Salt Lake City, plans to implement AI-assisted 911 call triaging to handle ~30% of about 450K non-emergency calls per year
Technology
143
Beiträge
77
Kommentatoren
3.0k
Aufrufe
-
AI is horrible at understanding context. remember when that lady was calling the police about her abuser and coded it to sound like a pizza order? yea I can see an AI hanging up
you have a link to that? interesting
-
This post did not contain any content.
Idiocracy was definitely a documentary.
-
This post did not contain any content.
Edit: I'm pretty sure the article is just going for shock value, and a lot of the commenters are getting baited. The City isn't looking to make 911 calls go to an AI. It's people who call the non-emergency line.
I design call centers (including for PubSec) for a living. We have a service offering for a non-emergency 911 bot. It's honestly not even that new of a feature, it was around before the generative AI boom. Dispatch Centers are chronically understaffed, the job is hella stressful, there's a lot of attrition and training new employees takes a lot of time because the calls can be sensitive or complex.
There is a pretty defined split in different cities (I mostly do state & local govt, not federal) in terms of who wants AI and who despises it. Some folks that lead dispatch groups are VERY adamant that everything needs to be a person, they often have big egos because their call center is "the most important" in any city.
And yeah, we've implemented the non emergency 911 bot for customers before. Our design starts with an agent though, and if the agent makes the determination that it's not an emergency, they transfer the call over to the automated line. Btw, roughly half of all calls into a 911 center are actual "emergencies". So they get a shit ton of calls they don't need to, my guess is just because 911 is easy to remember and a non emergency line isn't, I feel like we need another 3 digit line for "not life and death but still important" calls.
-
I've worked as a first responder for a number of years, our county like many have an emergency number, 911, and a non-emergency number, i.e. 123-456-7890. We actually carry cards with the nom emergency number on it with us in the truck to pass out if a call was less than an emergency for people in our county to put into their phones for future use.
We also are a smaller place and only ever have 2-3 dispatchers on at a time, so if the calls on the non-emergency line they got could be 'auto-filled' by the AI with the location, need, and everything and wasn't tieing up a dispatcher that would be great. The main 911 number needs to ALWAYS be human answered. If the dispatcher makes the decision that it is non-emergent and transfers it over to the AI when they're busy then great, but those first words you hear after you hit 911 needs to be human.I design call centers for my job, we have an AI bot that can handle non emergency calls and what you said at the end is how we do it.
911 calls always start with a person, and the dispatcher can make the determination to transfer to the non-emergency bot. Y'all get too many calls that aren't actual emergencies tbh.
Edit: I looked up Versaterm's solution, CallTriage, and it's important to note that the AI isn't for 911 calls, it's for non-emergency line calls only. The article is conflating the non-emergency calls with 911 calls for shock value.
-
So they have money to spend on AI that will absolutely not be able to do the job half as well as a human, but not any money to spend on humans. Got it
Unlikely. AI is cheaper than humans, that's the whole point. And you have no idea how well it'll be able to do the job. Neither do they, which is why they're planning a test first.
-
This post did not contain any content.
How many people are going to die before they switch back?
-
Unlikely. AI is cheaper than humans, that's the whole point. And you have no idea how well it'll be able to do the job. Neither do they, which is why they're planning a test first.
I absolutely do have an idea how well it'll be able to do the job, based on AI's past performances in basically every other area, knowing its strong and weak points and knowing the job very well myself. Obviously I don't know for sure, but I'm not hopeful!
-
Those questions can be asked about humans who are making the decision too. How long does it take for a human to determine that it's an emergency? How many are going to mis-identify as a non-emergency? There's nothing unique about AI here.
We know how to run excellent human 911 service whereas automation in this area has always been frustrating garbage unfit to handle a cable tv outage.
-
Seems like a good idea to have a mechanism to divert the non-emergency calls off of the 911 dispatchers, then.
How do you do that without screening calls through the AI. Now anyone calm will spend an extra minute begging a computer for help whist screaming people will spend an extra 20 seconds before being bumped up .
-
We know how to run excellent human 911 service whereas automation in this area has always been frustrating garbage unfit to handle a cable tv outage.
Did you read the article? The reason they're looking into this is because they're not able to run an excellent human 911 service, they can't hire enough people to handle the volume of non-emergency calls that are coming to the emergency line.
-
Did you read the article? The reason they're looking into this is because they're not able to run an excellent human 911 service, they can't hire enough people to handle the volume of non-emergency calls that are coming to the emergency line.
If anyone ever tells you they can't hire enough of blank they are lying to you. People have been running excellent 911 service all over the country for longer than I've been alive maybe they should ask someone?
