The media in this post is not displayed to visitors. To view it, please go to the original post.
I was a 911 dispatcher for a decade.
To really understand how absolutely terrible an idea it is to have AI answer 911 calls, I'll present you with a single real-life call:
Upon answer, the caller calmly began ordering a pizza. Large, pepperoni. Side order of chicken wings. To be delivered promptly.
The only sign the caller gave that things were amiss? She rather ignored the first few questions from the calltaker.
Because she was being physically abused, had already been beaten, and her abuser was *IN THE ROOM* listening to her order pizza.
Tell me how your AI is going to read between those lines.
AI has no place in 911 calls. From Mystery AI Hype Theater 3000’s latest episode “State of Emergency”.(with @emilymbender)
Jonathan Lamothe likes this.
reshared this
![Is there a way of writing the expression<br>\[ \sum_{n=1}^m f(n) \]<br>in a way that omits the term where $n=a$?</p><p>I know it could be written as<br>\[ \left( \sum_{n=1}^m f(n) \right) - f(a) \]<br>but this doesn't work when $f(a)$ is undefined.](https://social.jlamothe.net/photo/13528377976a61653463c9a386467268-1.png "Is there a way of writing the expression<br>\[ \sum_{n=1}^m f(n) \]<br>in a way that omits the term where $n=a$?</p><p>I know it could be written as<br>\[ \left( \sum_{n=1}^m f(n) \right) - f(a) \]<br>but this doesn't work when $f(a)$ is undefined.")
Ian Campbell 🏴
in reply to Ian Campbell 🏴 • • •This call happened more than twenty years ago, and I didn't take it.
The calltaker who did was brilliant, because it took her exactly one repetition, only one missed beat, to understand what the call really was.
We spent several hours in training *just* dissecting, sitting with, and understanding that one call, because the concept behind it was so important.
reshared this
Bernie Luckily Does It and 🇺🇦 haxadecimal 🚫👑 reshared this.
Ian Campbell 🏴
in reply to Ian Campbell 🏴 • • •I want to know how many caller's lives are worth not having to hire human 911 dispatchers, or pay them a decent wage.
Because this will kill people. Not get people killed. AI use in 911 answering will actively kill people.
reshared this
Shannon Prickett and Bernie Luckily Does It reshared this.
Irene Zhang
in reply to Ian Campbell 🏴 • • •Bernie Luckily Does It reshared this.
Irene Zhang
in reply to Irene Zhang • • •I just finished reading Careless people which is a totally unsurprising but trashy book about the choices made by Facebook when their product clearly has the potential to and then did kill people. Per the title, they just did not care. Although it would be more correct to say that they cared about other things, like stock price and growth rates more. It took me most of my 20s/30s to wrap my head around the fact that these people were not just dumb or racist or lazy or anything other than actively malicious. Because they don’t think of most people as people.
ETA: actually it was more like my 30s since it probably started with the first Trump presidency.
Articulated Intelligence
in reply to Irene Zhang • • •Hugs4friends ♾🇺🇦 🇵🇸😷
in reply to Articulated Intelligence • • •Sensitive content
fuzzyfuzzyfungus
in reply to Ian Campbell 🏴 • • •This seems like the massively important point to hammer home as often(and as loudly and publicly) as possible.
There might be some genuine kool-aide drinkers out there(who still shouldn't be allowed near important things); but more than a few "AI" initiatives are pretty cynical exercises in people trying to get away with adjusting spending to match how much they care; and that's not something you want people to get the benefit of the doubt for.
Bernie Luckily Does It reshared this.
Tanek Rune
in reply to Ian Campbell 🏴 • • •Sounds like an exceptionally stupid idea to get the machine that can't even properly take drive-thru food orders right to manage emergency calls.
There's already enough complexity with emergency services. Just getting the right people (fire, ambulance, police) to a particular location with the proper situational information quickly is enough of a challenge.
Solitha
in reply to Ian Campbell 🏴 • • •Holy shit. This was my career and I cannot even.
I remember calls that went both ways. Ones where tone of voice and evasive answers sent up red flags. Another where someone called yelling they'd been robbed, and it was the fact they sounded utterly pissed that clued me off (it was a civil dispute with a cab driver, not an emergency).
How the hell would an AI understand when someone talking suddenly starts screaming incoherently?
SomeVeganCheeseIsOk
in reply to Ian Campbell 🏴 • • •How the fuck will AI know what the drunk local in distress is referring to when they call in and use local descriptions for locations and you're rural enough that geolocation of calls sucks rancid ass through a straw?
Sorry about the language but really not sorry at all, AI for 911 is *stupid*.
