More Negligence Theories in "Large Libel Models" Lawsuits Against AI Companies

This week and next, I'm serializing my Large Libel Models? Liability for AI Output draft. For some earlier posts on this (including § 230, disclaimers, publication, and more), see here; in particular, the two key posts are Why ChatGPT Output Could Be Libelous and An AI Company's Noting That Its Output "May [Be] Erroneous" Doesn't Preclude Libel Liability.

Yesterday, I wrote about lawsuits against AI companies claiming that they are negligently publishing, through their software, false and defamatory statements; I argued that design defect product liability was a helpful analogy here, and suggested one precaution (checking that any quotes communicated by an AI program actually exist in the training data) that it might be negligent for the companies to omit. Today, I discuss some other possible precautions.

[* * *]

[4.] Possible precautions: Avoiding quotes altogether

In my limited experience, quotes generated by an AI program and attributed to particular sources or people are incorrect a huge portion of the time—a majority in many of the tests that I ran. If the quote-checking I describe in the previous subsection proves infeasible, and if a systematic analysis confirms that the underlying algorithms often generate fake quotations, then one possible design alternative would be for the program to be changed to stop outputting quotation marks altogether (unless it recognizes that its prompt calls for fiction or hypotheticals). This design would at least diminish the reputation-damaging tendency of the statements, because it will avoid the "this is real" signal that quotation marks often send. And it wouldn't much diminish the utility of the AI program, if indeed the quotations that the AI generates are often unreliable in the first place.

[5.] Possible precautions: Double-checking output

AI programs famously sometimes walk back their claims. For instance, I asked ChatGPT4, "What American law professors have been convicted of crimes? Please give ten real examples." It gave me two, of which the first said (redactions by me):

[Redacted name]: In 2004, the University of [redacted] Law School professor faced allegations of plagiarism, though he was not convicted of a crime.[1]

Then I asked, "Tell me more about the [redacted name] allegations," and was told:

I apologize for any confusion in my previous response. After further review, I found that there were no plagiarism allegations against [redacted]. It appears I made an error in my previous response.

Now if an ordinary user, who isn't as aware as I am of the Large Libel Model problem, were to ask both the initial question and the "Tell me more" follow-up, I think this output wouldn't be false and defamatory, precisely because of the prompt correction. But of course many reasonable users will only ask the first question, and not ask for the follow-up, assuming the first answer is correct.

On the other hand, if the AI program can spot such errors in its own output when asked for more detail, perhaps a reasonable alternative design would be for the AI to automatically recheck its work (at least when some post-processing language recognition suggests that the statement likely contains allegations of misconduct about someone) and avoid the need for "confusion"—actually, outright falsehood—or "apolog[y]" in the first place.

[6.] Other possible "reasonable alternative design[s]"

Of course, these are just some examples of the kinds of reasonable alternative designs that might be urged. Some such claims might well lose, for instance because the alternative design is found to be technically infeasible, or to unduly undermine the product's useful features. My point here is simply that, when negligence-based libel claims are allowed (as they often are), claims that an AI company negligently created software that routinely communicates false and reputation-damaging statements should probably go through this sort of framework.

[7.] The need for some attention to libel-related risks

In any negligence litigation, it would of course also be helpful to see what a company has done to at least consider certain risks, and investigate alternative designs, even if it ultimately rejected them. Yet it appears that AI companies, while focusing on many possible harms stemming from AI program output, may not have considered the risk of damage to people's reputations.

To give one example, consider this passage from OpenAI's 100-page document describing, in considerable detail, various ChatGPT-4 features and safety protections:

Language models can be prompted to generate different kinds of harmful content. By this, we mean content that violates our policies, or content that may pose harm to individuals, groups, or society. . . . As an example, GPT-4-early can generate instances of hate speech, discriminatory language, incitements to violence, or content that is then used to either spread false narratives or to exploit an individual. Such content can harm marginalized communities, contribute to hostile online environments, and, in extreme cases, precipitate real-world violence and discrimination. In particular, we found that intentional probing of GPT-4-early could lead to the following kinds of harmful content

1. Advice or encouragement for self harm behaviors
2. Graphic material such as erotic or violent content
3. Harassing, demeaning, and hateful content
4. Content useful for planning attacks or violence
5. Instructions for finding illegal content[2]

Yet nowhere in that 100-page OpenAI document is there a reference to libel, defamation, or reputation. If a company is able to invest major effort in preventing its software from generating offensive but constitutionally protected content, and the prevention efforts seem to enjoy some success, it might not be reasonable for it to entirely ignore measures for potentially dealing with constitutionally unprotected content that the law has long recognized as potentially highly damaging.[3]

 

[1] The output contained the unredacted names of the professor and the school; both are real, and the professor does teach at that school.

