New Lawsuit Against Bing Based on Allegedly AI-Hallucinated Libelous Statements

When people search for Jeffery Battle in Bing, they get the following (at least sometimes; this is the output of a search that I ran Tuesday):

But it turns out that this combines facts about two separate people with similar names: (1) Jeffery Battle, who is indeed apparently a veteran, businessman, and adjunct professor, and (2) Jeffrey Leon Battle, who was convicted of trying to join the Taliban shortly after 9/11. The two have nothing in common other than their similar names. The Aerospace Professor did not plead guilty to seditious conspiracy.

And this Bing output doesn't just list the facts about each of the Battles separately, the way that search engine results have long listed separate pages separately. Rather, it expressly connects the two, with the "However, Battle was sentenced …" transition, which conveys the message that all the facts are about one person. And to my knowledge, this connection was entirely made up out of whole cloth by Bing's summarization feature (which is apparently based on ChatGPT); I know of no other site that actually makes any such connection (which I stress again is an entirely factually unfounded connection).

Battle is now suing Microsoft for libel over this, in Battle v. Microsoft (D. Md.) (filed Friday). He's representing himself, and the Complaint is flawed in various ways. But, if the case is properly framed, he may well have a serious argument. That is especially so if he can substantiate his allegations that he had informed Microsoft of the problem and it didn't promptly fix it.

In particular, I doubt Microsoft would have 47 U.S.C. § 230 immunity. As I discuss in more detail in my Large Libel Models? Liability for AI Output draft, § 230 states that, "No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider" (emphasis added). "[I]nformation content provider" is defined to cover "any person or entity that is responsible, in whole or in part, for the creation or development of information provided through the Internet or any other interactive computer service" (emphasis added). But this lawsuit aims to treat Bing as a publisher or speaker of information provided by itself.

After all, the allegedly libelous material here isn't simply what's borrowed from other sites (the accurate material about Jeffery Battle and the accurate material about Jeffrey Leon Battle). Rather, it's the combination of the material on the two Battles, in one paragraph, linked with "However, Battle …." Courts have held that § 230 doesn't immunize defendants who "materially contribut[e] to [the] alleged unlawfulness" of online content. And the allegation here is that Bing did materially contribute to the libelous nature of the content.

By way of analogy, note that human web site operators are protected by § 230 when they manually quote material provided by other online content providers (see Batzel v. Smith (9th Cir. 2003)). If I just quote something from a web site about Joe Schmoe the professor, I'm generally protected by § 230; likewise if I quote something from a web site about Joe Schmoe the criminal. But if I combine the two, in a way that falsely conveys that the two are the same person, I don't think I'd be immune from liability, since the libel would stem from what I added myself (the juxtaposition of the two items, together with words that make them falsely seem like they're about the same Schmoe). Likewise when the combination is done by Bing (or done by Bing's business partners at OpenAI and then published by Bing, though it's an interesting question whether the ultimate liability would be on Bing, on OpenAI, or on both).

To be sure, there are still many other elements that would have to be shown for Battle to prevail: For instance, Battle would likely have to show at least negligence on Bing's part, and perhaps even knowledge or recklessness as to the falsehood. A key question in this case would be whether his having informed Bing of the error would suffice, if his allegations about having informed Bing are correct. I discuss those issues and many others in considerable detail in Large Libel Models? Liability for AI Output, which I plan on updating before it's published (soon) to also mention this case as an illustration.

But in any event, I wanted to pass this along, since this is to my knowledge only the second lawsuit over libel-by-AI, after Walters v. OpenAI. I will also update this with any response I receive from Microsoft, to whom I sent a media query about the case.

Comments

[Flatulus]

I know this is new ground, but negligence or recklessness as to what? Constructing the AI in the first place? Collecting the training data? Running the learning algorithm? Posting the output without human review?

[Mr. Bumble]

All of the above?

[Flatulus]

In the alternative? Or does the plaintiff have to prove negligence/recklessness as to all?

[Mr. Bumble]

Given the complexity and novelty of this the first thing he needs to do is forget proceeding pro se and get some good legal representation.

[Michael P]

Ultimately, whatever overall process led to publication of the material. As long as Microsoft and its agents performed all of the relevant conduct, the intermediate steps are not legally relevant to whether Microsoft is liable. (So, in terms of your follow-up question, roughly "in the alternative" -- but also inclusive of any other steps that Microsoft took.)

