It was a real struggle to upvote you after seeing your username. That's the name of the furniture rental company that tried to screw me over a few months ago.
The British Post office rolled out a hugely buggy piece of software that bankrupted small business owners, got some sentenced to years in prison, and caused thirteen people to commit suicide because “computers can’t be wrong”
Do you not know what a computer is? It's literally a digital logical accountant! Yeah yeah, we should probably blame the programmers and engineers instead when shit goes sideways, but now I think we need to also hold CEOs accountable when they decide to inject faulty AI into mission critical systems...
There’s a reason why license agreements often stay there there are no warranties express or implied, no guarantees, and no fitness for any particular purpose.
If a building collapses. You blame the people who built the walls and poured the concrete, or the ones who chose the materials and approved the project?
In any case, often programmers and engineers retain no rights to the software they worked on. So whoever profits from the software should also shoulder the blame.
Those in charge, that approve the continued use of proven faulty software, should take all the blame, after significant faults have been proven anyways.
I mean you have a point, but still, 1+2+3≠15, and a bag of Doritos is not a gun. When AI fucks up this badly, the real guilty parties
(my AI keyboard wanted to replace guilty with gullible BTW, and I'm using the FUTO keyboard no less),
the real guilty parties are the ones in charge that allow such proven faulty systems to continue running for mission critical systems.
That's why cops love using dogs too. Courts have ruled that dogs can't lie. That means if a dog indicates you have contraband, then a search is warranted, even if nothing is found. This of course ignores that it is entirely possible the dog indicated contraband because the cop trained it to do so on command.
You could hold developers of algorithms, logic and even symbolic AI accountable.
However, it's a completely different story for AI based on deep neutral networks. After training they're just a bunch of weights and parameters without individual meaning and it's not a few, it's billions or trillions of them. And almost none of them were individually set, they're often randomly initialized and then automatically tuned by deep learning algorithms during training until the behavior / predictions of the neural net are "good enough".
It's practically impossible to review the network and when you test it you just get the result for the concrete test cases, you can't interpolate or assume even slightly different cases will behave similarly. You also can't fix an individual bug. You can just train again or more and this effort might fix the problem but it could also destroy something that worked before (catastrophic forgetting).
I feel like you're missing the point.
They're not saying to jail computers, they're saying be ware of political leaders using computers to abdicate responsibility.
We shut down companies for it though, and what AI vendors are doing is basically selling the ability to turn job roles into “accountability sinks”, where your true value is in taking the fall for AI when it gets it wrong (…enough that someone successfully sues).
If you want to put it in gun terms: The AI vendors are selling a gun that automatically shoots at some targets but not others. The targets it recommends are almost always profitable in the short term, but not always legal. You must hire a person to sit next to the gun and stop it from shooting illegal targets. It can shoot 1000 targets per minute.
We also don't give the murderer a free pass because they used a gun.
A tool is a tool, and the person who designed it or used it is responsible depending on why it caused a negative outcome. I know you clarified it later but it is so stupidly obvious I wanted to add to it.
The gun isn't running software in the background when humans are away either. See my other comment, when shit goes sideways, blame the programmers, engineers, and now the CEOs that decided to jam screwy AI up our collective asses..
So when a kid commits suicide because the Generative AI LLM agreed with him in a harmful way?
Edit: In before someone says something about how the gun manufacturers still shouldn't be held accountable.
Gen AI LLM's in this instance are products working as intended/designed, and are being used in a way that the manufacturer knows is harmful and admits is damaging. They also admit that there are no laws to safeguard persons against how the AI is designed, implemented etc and these things don't even have warning labels.
Guns by contrast have lots of laws involving how and where they can be sold and accessed, as well as by whom, and with respect to informing the user of the dangers. You don't sign a EULA or a TOS when you buy a gun, waiving your rights to sue. You don't agree to only arbitration.