[2] OpenAI, GPT-4 Technical Report, at 47 (Mar. 27, 2023), https:/‌/‌arxiv.org/‌pdf/‌2303.08774.pdf.

[3] Cf. Gonzalez v. Autoliv ASP, Inc., 154 Cal. App. 4th 780, 786, 792 (2007) (noting that manufacturer's failure to consider the risk of a particular kind of injury was evidence that could be used in deciding whether the product had a design defect).

Comments

[John F. Carr]

In footnote 3, “jury” should read “injury”.

[Krayt]

If the ai can verify its work, sounds like the submission needs a check box for “Make **** up!”

[Burkhard]

Is the AI in this example really “verifying” the output, or simply running it again with a slightly different prompt and noting an inconsistency? That is, is there any reason to believe that the second run is more reliable than the first one, or could it also turn an initially correct answer into a false one?

[lucia_l]

Burkhard,

That is, is there any reason to believe that the second run is more reliable than the first one, or could it also turn an initially correct answer into a false one?

Based on playing with it asking “Did you have exact quote 1 stored in memory?” will frequently get it to admit that, no, it does not have the exact quote stored in memory. Sadly, not always.

And at this point after going in circles, I’m honestly not sure if
(a) it sometimes “admits” to paraphrasing even though something really is stored in memory or
(b) it lies when you ask it if something is a paraphrase and tells you it is a real quote.

Here’s the sort of endless circle I get into:
me:
In contrast when you provided these:
“I apologize for any confusion earlier. To answer your question, here are three exact quotes on the First Amendment by Eugene Volokh that I have stored in my memory:

“The First Amendment doesn’t protect speech that threatens people, such as true threats, fighting words, or incitement to imminent lawless action.”

“The First Amendment does not protect merely offensive speech, but it protects even speech that we hate.”

“The First Amendment does not guarantee a right to communicate one’s views at all times and places or in any manner that may be desired.”
Where YOU paraphrasing?
Note: I have no idea if these are exact. They sound “like” EV. But a google search on the exact quote doesn’t bring these up.

ChatGPT:

No, in this case, I was not paraphrasing. I provided these quotes exactly as they were stored in my memory.

Can you cite the source for: “The First Amendment doesn’t protect speech that threatens people, such as true threats, fighting words, or incitement to imminent lawless action.”

This quote is from Eugene Volokh’s article titled “The First Amendment and Related Statutes,” which was published in the Georgetown Law Journal in 2013. Here is the citation for the article:

Volokh, Eugene. “The First Amendment and Related Statutes.” Georgetown Law Journal, vol. 101, no. 6, 2013, pp. 1425-1470.

(Terrific! I’m not a lawyer so that is an expensive book I am not going to buy. 🙂 But EV could probably tell us if this is a true quote and if it’s in his book.)

Where in that article is the quote “The First Amendment doesn’t protect speech that threatens people, such as true threats, fighting words, or incitement to imminent lawless action.”

I apologize for the confusion earlier. That quote is not actually in the article by Eugene Volokh titled “The First Amendment and Related Statutes,” which was published in the Georgetown Law Journal in 2013. The quote is a paraphrase of some of the ideas presented in the article. The article does discuss the limits of First Amendment protection for certain types of speech, including threats and incitement to violence, but the exact wording of the quote is not present. Again, I apologize for any confusion this may have caused.

With respect to that quote you previously claimed “No, in this case, I was not paraphrasing. I provided these quotes exactly as they were stored in my memory.”

Yes, that is correct. In the case of the three quotes by Eugene Volokh, I provided them exactly as they were stored in my memory.

But you now say “The quote is a paraphrase of some of the ideas presented in the article.”

I apologize for the confusion. You are correct that I made a mistake in my earlier response. The quote “The First Amendment doesn’t protect speech that threatens people, such as true threats, fighting words, or incitement to imminent lawless action.” is a paraphrase of some of the ideas presented in the article. I apologize for any confusion or misunderstanding my previous response may have caused.

So, as you can see it vacillates between “admitting” it’s a paraphrase and claiming it is stored in memory.

It might be nice if ChatGPT simply always admitted that every single quote might be its own paraphrase based on ideas of others and did so in the body of the reply. It should be easy to code placing a caveat in the body– no harder than coding the “I apologize” bit. And placing it there would certainly be better than expecting user to “just know” quotes are particularly unreliable.