Matt Blaze
in reply to Ian Campbell 🏴 • • •AA
in reply to Ian Campbell 🏴 • • •Alison Meeks
in reply to Ian Campbell 🏴 • • •Mike Sheward
in reply to Ian Campbell 🏴 • • •@mttaggart I called non emergency 911 to report a cow loose in a residential area a few weeks back and was met with a voice chatbot
it didn’t understand me when trying to explain about the cow, so i hung up
a human dispatcher called me back 10 mins later
Justin Derrick
in reply to Mike Sheward • • •Arthur van der Harg
in reply to Justin Derrick • • •Justin Derrick
in reply to Arthur van der Harg • • •Epic Null
in reply to Ian Campbell 🏴 • • •A person calling 911 is also potentially in the worst point of their lives. Many of said points will fuck up peoples minds.
A 911 opperator has to get a caller from "ramble" to "I am covered in blood" to "I've been stabbed" during a time where blood is not in their head. I doubt AI can navigate that.
Bernie Luckily Does It reshared this.
Nicole Parsons
in reply to Ian Campbell 🏴 • • •newsweek.com/jess-edwards-teen…
theguardian.com/commentisfree/…
The people funding & enabling AI want the wife beating, marital murder/suicide, & child abuse.
Male supremacy is demonstrated through a hierarchy that preserves impunity, entitlement, & unearned privilege.
dailykos.com/stories/2026/3/13…
dailykos.com/stories/2026/1/24…
dailykos.com/stories/2026/1/3/…
Support the GOP & get a DOJ investigation closed.
christianitytoday.com/2025/03/…
nbcnews.com/politics/congress/…
newrepublic.com/post/210178/ma…
mprnews.org/story/2026/04/23/u…
Why aren’t Republicans thrilled by the fall in teen pregnancies? | Arwa Mahdawi
Arwa Mahdawi (the Guardian)Nicole Parsons reshared this.
Franceska Mann
in reply to Nicole Parsons • • •@Npars01
Child abuse is a polite way of saying these people are pedophile child rapists.
Nicole Parsons
in reply to Franceska Mann • • •They definitely support a culture of rape & child abuse, possibly because their own childhoods said violence & abuse was "normal" & saw it as a way to assert identity in a group.
psychologytoday.com/ca/blog/th…
childrenssociety.org.uk/what-w…
mediamatters.org/nick-fuentes/…
equimundo.org/making-the-conne…
texasmonthly.com/news-politics…
The mistreatment of "their inferiors" is a hallmark characteristic of Christian nationalism, white supremacy, & male privilege.
imprintnews.org/opinion/white-…
Is There a Link Between White Supremacy and Childhood Trauma?
Jeffrey B. Rubin Ph.D. (Psychology Today)Nicole Parsons reshared this.
Franceska Mann
in reply to Nicole Parsons • • •@Npars01
When you see other people as objects, not as human beings, abuse follows.
Tired Panda
in reply to Ian Campbell 🏴 • • •Helena B
in reply to Ian Campbell 🏴 • • •lupus_blackfur
in reply to Ian Campbell 🏴 • • •@alex
Easy Peasy...
The so-called "AI" will simply dispatch SWAT any time a pizza is ordered on a 911 call...
Why is this difficult??
FFS.
🤦♂️🤷♂️🤡🫏🖕🖕💩💩
Oregon Pacifist
in reply to Ian Campbell 🏴 • • •Dave from Burnout Café
in reply to Ian Campbell 🏴 • • •I definitely agree with the point you are making.
Emergency service call dispatchers definitely should not be replaced by AI tech, the human presence is still absolutely essential.
derptron
in reply to Ian Campbell 🏴 • • •The really cool thing here is that before I get to hear how an AI is going to be the last thing I hear when I dial 911 during my heart attack, I get to learn that if I'd made it even a decade ago it would have been recorded and sold to a foreign whoever so they could do whatever the fuck they want with my last words. "My" government already allowed, participated, and profited by this shit.
Every day I learn more and more that I'm just fucking food and so is everyone I care about.
Truone
in reply to Ian Campbell 🏴 • • •Sara
in reply to Ian Campbell 🏴 • • •somewhere I have the transcript of the 911 call I made the time I was hiding in the bathroom with the kids while my ex tried to kick in the doors between him and us
it took the officers 9 minutes to get there
I'm glad there was a human being on the other end of that line
Texan Tally
in reply to Ian Campbell 🏴 • • •Yeah, this is some scary crap. If someone is badly hurt and losing consciousness, or they're whispering because it's an active shooter situation and they can't speak clearly... what then? There are too many human variables to count that make it difficult to discern what's happening with human to human interaction (though good dispatchers try their damnedest for sure), how the hell is AI gonna manage?
Scary fucking times we're living in. As Ian Malcolm said in Jurassic Park, "God help us, we're in the hands of engineers."
Toni Aittoniemi
in reply to Ian Campbell 🏴 • • •Mensch, Marina
in reply to Ian Campbell 🏴 • • •Momo
in reply to Ian Campbell 🏴 • • •There is a video game where you play as a 911 dispatcher and this is exactly one of these call scenarios that can happen. Sadly, the first time I got such a call I missed it.