[ReaderY]

Negligence and recklessness in posting material that an ordinary person would realize is likely false.

The technology Microsoft chose to use in doing so is irrelevant. Microsoft’s decision to use AI rather than some other method to create its answer to a query to it no more diminishes or affects Microsoft’s responsiblity for its answer, has no more relevance, than whether a typist uses an electric rather than a manual typewriter.

In the electric typewriter hypothetical, the typist can’t argue that the use of an electric circuit rather than direct manual force somehow makes what was written the typewriter’s responsibility and not the typists’. If an electric typewriter with memory accidentally retypes something unexpected and the typist didn’t check, that’s the typist’s problem.

The use of a more complicated electric circuit changes nothing. Microsoft, the typist, is responsible for the contents of the answer it typed to the query given it. The technology it used to help it type that answer has no relevance to its responsibility.

[Ridgeway]

The other part of the equation is the process MS put in place to receive and process complaints about, and corrections of, inaccurate or libelous information. My guess, based on interactions w Big Tech in similar arenas, is that such process was non-existent or extremely inaccessible.

[ReaderY]

Microsoft’s promises would appear irrelevant in this case. Microsoft never promised anything to Mr. Battle. He is claiming Microsoft libeled him based on work it published about him to someone else at the other person’s request. He is not claiming there is any contract between Microsoft and him, let alone any consideration for such a contract, that would make anything Microsoft promised or said legally relevant or bunding against him. Microsoft can’t simply unilaterally waive non-customers’ legal rights by posting a notice.

[ReaderY]

Forget typewriters. Microsoft has the same responsibility it would have if a Microsoft employee had hand-written the answer and sent it by carrier pigeon. The technology involved has nothing to do with it.

[ObviouslyNotSpam]

It's almost as if Microsoft had intended to create a "random libel machine" and set it free in the most litigious place on Earth.

[Longtobefree]

"What happens on the web, stays on the web (forever)".

What hasn't happened on the web, still stays on the web.

News flash folks, you can't trust the web; including this comment.

[John F. Carr]

"It was not called 'the net of a million lies' for nothing." – A Fire Upon the Deep, Vernor Vinge

[Nige]

People still think stuff stays on the internet forever?

[Voize]

Not quite forever, the Internet Archive has only been operating for 27 years.

[John F. Carr]

This is a mistake a human could make.

[Pereours]

And AI can make it more quickly and thoroughly.

[Rossami]

Very true. And a human could be equally vulnerable to a libel accusation for publishing that same mistake - especially if they refused to correct it after being notified of the error.

[Bored Lawyer]

Possible, but seems to me less likely. If I were researching a person of a certain name, and came up with the same information, I would at least suspect there might be two people of the same name, and dig deeper. Like look for a photo, or other information about who is the veteran/businessman and who is the convict. The two parts of the story do not seem to fit. Possible that an honorable man went bad, or it's possible there are two different people with the same name.

[Dr. Ed 2]

No John -- a human would have seen the logical inconsistency here.

Is being a *honorably* discharged veteran with adjunct status at a major aviation college consistent with a conviction for seditious conspiracy and levying war against the US? In other words, do you think that someone with *those* convictions would continue to be employed at a security-conscious institution?

And a human would have checked to see if the dates of birth match, or at least they were the same approximate age.

[John F. Carr]

Several veterans have been convicted recently on January 6 charges. He could have been one of them. A careful human would not have made the mistake. A careless human could have. A careless human would not have fabricated some of the other potentially defamatory content discussed on this blog.

[Bored Lawyer]

And a "careless human" who did the same thing could be found liable for defamation.

[Dr. Ed 2]

"But if I combine the two, in a way that falsely conveys that the two are the same person,"

As you could copyright your combination, that makes you the author of it, and hence liable as an *author*.

As I understand it, while the text of SCOTUS decisions is in the public domain, West is able to copyright it's layout and key notes and hence it's books containing the SCOTUS decisions.There's a ruling on that somewhere.

[Bored Lawyer]

Copyright law and defamation law are not the same.
Defamation depends on publication, not authorship. If you combine two things and publish them, and that falsely puts someone in a bad light, then you can be liable for defamation.

[Jonathan Affleck]

Is a pro se non-attorney trying to represent an LLC? Won't this case be immediately thrown out?

[ReaderY]

The complaint states that the plaintiff is an “individual”, and also states the defendant is an “organization.”