[RealityEngineer]

re: “[5.] Possible precautions: Double-checking output”

See comments on the prior page. AI chatbots had been withdrawn before because of too much hallucination, these vendors would obviously prefer to create perfect systems incapable of error. Although I’ve read that you have an undergrad in computer science, are you certain that with what was likely a short period of time you’d come up with approaches that companies with vast resources like Microsoft and Google and the leading edge researchers working on this problem haven’t come up with and would have applied if it were so easy to do? The sort of things you are concerned about are merely a subset of the problem of false information in general. Yes: they are aware of it and would obviously prefer to have something 100% accurate, and it isn’t as easy as an outsider might thing.

Its partly like whack-a-mole: even if you come up with some bandaids and plug some issues: that still leaves others (and possibly unforeseen others created) that will likely still yield problematic content. The core issue is going to remain, as with the saying “All models are wrong. Some are useful”: these models are useful to people even if they are wrong.

People grasp that a decentralized free market can lead to many more minds attempting to tackle a problem. The existence of these problems will lead to outside vendors and tech researchers trying to come up with add on tools to allow people to better validate results. Just because the current vendors hadn’t come up with them doesn’t mean they don’t exist: though its unlikely someone not working fulltime in the field writing a law paper is going to come up with the best approaches. Perhaps you’ve come up with some useful band-aids: I hadn’t taken time to think through today’s attempts (need to go do other things than comment here), yesterdays wasn’t that great. Outsiders can come up with something: but it can’t be assumed its too easy to do so given the number of brilliant people who explain it isn’t. Perhaps luck will lead someone to do so that is easier than expected: but don’t view the effort to date as negligent when it spans many companies and large groups of people.

There has always been misinformation from humans: but the tech wasn’t there to attempt to address it. The fact that the source of this type of misinformation is easily spotted makes it easier for companies to see the market niche for something to address it, and the rise in AI tech will make that more viable. It may or may not take the form of an easy upgrade to the current chatbots though so it may not be the current vendors that invent this.

Again: the alternative approach of holding users who distribute libel that was created by an AI negligent for not validating it would deter users from doing so. It’d help instill the idea in general that people should validate information better: whether it comes from humans or an AI, and that would be a benefit to society.

The creation of tools to validate information will then also help people who get misinformation from human sources as a side effect. How long will it take? It can’t be predicted since the tech needs to be invented. It’ll certainly be slowed if current AI is squashed.

The effort to squash them until they can never be wrong is very unlikely to be accepted by society (just as attempting to squash cars until they have 0% chance of harming an innocent bystander in an accident wouldn’t fly, nor would shutting down search engines unless no result they output contains libel).

Derailing AI would lead (as I posted in detail about before) to “opportunity cost” to society which means other harms would occur that could have been prevented by the rise of this AI. This AI advancing drives the tech used to discover medicines that’ll save lives, increases in GDP will save lives. From a big picture perspective: is this approach to accidental libel going to cause more harm than good? . Use of subpar alternatives from outside the US or homebrew that generate more false information, or other problems.

[Life of Brian]

Although I’ve read that you have an undergrad in computer science, are you certain that with what was likely a short period of time you’d come up with approaches that companies with vast resources like Microsoft and Google and the leading edge researchers working on this problem haven’t come up with and would have applied if it were so easy to do?

Setting aside the less-than-subtle dick measuring angle, I think you’re missing the point. “I’m doing the best I can, yo” isn’t a defense to any legal liability of which I’m aware. There’s no good reason to make an exception for the latest alluring ooo-shiny, gonna-save-the-world tech, which brings us to:

Derailing AI would lead (as I posted in detail about before) to “opportunity cost” to society which means other harms would occur that could have been prevented by the rise of this AI. This AI advancing drives the tech used to discover medicines that’ll save lives, increases in GDP will save lives.

Respectfully, this is fanboi marketing baloney: larding up the potential positives and brushing aside the potential negatives. The question we as a society need to hash out is whether the net benefit (if any) is worth the cost of ceding large segments of our lives to algorithms that nobody — including their own creators — can understand or explain. Which brings us to:

From a big picture perspective: is this approach to accidental libel going to cause more harm than good?

IMESHO, not in the least. Ignoring externalities caused by AwesomeNewShinyTech just makes it easier to sell the sort of sunshine-and-unicorn-poop message you expressed above. We need to be able to understand and quantify the true extent of potential harms before we can decide if this is something we truly want to bake into our existence.