I'll agree with you there, I shot myself in the arm when I was only 3 with a pellet gun. My dad realized his mistake and kept all guns away from me, until age 10, when he took me out to shoot some bottles and cans, and teach me proper gun safety.
Yes that might have been an earlier childhood lesson than many parents might agree to, but he was proper about what and when he taught me. Like, aside from the obvious of keep the gun on safety and never point it at anyone or anything unless you intend to use it, who thinks of things like, don't lean on a rifle with the barrel in the dirt? The dirt can and will clog the barrel and cause the gun to explode!
Anyways, back on point of AI...
Most parents aren't just up and giving their kids guns, but major corporations are shoving this AI shit up everyone's asses, as much as they can anyways, knowing good and well that one AI model says 1+2+3=15 and another AI model is suggesting people suffering pain to use heroin...
So what's the answer, avoid AI? Well fuck Google then.....
You don't get that option with AI these days if you simply want to Google something, they force you to use it. Google, Alexa, Siri, fuck our own US government is now using Grok!
Law enforcement will seize and use computers and the data they hold as evidence to convict criminals, just like any other tool that they might be warranted to seize.
Courts will examine the evidence of what it did to determine what role it played in the offence and whether it supports the allegation.
Likewise police complaints authorities could do the same in principle against the police; if someone were to give them a warrant and the power to execute it.
If a thing happens in public that was unwarranted and can be traced back to a police force or how they deployed any equipment, they can be judicially reviewed* for any decision to deploy that bit of kit. It's more a matter of will they actually be JR'd and will that be review be just and timely. * - in my country.
I don't think it's much different from how they deploy other tech like clubs and pepper spray, tear gas, tazers or firearms. If they have no fear of acting outwith their authority , that's a problem.
In some ways it might be easier to have an 'our word' vs 'their word' defense when they shoot someone, compared to a computer program that might literally document the abuse of power in its code or log files.
"Oops i dropped my notebook", is maybe easier than, "oops i accidentally deleted my local file and then sent a request to IT - that was approved by my manager - asking them to delete instead of restore any onsite or offsite backups".
This is unironically one of the main drivers of AI. As soon as all crucial social systems are inundated with AI, the built-in bias will be excused as "minor glitches" of the system, but the real reason was always a total lack of accountability.
"While computers are mechanical, the processes must be dictated and implemented by a human. Therefore, the only person culpable when a computer does something, is the human who wrote the instruction set, code, or algorithm."
If we interpreted the problem with human culpability and consequences, a whole lot of bugs would stop disappearing from software right quick.
58 Comments
stepan@lemmy.cafe · 60 pts · 313d
the AI decided ¯_(ツ)_/¯
PillowTalk420@lemmy.world · 42 pts · 313d
Excuse me, you dropped this: \
Cort@lemmy.world · 18 pts · 313d
No, I'm sorry, the AI has decided you don't need that forearm, but you're welcome to keep the hand.
PillowTalk420@lemmy.world · 5 pts · 313d
It was a real struggle to upvote you after seeing your username. That's the name of the furniture rental company that tried to screw me over a few months ago.
Cort@lemmy.world · 5 pts · 313d
Oof, I've heard they're as bad as payday loans, so I've never used them. Though I've also heard that rent-a-center is worse.
shalafi@lemmy.world · 2 pts · 313d
You the guy that trained Roland?
Cort@lemmy.world · 1 pts · 313d
SatyrSack@quokk.au · 2 pts · 312d
Lemmy spoilers use a different syntax than Reddit
::: spoiler Roland spoiler He's dead! :::
regedit@lemmy.zip · 1 pts · 312d
Another reason to call me handsome!
ch00f@lemmy.world · 46 pts · 313d
Outside of law enforcement, this is certainly how shitty customer service policies get enforced. In other words, "Computer says no".