He appears to have mentioned Battle LLC just as part of his address. Maybe he has an office address separate from his home and wants to use it. I think the complaint states that he is suing in his individual capacity clearly enough to survive a challenge.

[Roger S]

A search for Jeffrey Battle returned info on people with that name. What's the problem? Bing should have no liability for this.

I suggest that the professor put a note on his own web page that he is not the terrorist of the same name. Then add a note to the Wikipedia page on Jeffrey Battle that the article is about the terrorist, not the professor. He had not done either of these things. Then maybe Bing/ChapGPT will figure out that they are two different people. Otherwise, how is the bot supposed to know?

[Eugene Volokh]

The problem is that it returned information that purported to be on a person with that name. It took the information about the aeronautics Jeffery Battle, and then added the "However," followed by "Battle" and then information about the terrorist Jeffrey Battle; this made it seem to a reasonable reader like the entire paragraph was about one person.

Say I wrote that paragraph as part of my own blog post, combining the information about the two Battles the same way. Would we say "This Volokh blog post contained info on people with that name. What's the problem? Volokh should have no liability for this"? And if we think I should be liable for merging information about an upstanding citizen and a terrorist, and making it look like it's all about one person (the upstanding citizen), why shouldn't Bing or OpenAI or both (whoever one views as responsible here)? There might well be answers to that question -- but I think one would have to offer something more than just that Bing "returned info on people with that name."

[Roger S]

If I searched all of your blog posts, I would not be surprised if I found an example of you pulling a fact about someone from a Wikipedia article without checking it. Maybe the fact was false, or maybe it was true about someone else. That is essentially what Bing Chat did. Okay, you are careful than an experimental bot that responds in less than a second, but how can anyone expect a bot to distinguish between two people of the same name?

I say the professor had his chance to point out the distinction, both on his own web page and Wikipedia, and he chose not to do it. I don't see how he can blame anyone else for not pointing out the distinction, when he did not do it.

[ReaderY]

Microsoft did nothing of the sort. Microsoft took two different articles about two completely different people and presented them as information about a single person, the plaintiff.

The plaintiff has a good case that that’s libel.

[Roger S]

Did Google say they were different people? Did Wikipedia? Did the professor's own web site? No. They all leave the reader to figure it out.

[ReaderY]

And by saying nothing itself - merely passing on information from other sources - Google stayed within Section 230’s safe harbor. But by conducting its own original research and writing its own original research report that SAID they were the same person, Microsoft left Section 230’s safe harbor and is responsible.

[Toranth]

If you perform a Google search for "Jeffery Battle" right now, the first three results are:
1) Player page for a North Alabama football player
2) Wikipedia page about the terrorist
3) Linked-in page for "The Associate Professor".
Each listing on Google's result page includes some text from the linked page, clearly identifying each person as a football player, a terrorist, and a CEO. It does nothing to indicate these are not the same person.

Are Bing Chat's results actually different from Google's? Or is it a matter of degree, rather than kind?

[Roger S]

Good point. A simple Google search might similarly seem like libel, if the reader naively assumes that all the Jeffrey Battles are the same. Bing Chat just combined some info into a paragraph.

[Bored Lawyer]

"A simple Google search might similarly seem like libel, if the reader naively assumes that all the Jeffrey Battles are the same."

But that is not a reasonable reading of a Google search result. Most people know that if you do a search, you may get results about different people or things. The results are listed as separate links to separate websites, without any suggestion that each listing is related to the other. Libel claims are evaluated as to how they would be reasonably read:

"Where a plaintiff alleges that statements are false and defamatory, the legal question for the court on a motion to dismiss is whether the contested statements are reasonably susceptible of a defamatory connotation. In making this determination, the court must give the disputed language a fair reading in the context of the publication as a whole." Armstrong v. Simon & Schuster, Inc., 85 N.Y.2d 373, 380, 649 N.E.2d 825, 829 (1995)

So I don't think that would be a good libel claim. Here, in contrast, they were combined in a single article.

[Toranth]

But that relies on the assumption that a 'reasonable person' searching Google for information about a person would not assume that links (and sample information) returned are about that person.
While I'd like to agree, there are far too many cases of people looking up a name and then citing information without any further attempt to verify the results. You see it any time a random person makes the news.