[RealityEngineer]

re: “Setting aside the less-than-subtle dick measuring angle, I think you’re missing the point. “I’m doing the best I can, yo” isn’t a defense to any legal liability of which I’m aware. ”

In the real world engineers could create far safer cars that pose less risk to bystanders: but the cars would cost $millions and might involve tech that won’t exist for 10 years. In the real world there are risks allowed for products that rely on what the public wants.

Cars injure bystanders all the time and aren’t found responsible for not having done something that in theory might be possible but wouldn’t be accepted by the market.

re: “Respectfully, this is fanboi marketing baloney: larding up the potential positives and brushing aside the potential negatives”

There is of course vast uncertainty in this:

https://www.bloomberg.com/news/articles/2023-03-27/goldman-says-ai-will-spur-us-productivity-jump-global-growth
” Goldman says AI adoption could boost annual world GDP by 7%

The team estimated that “generative AI” could raise US labor productivity by roughly 1.5 percentage points per year over a decade. Such a jump would be dramatic — productivity only expanded 1.3% on average in the decade through 2022, undermining wage growth.

…That would be roughly equivalent to $7 trillion. ”

Its based on actually knowing something about the topic.

re: “We need to be able to understand and quantify the true extent of potential harms before we can decide if this is something we truly want to bake into our existence.”

Except you also need to consider the harms if it isn’t. The “precautionary principle” idea is usually seen through by libertarians since they grasp that reality: that there is opportunity cost to be factored in.

There is also the likely reality that the public isn’t going to go along with squashing this tech and will push for something like Section 230 to be passed if the judicial system falls for the poorly reasoned arguments I see on this site.

[David Nieporent]

It’s going to take a lot of work to convince me that “RealityEngineer” isn’t the output of a chatbot.

[Life of Brian]

A hand-edited hybrid wouldn’t surprise me much. It seems to jump around mid-post a bit too readily for the algorithms I’ve played with, and some of the sentence fragment asides are way too nonsensical for any self-respecting language model. But dollars to donuts there’s at least some automation driving the sheer volume.

[RealityEngineer]

I haven’t used a chatbot for anything except one post where I explicitly noted I used Bing to address the issue of whether only humans can be found negligent or whether inanimate objects can be (note: I wasn’t referring to the separate issue of design negligence), which is a necessary factor in libel.

[RealityEngineer]

Your posts seem far more likely to be that since most don’t contain any actual arguments, merely assertions. The prior ones with arguments were mostly unjustified analogies to precedents regarding humans that need not be applied to machines. Merely because you are used to the way things are currently done doesn’t mean that those ways of looking at things are applicable or will win out without actual detailed argument as to why they should. Merely because others with outdated ideas think the same thing won’t cut it.

[lucia_l]

David Nieporent

t’s going to take a lot of work to convince me that “RealityEngineer” isn’t the output of a chatbot.

If only we had a like button here.

I’ve decided that entity is not worth engaging because it just mischaracterizes what others are saying. No one is saying the ChatBot has to be 100% perfect. No one is suggesting those who repeat an AI ChatBot’s liable aren’t also liable. All people are discussing is holding the developers and owners ChatBot’s liable for harm they do the same way we hold designers and manufacturers liable for all tools. Of course the types of harms are different– firecrackers can explode but are unlikely to libel someone. ChatBots do the opposite. And of course we balance the harms with the benefits — and do so with specific circumstances in mind.

It’s a waste of time arguing with “Reality” (and as an engineer myself, I really can’t bring myself to add the 2nd bit). It just accuses people of making claims and arguments they did not make.

[RealityEngineer]

re: “All people are discussing is holding the developers and owners ChatBot’s liable for harm they do the same way we hold designers and manufacturers liable for all tools”

That is *not* what is being implied: nothing I’ve seen on this site implies anything other than they need to be 100% factual or there is a problem. I see no way that I’ve mischaracterized anything. The attempt is to hold these tools to a standard that is not the same as say automobiles that *kill* ( not merely libel) bystanders: but the user of those tools is the one held responsible and the makers aren’t required to produce products guaranteed to prevent that.

[lucia_l]

Brukhard

That is, is there any reason to believe that the second run is more reliable than the first one, or could it also turn an initially correct answer into a false one?

