OpenAI sued for defamation after ChatGPT fabricated yet another lawsuit
- Thread starter JournalBot
- Start date
One would think that the "loudest voice in America fighting for gun rights" would be enough of a public figure that the "actual malice" standard would apply.
Upvote
127
(133
/
-6)
Upvote
94
(97
/
-3)
If you think its not liable because people should know its not facts, consider if someone could get around the liable laws by having their "AI" say stuff for them. You could say anything about anyone - just program the AI with that specific bias.
Upvote
70
(94
/
-24)
Upvote
166
(167
/
-1)
Good. It's about time companies stop escaping legal actions by hiding behind the algorithm that they programmed.
Upvote
38
(94
/
-56)
The complainant would agree with you from the lawsuit ...
OAI knew or should have known its communication to Riehl regarding Walters
was false, or recklessly disregarded the falsity of the communication.
Upvote
20
(21
/
-1)
“Guns don’t kill people, people kill people. A gun is just a tool.”
By gun rights/ 2A logic you can’t blame a toolmaker for what someone does with it, so why doesn’t this logic apply to AI?
By gun rights/ 2A logic you can’t blame a toolmaker for what someone does with it, so why doesn’t this logic apply to AI?
Upvote
225
(239
/
-14)
Given the accuracy clause in ChatGPT terms of use at https://openai.com/policies/terms-of-use, presumably the decision to ‘publish’ inaccurate information is not something to be laid at OpenAI’s door in any strict legal sense. Suing OpenAI looks like more publicity not a serious legal case. IANAL
Upvote
113
(124
/
-11)
And the content isn't generated by machine learning so was a really bad comparison.
Upvote
77
(79
/
-2)
Per the OpenAI accuracy clause then, does it make a difference that OpenAI not only knew the output could be inaccurate but explicitly states that it could be in advance, and made awareness of that part of the terms of use.
Upvote
38
(43
/
-5)
Except that I think the person using the AI should be held responsible. Surely it qualifies as 'reckless disregard for the truth'?
The AI doesn't have agency here. It cannot be responsible, and thus cannot be used to dodge responsibility. If you use it, you should be liable for what that use results in.
Upvote
115
(121
/
-6)
"One problem with asking AI chatbots like ChatGPT for case summaries is that case law is not widely published online, the Free Law Project tweeted."
No, the problem with chatbots is not that other people are not making more work available for the companies that make chatbots to steal. The problem is that what comes out the end of the "AI" chatbot word-chipper is no more the synthesis of intelligent processing than the vomit spewed by a drunk college student on a Saturday morning. There may be some novelty value in it but the only place it really belongs is in the toilet as soon as possible.
No, the problem with chatbots is not that other people are not making more work available for the companies that make chatbots to steal. The problem is that what comes out the end of the "AI" chatbot word-chipper is no more the synthesis of intelligent processing than the vomit spewed by a drunk college student on a Saturday morning. There may be some novelty value in it but the only place it really belongs is in the toilet as soon as possible.
Upvote
19
(50
/
-31)
This article is a bit of a mess… like a ChatGPT response.
1. The newsworthy part here is suing an AI vendor as liable for their product’s content. IANAL, but just from the article, it seems clear it has to do with “spreading” and with damages. If the same piece of liebel is only generated once per user, can it be considered “spreading” disinformation at all, and is it damaging enough to win a lawsuit?
So I suspect this suit (aside from being filed by a far-right nut), would be tossed, setting a precedent which may actually not be so good for society en large.
2. The other question is, is this hallucination reproducible? The article says nothing of it, and the lawsuit does not. It simply describes a single instance. As we know, the AI may or many reply with completely different information in response to the same (or all so slightly differently-phrased) query. It just spews well-phrased text.
3. “It's not the first time that ChatGPT has completely fabricated a lawsuit.”
Obviously, for anybody who’s used ChatGPT or been following up on LLM’s (i.e. most of us Ars readers). ChatGPT likely completely fabricates lawsuits many times a day, alongside any other fabrications.
1. The newsworthy part here is suing an AI vendor as liable for their product’s content. IANAL, but just from the article, it seems clear it has to do with “spreading” and with damages. If the same piece of liebel is only generated once per user, can it be considered “spreading” disinformation at all, and is it damaging enough to win a lawsuit?
So I suspect this suit (aside from being filed by a far-right nut), would be tossed, setting a precedent which may actually not be so good for society en large.
2. The other question is, is this hallucination reproducible? The article says nothing of it, and the lawsuit does not. It simply describes a single instance. As we know, the AI may or many reply with completely different information in response to the same (or all so slightly differently-phrased) query. It just spews well-phrased text.
3. “It's not the first time that ChatGPT has completely fabricated a lawsuit.”
Obviously, for anybody who’s used ChatGPT or been following up on LLM’s (i.e. most of us Ars readers). ChatGPT likely completely fabricates lawsuits many times a day, alongside any other fabrications.
Upvote
59
(60
/
-1)
I'm sorry, but this "suit" reads like the proverbial ambulance chasers wrote it, up to an including all the bullshit bravado trollish language. I'm surprised they didn't write "we bring the smartest suits, the best suits!" :/
Upvote
39
(41
/
-2)
Thing is it can't there is no difference between true and false information to the model, it's just a statical likely hood that the output is coherent based on training data. The model doesn't "know" it's lying
Upvote
105
(107
/
-2)
Do we know if openai was trained on the cases in question, and then made stuff up? Or just trained on other cases, and then made stuff up. The first would seem to more likely fall into the "knowingly made false statements" while the second might get away with unknowingly making false statements.
Upvote
-18
(1
/
-19)
Exactly! Also the model doesn't know it's telling the truth! It knows nothing!