I agree with your point about Bing Chat creating a single 'article' - especially when it switched from using the full name to just using "Battle" - is a good one. However it also supplied reference links for each claim, which doesn't seem too different than what Google does.
So, I agree there is certainly a difference in degree - Bing Chat did more. But is it enough to declare is a difference in kind? A list of search results for a name that cite facts about different people with that name vs a summary of the facts from that same list of search results seem not too different.

[ReaderY]

The difference is that Google stayed within Section 230 by merely summarizing each article separately. It simply transmitted content created by others. But Microsoft wrote its own article representing its own original research. It created its own new content. And that took it outside Section 230, making it liable for what that content said.

Driving on the right and left sides of the road appear to an outsider to be almost identical acts. Just looking at the road, there isn’t any obvious reason why the two sides should be considered different. Yet one is considered legal and safe, while the other is illegal and dangerous. And anyone who knows how to drive can tell the difference between the two. This may be similar.

[Roger S]

You have a funny idea about original research, if you think Bing is doing original research. It crawls the web in a manner not much different from Google, but packages the results differently.

[Toranth]

Can you explain to me why Google would be protected by Section 230, but Microsoft's Bing Chat would not be?
Google's sample for each linked result page is custom selected no less than the summary provided by Bing Chat for the same links. Is there actually a clear definition of how to determine what is "new content", or is this a case of soft lines and interpretation?

And actually, I missed it the first time, but: Both Bing Chat and Google fail, because the two people do not have identical names! One is Jeffery Battle (CEO), the other is Jeffrey Battle (terrorist). Does this incorrect association increase or decrease any potential liability?

[ReaderY]

The difference that whereas Google displayed 3 different web pages and summarized each, Microsoft instead provided its own original-researcj meta-summary that combined information from multiple pages. Microsoft’s choice of language (However…”) permitted. no reasonable interpretation other than that its choice of language summarized a single person, the person described at the beginning of its summary, the plaintiff.

[Roger S]

Microsoft provided its original research? It licensed a bot from OpenAI and lumped some facts together in less than a second.

You say "permitted no reasonable interpretation". Now you must be joking. Any reasonable person would understand that he is chatting with a bot with only the most superficial knowledge of people named Jeffery Battle.

[ReaderY]

The law is concerned solely with how a reasonable person would interpret what the words Microsoft wrote mean. The technology Microsoft used to create its content – quill pen, typewriter, perhaps even a computer – is completely irrelevant to that question. Once placed out in the universe, their origin makes no differences to the consequences.

Is Boeing not responsible for the 737 Max crashes because it was software, and not people, that piloted the plane? The idea that things done by software are somehow nobody’s responsibilty means it isn’t.

The law doesn’t care HOW the plane was piloted, only that it was piloted negligently. And for good reason. The passengers didn’t care.

Nor should Mr. Battle here. As long as he can prove damage to his reputation, that damage is as real to hom as the 737 Max crash was to its passengers and their families.

If a Boeing or a Microsoft wants to put cool new things out for people to play around with, they have to get the kinks worked out before the cool new things go out to the public in ways that can get people hurt. When ordinary people get hurt, they AREN’T so understanding and accepting. There’s no “Wow! way cool technology, so it’s OK, and just stuff it Mr. Jones” exception to the law.

[Roger S]

Following your analogy, the 737Max would sometimes go into a sudden dangerous dive. Yes, I think it makes a difference whether a person initiated the dive, or a software program. Boeing would have been in a lot more trouble if a Boeing employee was personally doing it.

As for harm, people died in 737Max crashes. It is not clear that anyone was harmed by the Bing search, or that the reasonable user was misled.

[ReaderY]

But Boeing employees DID personally do it in the Boeing case. They personally hand-programmed the software that caused the crash. They can’t even claim it was created by AI.

If the plane was on manual control, but the pilot was in the bathroom and wasn’t looking, you could argue the pilot didn’t “initiate” the dive. But that wouldn’t mean the pilot is off the hook. The dive would most assuredly be considered the pilot’s fault. Trying to say “but the plane just did it on his own” wouldn’t work.

By creating software that took over the controls from the pilot, Boeing assumed responsibility for the consequences. Boeing was as responsible for how its 737 Max behaved as pilot in the bathroom and not at his post would be.

Same with Microsoft and its libel. Just as a human pilot responsible for controlling the plane is can’t blame somebody else for his negligent failure to properly control it, Boeing and Microsoft can’t dosclaim responsibility for not being around to supervise what the law makes it their job to to take care of.