There were also calls with people in medical distress unable to speak but able to answer basic yes/no questions by pushing or not pushing a random number button on the phone generating a beep. Another scenario where I can see AI will fail.
Hammerwell
in reply to Momo • • •Purple
in reply to Ian Campbell 🏴 • • •BeeCycling
in reply to Ian Campbell 🏴 • • •14mission
in reply to Ian Campbell 🏴 • • •🔶Mark Nicoll 3.5%🏴🇬🇧🇪🇺🇺🇳
in reply to Ian Campbell 🏴 • • •Cavyherd
in reply to 🔶Mark Nicoll 3.5%🏴🇬🇧🇪🇺🇺🇳 • • •& this probably doesn't account for the proportion of customers who just repeat "representative" until they get alive human on the phone 😂
Frank Heijkamp
in reply to Ian Campbell 🏴 • • •That sounds like a terrible idea.
'What could possibly go wrong?'
refraction 折射
in reply to Frank Heijkamp • • •Happy Media
in reply to Ian Campbell 🏴 • • •paelnever
in reply to Ian Campbell 🏴 • • •Ian Campbell 🏴
in reply to paelnever • • •@paelnever A fair point, but keep in mind that you need to meet people where they're at in some ways, and "AI" is the cognitive frame most people operate with around this.
You're not wrong that there are drawbacks to starting out in that cognitive frame though.
Colin Rowat
in reply to Ian Campbell 🏴 • • •I have no general view on LLMs reading subtle cues.
Your challenge - "Tell me how you AI is going to read between those lines" - is both interesting and easily testable, though.
Here is the response of Kimi's default 'fast' model:
> A caller who "orders a pizza" on a 911 line while ignoring direct questions is almost certainly in a situation where they cannot speak freely—often domestic violence, home invasion, or a hostage situation where an assailant is within earshot.
Colin Rowat
in reply to Colin Rowat • • •Your sentence stressing the importance of this case led me to wonder if the LLM might get it right:
"We spent several hours in training *just* dissecting, sitting with, and understanding that one call, because the concept behind it was so important."
Thus, I thought that there was a good chance that some call like this had been written about, making its way into the training data.
If so, then the LLM doesn't need to understand anything subtle - just stochastically parrot.
Ian Campbell 🏴
in reply to Colin Rowat • • •@rowat_c Sorry, "easily testable" betrays a failure to understand context, training, and differing incentives in deployment.
Please do not mistake your prompt for 10 years of field experience.
Kye Fox
in reply to Colin Rowat • • •@rowat_c Numerous instances are in the training data now. It would probably handle a new form of the same kind of call well, but not a new form with the cheap models most of these systems use. The latest and greatest and most expensive _might_ get it.
It's because I use LLMs and know how they work that I know the likely implementation means they aren't ready for life and death scenarios and may never be.
2¢
in reply to Ian Campbell 🏴 • • •Ian Campbell 🏴
in reply to 2¢ • • •dennislong82
in reply to Ian Campbell 🏴 • • •dennislong82
in reply to Ian Campbell 🏴 • • •Ian Campbell 🏴
in reply to dennislong82 • • •Timothy Wolodzko
in reply to Ian Campbell 🏴 • • •ChookMother 🇦🇺🦘
in reply to Ian Campbell 🏴 • • •Maria Langer
in reply to Ian Campbell 🏴 • • •Ray Greenwell
in reply to Ian Campbell 🏴 • • •@alex
“You’re absolutely right, that was my mistake and I own it. I told you that help was on the way, but it wasn’t.”
Ian Campbell 🏴
in reply to Ray Greenwell • • •Farms In Schools Initiatives
in reply to Ian Campbell 🏴 • • •Flash Mob Of One
in reply to Ian Campbell 🏴 • • •David Scott Moyer
in reply to Ian Campbell 🏴 • • •Aleksei � Matiushkin
in reply to Ian Campbell 🏴 • • •② Each and every student in applied math would tell you that particular case you’ve described would be perfectly caught by AI (unlike many others though.)
Ian Campbell 🏴
in reply to Aleksei � Matiushkin • • •@mudasobwa 75% of the *single-context, specialized* deployments that I've encountered can't understand basic requests (think: restricted-context customer environments for a specific business).
911 dispatching is the definitional opposite of a restricted context. It requires deep general knowledge across a number of contexts that have to be integrated and lead to correct inference *the first time.*
Suggested reading: en.wikipedia.org/wiki/Dunning%…
Dunning–Kruger effect - Wikipedia
Contributors to Wikimedia projects (Wikimedia Foundation, Inc.)Cavyherd
in reply to Ian Campbell 🏴 • • •Ian Campbell 🏴
in reply to Cavyherd • • •Process Parsnip
in reply to Cavyherd • • •Sensitive content
@cavyherd fact. I've never seen so many blockable replies on a thread started by a man before.
@neurovagrant
Steve Holden
in reply to Ian Campbell 🏴 • • •Ajira
in reply to Ian Campbell 🏴 • • •