Semi_Hemi_Demigod@lemmy.world · 32 pts · 313d
The British Post office rolled out a hugely buggy piece of software that bankrupted small business owners, got some sentenced to years in prison, and caused thirteen people to commit suicide because “computers can’t be wrong”
ch00f@lemmy.world · 24 pts · 313d
Oh yeah! I forgot about that!
https://en.wikipedia.org/wiki/British_Post_Office_scandal
Seleni@lemmy.world · 14 pts · 313d
Ewwww
dogfoodeater@lemmy.blahaj.zone · 6 pts · 313d
Here in Australia they rolled out an automated system to calculate welfare overpayments and issue debts. It didn't quite work of course and hundreds of thousands of the poorest Australian were issued with false debts, some of whom died or committed suicide before they could be repaid. People still keep floating the idea of automation and AI in our welfare systems...
Xaphanos@lemmy.world · 4 pts · 313d
Not to be that guy but... Link?
Semi_Hemi_Demigod@lemmy.world · 3 pts · 313d
Here ya go
Xaphanos@lemmy.world · 1 pts · 313d
Thank you
Kyrgizion@lemmy.world · 3 pts · 313d
Not only UK's fuckup, also Fujitsu's.
over_clox@lemmy.world · 27 pts · 313d
Now where does this thought come from?
Do you not know what a computer is? It's literally a digital logical accountant! Yeah yeah, we should probably blame the programmers and engineers instead when shit goes sideways, but now I think we need to also hold CEOs accountable when they decide to inject faulty AI into mission critical systems...
https://lemmy.dbzer0.com/post/55990956
henfredemars@infosec.pub · 17 pts · 313d
There’s a reason why license agreements often stay there there are no warranties express or implied, no guarantees, and no fitness for any particular purpose.
Semi_Hemi_Demigod@lemmy.world · 19 pts · 313d
“This software is useless and should not be used by anyone for any purpose” is my favorite part of license agreements.
mogranja@lemmy.world · 8 pts · 313d
If a building collapses. You blame the people who built the walls and poured the concrete, or the ones who chose the materials and approved the project?
In any case, often programmers and engineers retain no rights to the software they worked on. So whoever profits from the software should also shoulder the blame.
over_clox@lemmy.world · 1 pts · 313d
Those in charge, that approve the continued use of proven faulty software, should take all the blame, after significant faults have been proven anyways.
I mean you have a point, but still, 1+2+3≠15, and a bag of Doritos is not a gun. When AI fucks up this badly, the real guilty parties
(my AI keyboard wanted to replace guilty with gullible BTW, and I'm using the FUTO keyboard no less),
the real guilty parties are the ones in charge that allow such proven faulty systems to continue running for mission critical systems.
Like fuck, a bag of Doritos is not a fucking gun!
Zorsith@lemmy.blahaj.zone · 17 pts · 313d
IT disagrees. Misbehaving hardware can be taken out back and shot.
faythofdragons@slrpnk.net · 9 pts · 313d
We all remember what happened to the printer in Office Space
popekingjoe@lemmy.world · 7 pts · 313d
Damn it feels good to be a gangsta.
ChonkyOwlbear@lemmy.world · 16 pts · 313d
That's why cops love using dogs too. Courts have ruled that dogs can't lie. That means if a dog indicates you have contraband, then a search is warranted, even if nothing is found. This of course ignores that it is entirely possible the dog indicated contraband because the cop trained it to do so on command.
affenlehrer@feddit.org · 10 pts · 313d
You could hold developers of algorithms, logic and even symbolic AI accountable.
However, it's a completely different story for AI based on deep neutral networks. After training they're just a bunch of weights and parameters without individual meaning and it's not a few, it's billions or trillions of them. And almost none of them were individually set, they're often randomly initialized and then automatically tuned by deep learning algorithms during training until the behavior / predictions of the neural net are "good enough".
It's practically impossible to review the network and when you test it you just get the result for the concrete test cases, you can't interpolate or assume even slightly different cases will behave similarly. You also can't fix an individual bug. You can just train again or more and this effort might fix the problem but it could also destroy something that worked before (catastrophic forgetting).