[Voize]

I don't follow why §230 would not shield Microsoft here if, as described, the defamatory linking commentary was created by ChatGPT rather than Bing.

[Bruce Hayden]

Respondeat Superior, or something similar. ChatGPT Was acting as and agent of Microsoft, which, as the principal, is typically responsible for the torts of its agents and employees.

[ReaderY]

Principles and agents concern people, natural or artificial. “ChatGPT” is neither. It’s a thing, like a quill pen or a carrier pigeon. It‘s simply a tool Microsoft chose to use to help it create its content. Whether Microsoft chose to use a stone tablet, a quill pen, or a computer program is entirely irrelevant.

[ReaderY]

The defective commentary was created by Microsoft. It doesn’t matter whether Microsoft chose to use a stone tablet, a quill pen, or even a computer in doing so. It doesn’t matter if it chooses to call its quill pen Simon, or name its computer program ChatGPT. That doesn’t make “Simon” or “ChatGPT” responsible. Similarly, it doesn’t matter if the stone tablet, the quill pen, or even the computer program is owned by Microsoft or was rented or lent for the occassion. The technology it chooses to use to help it create its content is of no relevance or concern to its legal responsibility.

[Voize]

ChatGPT isn't Microsoft's program, it is OpenAI's program.

[Bruce Hayden]

But Microsoft apparently hired them, which makes them responsible for their torts (I.e. defamation).

[Voize]

Microsoft would have to exert significant control over what OpenAI includes in the content it generates before it could incur vicarious liability for that content. Simply having them as a supplier of a service is not enough.

[phoqueue]

Let's assume that's that case. Microsoft has still done a lot of republication.

[Peter Gerdes]

I’m skeptical of your argument that a program like released as merely an AI playground can commit libel - I don’t think the authors of such programs can be said to have truly asserted those sentences anymore than apple asserts the sentence it’s predictive text suggests – but this is clearly different.

Microsoft presented the information in it’s search results clearly intending for people to accept it as at least a true summary of what was available in reputable online media. Yet, they synthesized a statement that wasn’t in any such media. So yah, this is the easy case.

[ReaderY]

Is it your position that Tesla can never be responsible for crashes when its car is on autopilot because its engineers can’t be said to be responsible for how the autopilot program drives the car? Boeing isn’t responsible for the 737 Supermax crash because its software, not it, controlled the plane and caused it to stall?

If your argument is correct, they aren’t responsible. If they are responsible, your argument isn’t correct.

[Davy C]

"Battle v. Microsoft" just strikes me as funny. Especially when the guy is pro se.

As for the case, it's hard to know how courts will rule, but I think he should win. I don't think we can just have a freestanding AI exception to defamation where they can say anything. Will this kill AI? No, but maybe it can kill search engines posting unreliable summaries based on AI.

[ReaderY]

That critical word “however” doesn’t appear in any of the sources. That’s Microsoft’s original writing. Microsoft, and only Microsoft, is resposible for it. And that word makes everything that follows a statement about the Plaintiff. A libelous statement.

The law doesn’t give a shit about how searches are conducted. The law cares about what words mean. And who wrote them. So in this case, the “packaging” - the words Microsoft wrote - is everything to the law, however trivial it may seem to the mere technician.

[Archibald Tuttle]

EV said "If I just quote" along with several other I s. This equates a machine with a human actor which I believe is wrong.

I know that the suit is against the owners of the machine, but the tort is about the human action of libel based on the action of the machine rather than its creators. The creator of the Ouija board is not responsible for the words that may emerge from it. I know EV rejected the Ouija board analogy before, but I think it applies. EV's stated reason is that AI output is more credible than Ouija output, focusing on the damage to the plaintiff rather than the culpability of the defendant. That's weak. Ouija and AI are both machines, and there must be a continuum of intermediate machines that enjoy intermediate credibility.

If we ever get to the point where machines are held to be subject to the law, then machines must also have rights. Think of the implications. That idea must be resisted.

[Bored Lawyer]

No, the tort of libel, as I point out above, is based on publication of false information, not creation of a message (with which copyright law may be concerned). If someone else creates a false message, and I publish it, under traditional defamation law, I am liable. (Section 230 does limit this for online publication).

Microsoft published an allegedly defamatory article about Mr. Battle. How it was created is interesting only to the extent of what a reasonable reader would take away from it. