A neighboring county to the one I work for briefly used an AI on their non-emergency lines (which are also mostly handled through their 911 center but go further back in the queue) just to route calls to the appropriate station or dispatcher.
It had one job, it had to understand the name of the town you were saying and transfer your call to whoever was responsible for that town. It wasn’t gathering any other info, it wasn’t dispatching police/fire/EMS, it wasn’t providing medical instructions or any of the other things you would expect
And it could not do that. When I had to transfer a caller to it, almost without fail I would say the name of the town they were calling for a couple times, being misunderstood every time, until the cranker would ask me “would you like to speak to a dispatcher?” And I’d say “yes” because that’s all I really wanted in the first place. Luckily it at least could understand “yes”
And I’m a reasonably clearly-spoken person who’s transferring calls with a level head. I’m not someone with a strong accent on top of a speech impediment, who’s drunk off my ass, and talking a mile a minute because they’re worked up over some emergency. The same can’t be said for all of my callers.
It lasted maybe a month before they got rid of it.
Now it could be that neighboring county really cheaped out on their AI assistant, and maybe there’s others out there that could do better, but every time I’ve been forced to use any kind of voice recognition , AI or not, that’s pretty much been my experience.
–
I also once had a call come in from one of our stations. They were asking for a call from another of our police departments regarding an incident that occurred at a store in that town that was related to a case they were working on.
There was a problem though, that store doesn’t have a location in that town. It does have locations in a couple nearby towns though, and we get that kind of confusion a lot because mailing addresses can be a little different than the actual municipality and sometimes people just casually lump a couple towns together, so I asked if they had the address for the store to make sure I was getting them in touch with the right department.
They gave me an address, complete with a town name, that is about 3 hours away from here.
I asked them why they were asking for the town they did when that address is clearly nowhere near here, and they answered that they asked Google’s AI where that address was and it told them that it was in that other town in our county.
And I’m not sure which is worse- that the AI was that far off on what should be a pretty simple query (“where is [town]?” Obviously [town] isn’t in [town] it’s in [totally different town] /s) or the fact that the human working the desk at that police station didn’t even bother to question that blatantly wrong answer.
So anyway, I really don’t like AI having anything to do with my job, it’s just bad at it.
This saddens me, and here’s why. We’ve all lost so much privacy and expectation of privacy between cameras everywhere, EULA’s wanting the rights to everything, “Smart” applicances phoning home, cars, stores, etc. Wherever there’s space to slap a device with a solar panel that can run a camera, microphone, and cellular modem there’s either one there, or there will be one.
So, given all that loss of privacy we turn to 911 calls. To a first approximation nobody calls 911 for the fun of it. There are idiots everywhere, and butt dialing is always going to be a thing, but you have an interface which people A) call when they deliberately decide they can’t handle the situation they’re in and need emergency help, and B) have the expecation that their call is going to be recorded.
So we have an audio dataset preselected for a purpose, dispatcher notes that are made with the expectation they may flow down into a court case somewhere along the line, and should have years of this stuff backed up as text and phone quality audio take up relatively little space.
And we don’t have an AI that’s trained effectively on that dataset? What went wrong? That data probably gets bought by insurance and realty companies for things like setting insurance rates or deciding where to build new subdivisions. Did we just collectively let the one aspect of the whole thing that would help people fall on the floor?
I work in 911 dispatch
A neighboring county to the one I work for briefly used an AI on their non-emergency lines (which are also mostly handled through their 911 center but go further back in the queue) just to route calls to the appropriate station or dispatcher.
It had one job, it had to understand the name of the town you were saying and transfer your call to whoever was responsible for that town. It wasn’t gathering any other info, it wasn’t dispatching police/fire/EMS, it wasn’t providing medical instructions or any of the other things you would expect
And it could not do that. When I had to transfer a caller to it, almost without fail I would say the name of the town they were calling for a couple times, being misunderstood every time, until the cranker would ask me “would you like to speak to a dispatcher?” And I’d say “yes” because that’s all I really wanted in the first place. Luckily it at least could understand “yes”
And I’m a reasonably clearly-spoken person who’s transferring calls with a level head. I’m not someone with a strong accent on top of a speech impediment, who’s drunk off my ass, and talking a mile a minute because they’re worked up over some emergency. The same can’t be said for all of my callers.
It lasted maybe a month before they got rid of it.
Now it could be that neighboring county really cheaped out on their AI assistant, and maybe there’s others out there that could do better, but every time I’ve been forced to use any kind of voice recognition , AI or not, that’s pretty much been my experience.
–
I also once had a call come in from one of our stations. They were asking for a call from another of our police departments regarding an incident that occurred at a store in that town that was related to a case they were working on.
There was a problem though, that store doesn’t have a location in that town. It does have locations in a couple nearby towns though, and we get that kind of confusion a lot because mailing addresses can be a little different than the actual municipality and sometimes people just casually lump a couple towns together, so I asked if they had the address for the store to make sure I was getting them in touch with the right department.
They gave me an address, complete with a town name, that is about 3 hours away from here.
I asked them why they were asking for the town they did when that address is clearly nowhere near here, and they answered that they asked Google’s AI where that address was and it told them that it was in that other town in our county.
And I’m not sure which is worse- that the AI was that far off on what should be a pretty simple query (“where is [town]?” Obviously [town] isn’t in [town] it’s in [totally different town] /s) or the fact that the human working the desk at that police station didn’t even bother to question that blatantly wrong answer.
So anyway, I really don’t like AI having anything to do with my job, it’s just bad at it.
This saddens me, and here’s why. We’ve all lost so much privacy and expectation of privacy between cameras everywhere, EULA’s wanting the rights to everything, “Smart” applicances phoning home, cars, stores, etc. Wherever there’s space to slap a device with a solar panel that can run a camera, microphone, and cellular modem there’s either one there, or there will be one.
So, given all that loss of privacy we turn to 911 calls. To a first approximation nobody calls 911 for the fun of it. There are idiots everywhere, and butt dialing is always going to be a thing, but you have an interface which people A) call when they deliberately decide they can’t handle the situation they’re in and need emergency help, and B) have the expecation that their call is going to be recorded.
So we have an audio dataset preselected for a purpose, dispatcher notes that are made with the expectation they may flow down into a court case somewhere along the line, and should have years of this stuff backed up as text and phone quality audio take up relatively little space.
And we don’t have an AI that’s trained effectively on that dataset? What went wrong? That data probably gets bought by insurance and realty companies for things like setting insurance rates or deciding where to build new subdivisions. Did we just collectively let the one aspect of the whole thing that would help people fall on the floor?