QuantumTickle@lemmy.zip · 9 pts · 313d
We don't jail the gun for murder.
PeriodicallyPedantic@lemmy.ca · 5 pts · 313d
I feel like you're missing the point.
They're not saying to jail computers, they're saying be ware of political leaders using computers to abdicate responsibility.
kibiz0r@midwest.social · 3 pts · 313d
We shut down companies for it though, and what AI vendors are doing is basically selling the ability to turn job roles into “accountability sinks”, where your true value is in taking the fall for AI when it gets it wrong (…enough that someone successfully sues).
If you want to put it in gun terms: The AI vendors are selling a gun that automatically shoots at some targets but not others. The targets it recommends are almost always profitable in the short term, but not always legal. You must hire a person to sit next to the gun and stop it from shooting illegal targets. It can shoot 1000 targets per minute.
JoeBigelow@lemmy.ca · 1 pts · 313d
Sounds like a fun job if the acceptable failure rate is like, 50%
snooggums@piefed.world · 1 pts · 313d
We also don't give the murderer a free pass because they used a gun.
A tool is a tool, and the person who designed it or used it is responsible depending on why it caused a negative outcome. I know you clarified it later but it is so stupidly obvious I wanted to add to it.
QuantumTickle@lemmy.zip · -1 pts · 313d
snooggums@piefed.world · 3 pts · 313d
No, I agreed with you in a slightly different way.
Great reading comprehension.
QuantumTickle@lemmy.zip · -3 pts · 313d
over_clox@lemmy.world · -1 pts · 313d
The gun isn't running software in the background when humans are away either. See my other comment, when shit goes sideways, blame the programmers, engineers, and now the CEOs that decided to jam screwy AI up our collective asses..
QuantumTickle@lemmy.zip · 11 pts · 313d
We don't jail gun manufacturers either.
When a tool is used to kill a human, the user of the tool is guilty.
atrielienz@lemmy.world · 2 pts · 313d
So when a kid commits suicide because the Generative AI LLM agreed with him in a harmful way?
Edit: In before someone says something about how the gun manufacturers still shouldn't be held accountable.
Gen AI LLM's in this instance are products working as intended/designed, and are being used in a way that the manufacturer knows is harmful and admits is damaging. They also admit that there are no laws to safeguard persons against how the AI is designed, implemented etc and these things don't even have warning labels.
Guns by contrast have lots of laws involving how and where they can be sold and accessed, as well as by whom, and with respect to informing the user of the dangers. You don't sign a EULA or a TOS when you buy a gun, waiving your rights to sue. You don't agree to only arbitration.
QuantumTickle@lemmy.zip · 5 pts · 313d
If a child shot themself, I'd blame the parents.
Keep 'em coming. I can do this all day.
over_clox@lemmy.world · 3 pts · 313d
I'll agree with you there, I shot myself in the arm when I was only 3 with a pellet gun. My dad realized his mistake and kept all guns away from me, until age 10, when he took me out to shoot some bottles and cans, and teach me proper gun safety.
Yes that might have been an earlier childhood lesson than many parents might agree to, but he was proper about what and when he taught me. Like, aside from the obvious of keep the gun on safety and never point it at anyone or anything unless you intend to use it, who thinks of things like, don't lean on a rifle with the barrel in the dirt? The dirt can and will clog the barrel and cause the gun to explode!
Anyways, back on point of AI...
Most parents aren't just up and giving their kids guns, but major corporations are shoving this AI shit up everyone's asses, as much as they can anyways, knowing good and well that one AI model says 1+2+3=15 and another AI model is suggesting people suffering pain to use heroin...
So what's the answer, avoid AI? Well fuck Google then.....
atrielienz@lemmy.world · 4 pts · 313d
They knew what I meant and chose to ignore the meaning to avoid the question.
