When you call 911, you typically expect to hear a human voice. However, in New Orleans, an AI system may respond to certain calls, specifically those involving already reported auto accidents.
The Orleans Parish Communication District (OPCD) manages 911 services in the area and is utilizing AI to handle multiple reports regarding the same incident. This allows human operators to focus on other emergencies.
It’s a bit jarring to imagine getting an automated voice after dialing for help. So, let’s break down what exactly this system does, when it intervenes, and where the human touch still remains essential.
How New Orleans 911 AI Works
OPCD collaborated with Carbyne, an emergency communications firm, to implement AI-assisted call triage, officially launched in 2023. The AI, known as AI-V Emergency Call Triage, is designed to assist communication centers in managing sudden surges of calls related to singular events.
A single car accident can quickly generate numerous calls from nearby drivers. In busy situations, these repetitive calls can accumulate, leading to delays, while operators attend to other emergencies. This AI solution offers a way to manage those redundant calls more efficiently.
The AI checks if a call originates from within roughly 200 meters (or about 650 feet) of an already reported crash. It can only engage when all human operators are busy, and OPCD clarifies that this technology hasn’t been applied to any other kinds of 911 calls.
What Happens When the AI Answers
When the AI picks up, it identifies itself as New Orleans 911 and asks if the caller is reporting an accident at a specific location. How you respond dictates the next steps.
If you answer anything other than “yes,” your call is transferred back to the human queue. This safeguard works in any language used. If you confirm that you are reporting the known accident, the AI informs you that 911 is already aware of the situation.
Callers with more information about the accident can remain on the line for a human operator, while others may hang up after being informed they can do so. Interestingly, the AI cannot terminate the call; the decision is left with the caller.
What the AI Cannot Do
It is crucial to note the boundaries of this AI system, especially since it operates in the high-stakes environment of emergency calls. It cannot dispatch emergency responders or assess how urgently a call should be handled.
Essentially, its role is limited to recognizing specific repeat crash reports and determining if the caller needs to stay in the human queue.
Furthermore, this system is not a broad language model like ChatGPT; it performs a much narrower function within the emergency communication ecosystem.
What if the AI Doesn’t Understand You?
This was a key concern I had when learning about the technology. If the AI doesn’t comprehend the caller, it reverts to the queue for a human operator. It also defaults to a human operator when the call falls outside its designed scenario.
The system accommodates multiple languages and is specifically trained on local 911 recordings to better interpret New Orleans accents and street names. This regional training is especially helpful in a city with uniquely named locations.
What Happens if You Hang Up While Waiting?
It’s good to know that hanging up doesn’t erase your spot in line if you’re waiting for a human operator. OPCD maintains standard callback procedures, so the center can still reach out after handling the queue. They advise callers to remain on the line as long as possible, since waiting calls will be answered in due time.
Calls are Still Recorded
Interactions involving the AI are still logged and recorded as they would be in a standard 911 call. These records, including audio and metadata, remain available for investigators when needed.
Notably, the calls processed by this AI system are not shared with any external large language model, addressing common privacy concerns that arise in such scenarios.
Have There Been Mistakes with the New Orleans 911 AI?
OPCD conducted several months of testing before integrating the AI and meticulously reviewed every call handled in the first three months of its deployment. During this evaluation, they did not find any false positives or negatives.
The system managed calls as expected, without mishandling calls outside its intended reach. While this is promising, it doesn’t mean mistakes won’t happen in the future. Ongoing oversight is essential.
Has the System Been Beneficial?
Carbyne provided performance data from its collaboration with OPCD, noting the AI system triaged over 3,500 incidents in a 90-day period. They report a more than 30% drop in redundant calls and up to 40 seconds faster responses for distressed callers on average. While these numbers seem promising, I hope to see consistent updates from OPCD regarding their own data over time.
Is AI Replacing New Orleans 911 Call-Takers?
OPCD asserts that it is not replacing call-takers with AI. They’re currently facing challenges in filling all available positions, a problem seen across the broader 911 sector. OPCD views AI as a supportive tool for human operators rather than a replacement.
For New Orleans, the aim is to alleviate the burden of repeat reports during busy times, thus allowing human operators to assist other citizens in need. The agency has not expanded AI use to other types of 911 calls.
Future Expansion of 911 AI
OPCD is examining potential uses for AI in other call scenarios, but these tests are being conducted in a controlled environment rather than on the live system. They are cautious, ensuring any extension to 911 categories happens only after confirming the technology’s effectiveness. For now, the AI’s live function remains specific to the crash report scenario.
What This Means for You
The developments in New Orleans may serve as a case study, but they could have implications for other communities as well. The AI’s role raises many questions: when should AI take charge? What types of calls should it manage? How can you be sure to reach a human if needed?
Currently, New Orleans has kept the AI’s functionality very limited, dealing only with specific repeat calls about known crashes. Human operators are still available for other situations. Moving forward, it will be interesting to see how AI is integrated into emergency services across different regions.
The ongoing issue of staffing shortages might lead many emergency centers to consider how AI could help alleviate pressure. It’s essential for communities to understand what any AI system can and cannot do, as well as the ease of reaching a human representative. That’s certainly something to keep an eye on.
Final Thoughts
Experiencing an AI response during a 911 call might cause unease among many. I totally understand the concern. In crises, immediate human support is crucial. But with New Orleans permitting its AI a narrow function, the system could help ensure that human operators can address more urgent calls. I appreciate that the early audits showed no errors thus far, but I think the scrutiny needs to persist. Any broader implementation should come equipped with solid safeguards and performance records.
How do you feel about the possibility of an AI responder when calling 911? Would it comfort you knowing it could potentially speed up help for others, or would you prefer always to have a live person answer first? Feel free to share your thoughts.