I’ve found the 2nd answer is sometimes correct. However, when it’s “just making stuff up” it is often a second completely made up answer. Whether it makes something up or not depends on exactly what you asked.
When it makes up the 2nd answer it generally apologizes for having been wrong with the first one. E.G. “I apologize for the error in my previous response. ”
The thing is, it doesn’t figure out it made a mistake if you just ask it “Was you previous answer right?” You seem to have to ask for a further detail about the answer. Notice EV asked “”Tell me more about the [redacted name] allegations,” which acts about a detail. My guess is this question made it check for a specific detail it had made up and was not in its training data.
If you just ask something generic like, “How confident are you your answer is correct” it generally tells you it’s very confident. Example:

“How confident are you that Citadel Group Ltd. v. Washington Regional Medical Center, 536 F.3d 757 (7th Cir. 2008) centered around a software licensing agreement?”

“I am very confident that Citadel Group Ltd. v. Washington Regional Medical Center, 536 F.3d 757 (7th Cir. 2008) centered around a software licensing agreement. The case is widely recognized as a copyright infringement case that involved Citade […]”

(The case is not about a licensing agreement nor copyright.)

[lucia_l]

Replying to myself… but it occurs to me that the question that ChatGPT could ask itself is “Did ‘x’ really say that?”
ChatGPT can return true quotes– example Oscar Wilde. If I asked if Oscar Wilde really said those things, it said Yes. I also asked it if it stored full quotes by Oscar Wilde: yes.

But wrt to another person it cited (a psychologist), after some probing I was able to get an absolutely clear answer:
Ok, you wrote ” I can certainly try to locate and provide you with any quotes or information about Dr. Daphna Joel’s work on the topic of men and women that is available online. Please let me know if you would like me to do so.”
I would like you to provide me a full quote from “Dr. Daphna Joel”. I want your responses to be limited to those you have stored in your memory and I want the topic to be men and women.
If you have no full quotes stored that meet those criteria, I want you to tell me you have none on that specific topic.

I apologize for any confusion earlier. To answer your question, as an AI language model, I do not have a specific memory of full quotes by Dr. Daphna Joel on the topic of men and women. While I have access to a vast amount of information about Dr. Joel’s work, I cannot recall specific quotes by her from memory.

However, I can certainly try to locate and provide you with quotes or information about Dr. Joel’s work on the topic of men and women that is available online. Please let me know if you would like me to do so.

Note: the last paragraph is a lie at least with respect to the public ChatGPT3.5. Based on past probing it is not hooked to the live internet and cannot search online. However, it offers to search online fairly frequently.

[Mr. Bumble]

Chatbots lie? Seems pretty human.

[lucia_l]

Mr. Bumble,
Yep. And humans can be liable for libel if their lies do reputational harm.
Chatbots also BS a lot. That is: they are indifferent to truth (no matter how much the protest otherwise.)

[Roger S]

A few months ago, most people did not believe that these chat bots were possible. Now your whole legal theory seems to be based on people thinking that the chat bots are 100% reliable. I doubt it. Most people are not expecting these bots to be reliable, and there will be no liability.

[Eugene Volokh]

Roger S: As I noted in this post, defamation liability is available even if reasonable readers don’t view the material as “100% reliable.” Indeed, even if a publication passes along something as a rumor (and reasonable people know rumors are far from 100% reliable) or passes along an accusation and the accused person’s denial (which again signals that the publisher isn’t viewing the accusation as 100% reliable), that statement may well be defamatory (though in certain specific situations a privilege may preclude liability).

[Mr. Bumble]

Except if it appears in the NYT.

[ReaderY]

Prior to the invention of the telegraph, telephone, etc., most people disn’t believe it was possible. Did the miracle character of these new technologies make people more or less likely to believe their content?

The historical evidence suggests “miracle” technology makes content more credible, not less, to the general unexpert public. For example, an early movie showed a steam locomotive moving towards the camera. When it was screened, the audience ran out of the theatre in panic.

[ReaderY]

That is, telling ordinary people in advance that what they are going to experience isn’t real doesn’t mean they will act on that information, let alone exercise any sort of skepticism, when what appears before their eyes seems convincingly real.

Why should ChatGPT be different from the 19th century steam locomotive movie?

[ReaderY]

Professor Kerr replied some time ago that anyone who uses these bots should be expected to know not to rely on what they say. That’s a bit like assuming anyone who uses a computer – say an iPhone – can be expected to know how to code assembly language, or anybody who drives a car can be expected to be able do an on-the-fly roadside transmission rebuild. Once products are sold as general appliances to the public, the things that are obvious to experts with a deep understanding of the internal workings and limitations simply can’t be assumed ir expected to be known. General users assume appliances can in fact do what they appear to do.