QuantumTickle@lemmy.zip · 2 pts · 313d
Why can guns be something that can be responsibly used but AI cannot?
over_clox@lemmy.world · 0 pts · 313d
One can simply decide to never use a gun.
You don't get that option with AI these days if you simply want to Google something, they force you to use it. Google, Alexa, Siri, fuck our own US government is now using Grok!
NewPerspective@lemmy.world · 1 pts · 313d
You are so close to getting it...
over_clox@lemmy.world · 1 pts · 313d
Okay, I'll bite.
What am I missing here NewPerspective?
skulblaka@sh.itjust.works · 0 pts · 313d
Be a lot cooler if you did
QuantumTickle@lemmy.zip · 5 pts · 313d
When guns have no legal uses, this is a direction we should go. Until then, this holds people accountable for other people misusing their product.
In a chain of responsibility, the last person who's accountable should be punished, not the first.
QuantumTickle@lemmy.zip · 0 pts · 313d
PeriodicallyPedantic@lemmy.ca · 0 pts · 313d
When a human dies because a tool was designed with needless danger, the manufacturer is often prosecuted.
But again, I think you're missing the point.
mogranja@lemmy.world · 3 pts · 313d
Blame the programmers? Yeah, no. The software is owned by the company, blame them.
over_clox@lemmy.world · 1 pts · 313d
So, I think that the open source developers should file a class action lawsuit for stealing their code.
Go ahead, ask Linus Torvalds, I bet he's not exactly happy with the current trajectory...
individual@toast.ooo · 8 pts · 313d
infinitesunrise@slrpnk.net · 6 pts · 313d
All technology has the potential to be both liberatory and oppressive, all that ever matters is who wields it and to what end.
Lewis Herber (Murray Bookchin) - Towards a Liberatory Technology
bryndos@fedia.io · 6 pts · 312d
Law enforcement will seize and use computers and the data they hold as evidence to convict criminals, just like any other tool that they might be warranted to seize.
Courts will examine the evidence of what it did to determine what role it played in the offence and whether it supports the allegation.
Likewise police complaints authorities could do the same in principle against the police; if someone were to give them a warrant and the power to execute it.
If a thing happens in public that was unwarranted and can be traced back to a police force or how they deployed any equipment, they can be judicially reviewed* for any decision to deploy that bit of kit. It's more a matter of will they actually be JR'd and will that be review be just and timely. * - in my country.
I don't think it's much different from how they deploy other tech like clubs and pepper spray, tear gas, tazers or firearms. If they have no fear of acting outwith their authority , that's a problem.
In some ways it might be easier to have an 'our word' vs 'their word' defense when they shoot someone, compared to a computer program that might literally document the abuse of power in its code or log files.
"Oops i dropped my notebook", is maybe easier than, "oops i accidentally deleted my local file and then sent a request to IT - that was approved by my manager - asking them to delete instead of restore any onsite or offsite backups".
Kyrgizion@lemmy.world · 3 pts · 313d
This is unironically one of the main drivers of AI. As soon as all crucial social systems are inundated with AI, the built-in bias will be excused as "minor glitches" of the system, but the real reason was always a total lack of accountability.
Damage@slrpnk.net · 2 pts · 313d
Yeah, not like people in power, who are held accountable all the time!
JeSuisUnHombre@lemmy.zip · 2 pts · 313d
Same argument for cars
netvor@lemmy.world · 1 pts · 306d
...or any inanimate objects, really.
KingThrillgore@lemmy.ml · 1 pts · 313d
"While computers are mechanical, the processes must be dictated and implemented by a human. Therefore, the only person culpable when a computer does something, is the human who wrote the instruction set, code, or algorithm."
If we interpreted the problem with human culpability and consequences, a whole lot of bugs would stop disappearing from software right quick.
shalafi@lemmy.world · 2 pts · 313d
If a mistake in my code exposes me to criminal investigation, I'm getting a new job.