That's one reason why I hate the term "hallucinate" for these models. When it gets an answer wrong everyone says it "hallucinated" the wrong answer. But it didn't. It's process for getting the answer wrong is exactly the same as its process for getting the answer right - it just followed a different random path to generate the text string it output and put out something we judge as incorrect instead of correct. If you want to say it "hallucinates" wrong answers you might as well say it "hallucinates" right answers because it's doing the exact same thing in both cases.
Upvote
146
(148
/
-2)
We now have a plaintiff that knows ChatGPT will publish mistruths about him when prompted. True those mistruths are not published to the world at large, but it is clear that ChatGPT will publish them to anyone that asks. This is most definitely a ChatGPT problem regardless of the TOS which do NOT apply to the person that the mistruths are generated about -- just like I cannot sign you into a contract for which you know nothing about.
This will be an interesting case as, if I remember correctly, libel requires others to see it. I don't know what happens for a collection of individually tailored publications that are generated on request. And it may be a state-by-state issue which is going to be awful.
A whole bunch of law students and professors will have competing journal articles soon.
Upvote
-3
(18
/
-21)
There is a fundamental difference between a tool inadvertently saying some incorrect thing - just like a search engine or any statistical tool - and intentionally inducing the tool to produce a predetermined result. That is, the former is other people not understanding what the tool is and getting wrapped up in media and marketing hysteria while the latter has nothing really to do with the tool, which is used as a veil for traditional disinformation. The latter is an expression of intent, which is a necessary component for defamation under the law (or negligence, but that's not part of the hypothetical); the former has no intent and is incidental.
There will be murky areas in the middle and establishing requirements for these cases is almost always difficult, but it's a false dichotomy. I just hope that judges don't get as confused and duped by the hype as these lawyers have been.
Upvote
21
(21
/
0)
I have a feeling that this is going to get a variation of the classic "right wing personality ChatGPT is purely for entertainment, only and idiot would take it seriously" defense.
Last edited:
Upvote
6
(8
/
-2)
Might as well sue Hewlett-Packard because a day-drunk engineer misused their calculator and a bridge ended up falling down.
ShitGPT, as little regard as I have for it, was performing exactly as designed and advertised.
ShitGPT, as little regard as I have for it, was performing exactly as designed and advertised.
Upvote
-19
(7
/
-26)
It's not like any of this is new, except for the media attention that started last year. Our inept media spheres are fueling the misinformation about these tools and that's the biggest threat for the time being, Altman's unhinged "human extinction" quips and calls for legal motes or Musk's cries for attention notwithstanding.
Upvote
24
(26
/
-2)
It's frustrating because when you know how these LLMs work you're looking at all of this like "what the heck is going on? Am I taking crazy pills or something?" Because it's just beads on a string. Really good context to get a really high probability that the next word on that string is going to be a good one, but that's all it is.
I'll give Altman this - he's damn good at being a hype machine. I know that's his skillset and how he's done so well in Silicon Valley up to this point, but to manage to get so many people to think that his LLM is somehow more than just a really damn fine autocomplete tool - PT Barnum would applaud. If anyone can manage to make his eyeball scanning cryptocoin dreams work, it'll be him.
Upvote
54
(55
/
-1)
The word "uncovered" seems misleading here. The text, and the claims made in the text, did not exist before Riehl asked the LLM a question. This might seem like nitpicking, but I think it's important enough since this is a story about legal issues.
Upvote
62
(62
/
0)
While we’re suggesting laws about LLMs, can we have a law that requires them to be called “believable bullshit generators”? Because that’s literally what they are.
Some AI (neural net) systems are actually trained to be accurate/truthful (like the ones that are trained to find medical problems in cat scans or X-rays). IBM’s Watson was trained to give correct answers to Jeopardy prompts. LLMs are not trained that way; they are trained to sound like something somewhere might actually say.
Some AI (neural net) systems are actually trained to be accurate/truthful (like the ones that are trained to find medical problems in cat scans or X-rays). IBM’s Watson was trained to give correct answers to Jeopardy prompts. LLMs are not trained that way; they are trained to sound like something somewhere might actually say.
Upvote
7
(15
/
-8)
I'm not sure if falling back on the terms of service will work here. The terms of service (ToS) is a contract between the company providing the tool and that company's user/customer. It can't override actual law, which includes defamation / libel law. And it can't force a 3rd party, who is not a user/customer of that company and never agreed to its ToS, to surrender any rights.
At most, the ToS might assign responsibility for the consequences of any ChatGPT-generated content to the user who prompted it. If the company can throw that user under the bus via the ToS, then I can see the case tipping in favour of the "it's just a tool, just like how you can't sue Microsoft because someone wrote a ransom letter in Word" angle.
I can also see how a court might conclude that, since OpenAI's product is producing libellous / defamatory material about a 3rd party who has no relationship with OpenAI, then OpenAI has an obligation to remedy that in some yet-to-be-determined fashion.
OpenAI might not be the publisher of the material in question, which could be a point in their favour. Or they may indeed be the publisher, if their tool is spitting the false statements out to anyone in the general public who prompts it with a particular name. I can write all the defamatory material I want in private; it only becomes a legal problem if I publish it for someone else to read. The person putting it out in public bears the brunt of the legal responsibility.
The "I just want monetary damages" as opposed to some kind of injunction or call for specific performance might be another point in OpenAI's favour. It makes it look to a judge & jury like Mr. Walters is just after cash, rather than actually protecting a reputation.
But I don't see a legally sound way to a "nobody's responsible, everything is fine, shrug, let's all go home" conclusion.
It's going to be an interesting case.
Last edited:
Upvote
11
(14
/
-3)
Upvote
11
(11
/
0)