Anoka County, Minnesota, is testing an artificial intelligence system to screen non-emergency calls to its 911 call center. The goal is simple: let human dispatchers focus on life-threatening emergencies while a machine handles the barking dogs, fireworks complaints, and noise issues that flood the lines every day.

The county's call center receives about 1,100 calls per day. According to dispatcher Samantha Gust, who handles up to 150 calls during a 12-hour shift, roughly two-thirds of those calls are not emergencies. That means about 733 calls each day tie up dispatchers who could otherwise be handling heart attacks, car crashes, or active threats.

The AI program, costing $60,000, is currently in a testing phase. The county expects it to go live by mid-May. When a caller dials the non-emergency number, the AI answers, asks for the nature of the call, and either handles the request or transfers a transcript to a human dispatcher if the situation escalates.

"We can screen the non-emergency calls so we can answer those 911 calls quicker and dedicate more time to those people without having to put someone on hold," a county official told local station KSTP.

During a test of the system, a reporter called the non-emergency line and described what they believed was a heart attack. The AI recognized the urgency and responded, "I'm transferring you to an agent." In a real deployment, that transfer would go to a human call taker. The system is designed to never handle an actual 911 call — anyone dialing 911 will reach a person directly.

How the AI triage works

The system operates on a simple triage model. It listens to the caller's description, parses key words and phrases, and routes the call accordingly. For routine non-emergency issues — noise complaints, lost property, minor traffic incidents — the AI can log details, provide standard instructions, and close the call without involving a human dispatcher. For calls that involve medical distress, violence, or any indication of an emergency, the AI flags the conversation and passes a real-time transcript to a human operator.

This is not voice recognition alone. The system analyzes the content of the speech, not just keywords. It can detect urgency in tone and language. The county says the AI is trained on thousands of hours of actual non-emergency calls, though officials declined to specify the training data or the vendor supplying the software.

The appeal for dispatchers is both operational and mental. "It's good for us mentally as well, to be able to recover from some of the higher priority calls that we're taking, and then focus on those after to make sure we have all the correct information in there," Gust said. Handling a steady stream of non-emergency calls between actual emergencies wears on dispatchers. The AI gives them breathing room.

Broader adoption of AI in public safety

Anoka County is not alone in pushing AI into public safety operations. The same KSTP investigation found that police departments in Brooklyn Park, Eagan, and Bloomington are using artificial intelligence to write police reports. Officers dictate incident notes, and the AI generates a formatted report. The departments require officers to review every AI-written report before filing it and to disclose when the technology was used.

The argument in favor is the same as in Anoka County: time. "If this helps us save on personnel costs, if this helps us save time and allows our police officers to be out on the street more, which I absolutely believe it will, I've seen that it does, then no question," a police official told KSTP.

But the spread of AI in law enforcement and emergency services raises concerns that critics say are not being addressed fast enough.

The ACLU's warning: bias and errors

The American Civil Liberties Union has warned that using AI to write police reports could introduce errors and bias into official documents. Police reports are foundational to criminal investigations, prosecutions, and public records. If an AI misinterprets an officer's dictation or inserts language that reflects training data biases, the consequences could be serious.

Anoka County's call-screening system faces similar risks. An AI that misclassifies a non-emergency call as an emergency could unnecessarily dispatch resources. A more dangerous failure would be the reverse: an AI that hears a genuine emergency but treats it as routine, delaying a human response.

The county says the system is designed to err on the side of caution. "If there's any doubt, it escalates to a human," an official said. But during the testing phase, the reporter's simulated heart attack call was treated correctly. The system identified the urgency and transferred it. Whether that reliability holds at scale, with varied accents, speech impediments, panic, and attempts to deceive, remains to be seen.

What the AI cannot handle

Non-emergency calls are not just simple. They can be ambiguous. A caller reporting a suspect vehicle might sound calm but the situation could be connected to a nearby armed robbery. AI cannot read context that a human dispatcher would pick up from neighboring calls, radio chatter, or institutional knowledge.

That limitation is why Anoka County is limiting the AI to the non-emergency line. Anyone who dials 911 will bypass the system entirely. The county also says human dispatchers will monitor the AI's performance after launch and can override its decisions.

Cost and benefit analysis

At $60,000, the AI system is relatively cheap compared to the cost of hiring additional dispatchers. A single dispatcher's annual salary and benefits in the Minneapolis metro area can exceed $70,000. If the AI handles 733 calls per day that would otherwise go to humans, the savings in labor and reduced burnout could be substantial.

But the cost of a wrong classification is measured in human lives and legal liability. If the AI misses a stroke or a child drowning because a caller described it as "someone acting strange" or "a small problem," the county could face lawsuits and public outrage.

What other municipalities are doing

Several other 911 call centers across the country are experimenting with similar AI triage systems. Dallas, Texas, launched an AI chatbot for non-emergency reporting in 2023. Los Angeles County uses an AI system to help dispatchers prioritize calls based on severity. The technology is not new, but it is maturing rapidly as natural language processing improves.

What sets Anoka County's test apart is the explicit separation between the AI's domain (non-emergency) and the human's domain (911). Other systems blur that line by having AI assist on all calls. Anoka County's approach is more conservative but also more defensible in the court of public opinion.

The human element

Dispatchers like Gust are not worried about being replaced. "I'll get a human call taker," she said. The AI is a tool, not a replacement. The county has made clear that no dispatcher jobs will be cut. Instead, the goal is to reduce stress and burnout in a profession that already faces high turnover.

"It's good for us mentally as well, to be able to recover from some of the higher priority calls," Gust said. The AI creates a buffer, a few seconds to breathe between the traumatic calls that stick with dispatchers for years.

What comes next

Anoka County will continue testing the AI through mid-May. If the rollout goes smoothly, other counties in Minnesota could follow. The state's public safety department is watching the test for potential statewide adoption. The ACLU and other civil liberties groups will be watching for errors.

The broader question is whether AI can earn public trust in a domain where mistakes are not just annoying — they can be fatal. Anoka County is betting that a carefully limited deployment, with human oversight, can prove the technology while keeping the public safe.

For now, if you dial the non-emergency line in Anoka County and hear an AI voice, you can be confident that a real person is still just one "emergency" keyword away.