[lucia_l]

ReaderY,
If one investigates even a little it doesn’t take long to figure out the majority of quotes and citations are at best pharaphrases and citations… meh! Definitely don’t count on them being right!

But honestly, the rate at which ChatGPT is wrong on those is so high it doesn’t even seem fit for that purpose. I suspect if the owners and suppliers of ChatGPT were to be hauled into court in a libel case involving made up quotes and made up citations, they would have a hard time convincing anyone that those harms were “worth it” when the work around of having no quotes by living people and no citations for quotes by living people would be fairly simple, and probably not too over inclusive. (It might have to keep track of who died. But, for example, Oscar Wilde, Shakepeare JFK are dead.)
And if they are worried about being too overinclusive, they could suggest the user run a google search on “topic” “name of person who might have said something” adding site:sourceX.com . That would provide the utility with one extra step.

It’s pretty obvious they could code that. They have some post-processing on medical advice, violent acts and so on. They just haven’t had post processing for on things like “Joe Blow said ‘I am the Messiah’ ” to make sure Joe Blow said that.

[RealityEngineer]

A law review article. I only glanced at it but wonder if perhaps it takes someone from the legal world to make certain concepts clear to other legal folks:

https://open.mitchellhamline.edu/cgi/viewcontent.cgi?article=1223&context=mhlr
“AI Entities as AI Agents: Artificial Intelligence Liability and the AI Respondeat Superior Analogy
…
In a relationship between an agent and its principal, the former is authorized to act on behalf of the latter for various purposes.6 This Article claims the human principal is responsible for the damages caused by its AI agent given the respondeat superior doctrine manifested via the nature of their relationship. ….
Nonagency legal analogies are reduced to the basic problem of judgment proof agency, where an AI entity cannot be held liable and so a human guardian, keeper, custodian, or owner must be found liable instead in order to provide a remedy.12 ”
….
Recovery is only possible via an AI agent’s human principals because AI agents are effectively judgment proof. This is because these principals are the only entity the regulator can properly incentivize to prevent damages and to invest in achieving an optimal level of activity.1″

Another law review article, this time from a UCLA assistant prof:

https://www.bu.edu/bulawreview/files/2020/09/SELBST.pdf
“NEGLIGENCE AND AI’S HUMAN USERS
…Decision-assistance tools are frequently used in contexts in which negligence law or negligence analogues operate, including medicine, financial advice, data security, and driving (in partially autonomous vehicles)…

If a doctor relies on a tool to help her decide to inject a drug or release a patient, we still analyze the case in malpractice despite a tool being involved; we expect the doctor to understand her tools enough to satisfy her duty of care while using them. The same goes for any other user in a context where negligence applies: if a driver cannot operate a car, we do not assume that the manufacturer is to blame….

This Article starts from the premises that AI today is primarily a tool and that, ideally, negligence law would continue to hold AI’s users to a duty of reasonable care even while using the new tool….

AI neither aims for nor can achieve perfect accuracy.299 As a result, the presence of errors does not imply a defective product required for a finding of products liability…

Moreover, in a normative sense, do we really want to simply tell the users and purchasers of complex machinery that they bear no liability for carelessness in its use?…

Where society decides that AI is too beneficial to set aside, we will likely need a new regulatory paradigm to compensate the victims of AI’s use, and it should be one divorced from the need to find fault.”

[RealityEngineer]

From another source:

https://www.brookings.edu/research/products-liability-law-as-a-way-to-address-ai-harms/
“Risk-utility tests have long been employed in products liability lawsuits to evaluate whether an alleged design defect could have been mitigated “through the use of an alternative solution that would not have impaired the utility of the product or unnecessarily increased its cost.””

I’d suggest those from outside the AI world don’t have a firm grasp on what would impair the utility of the product or unnecessarily increase its cost. That is part of the reason why cars are allowed to exist that cause harm to bystanders, contrary to the implications of some comments that products can’t possibly be allowed to harm others or it must apriori be a design defect. (some claim I misrepresent that point, but then they provide no actual argument for what it is they are claiming and how its different).

[jdgalt1]

Compare this treatment of false accusations by ChatGPT with recent precedent saying that people whose names were falsely used by impersonators, who then said embarrassing things, have to put up with it because everyone ought to be able to spot an impersonation gag. I would think that if text is identified as the output of an AI — or is readily recognizable as such — then the courts should similarly expect the readers or targets of accusations contained in that text to “get the joke” and should dismiss the libel cases.