humans — not yet, anyway. It's a triage assistant, a second pair of ears. But the test is being watched closely by other cities, because if it works, this could become the new normal for emergency response across the country.
What Dispatchers Are Saying
I talked to three dispatchers who've used the system in New Orleans. All of them asked to remain anonymous because they weren't authorized to speak publicly. Their verdict? Not impressed.
"It misses context," one said. "I had a caller who was whispering because someone was in the house. The AI flagged it as a low-priority call because the caller wasn't shouting or crying. If I hadn't been listening, we might've sent a patrol car instead of a full unit."
Another told me about a call where the caller was having a heart attack but spoke calmly — a symptom of the condition that makes people seem composed when they're dying inside. The AI categorized it as "non-urgent." The dispatcher overrode it. But what happens when the override doesn't happen?
"We're not robots," the third said flatly. "We know the neighborhoods. We know the regulars. We know when a caller says 'everything's fine' but it's not. That's not in the training data."
The Tech's Track Record
Carbyne isn't some fly-by-night startup. They've been around for over a decade, and their platform is used in many cities for real-time call tracking, video, and data sharing. They claim their AI is built on millions of calls and continually improves through machine learning. But for a system meant to handle life-or-death decisions, the margin for error is zero.
The company's PR materials say the AI is "designed to augment, not replace" human decision-making. That's good to hear. But the reality is, when budgets are tight, and city officials see a way to reduce staffing costs, augmentation has a way of becoming replacement.
And let's talk about bias. There's a well-documented history of AI systems failing when it comes to race and socio-economic factors. If the training data is skewed, the AI will be skewed. A call from a neighborhood with a high crime rate might get flagged as high-priority just because of the zip code, while a domestic disturbance in a wealthy area might get downgraded because the caller sounds calm and articulate. That's not speculation — that's the pattern we've seen in predictive policing and other AI-driven public safety tools.
The Political Angle
New Orleans is a city that's been through hell and back. Hurricane Katrina, corrupt leadership, and now a surge in violent crime that's left police departments short-staffed and overwhelmed. The mayor's office is touting this AI test as a technological leap forward — a way to do more with less, to use innovation to address challenges that have stumped human administrators for years.
But you have to ask: who's pushing this? Carbyne's parent company has lobbied for government contracts aggressively, and the sales pitch is always the same — efficiency, cost savings, accuracy. Cities are buying it because they're desperate. And desperate times lead to desperate measures.
A city council member I spoke with, who voted to approve the pilot, said, "We can't hire enough dispatchers. The turnover is insane. If this tool can take some of the load off, we owe it to the public to try." Fair point. But when you're testing with real 911 calls, the stakes aren't a spreadsheet — they're a body on the floor.
The Human Element
Here's what the algorithm can't measure: emotion. The catch in a voice when someone says "Please hurry." The sound of a door creaking open in the background. The silence that means the caller is in danger and can't speak.
One dispatcher told me about a call where a mother was reporting her son missing. She was calm, measured, almost clinical. The AI flagged it as low urgency. But the dispatcher heard the controlled panic in her voice. She probed a little more and found out the son had a history of suicidal ideation. That call went from low to critical in seconds. The AI didn't see it coming.
"We're not robots. We know the neighborhoods. We know when a caller says 'everything's fine' but it's not. That's not in the training data." — anonymous NOLA dispatcher
Technology can do a lot of things. It can process data faster than any human, it can find patterns in a million calls, and it can route resources more efficiently. But it can't be a good neighbor. It can't know that old Mrs. Johnson always calls about her cat when she's lonely, and that today she's not calling about the cat — something's different.
The Verdict
I'm not saying AI has no place in emergency response. It does. It can help with data entry, with transcription, with routing. It can cut down on hold times and flag repeat callers. There's real potential here.
But triage? The moment you let a machine decide who gets help first, you're making a values judgment. You're saying that a calm voice is less urgent than a screaming one. And that's a dangerous assumption.
The test in New Orleans is being watched. If it succeeds, other cities will follow, and before long, we'll have a system where the first responder isn't a person but a server farm in some data center. And when that happens, we better hope the algorithm got it right.
Because when the algorithm gets it wrong, the price is a life. And no amount of efficiency can buy that back.
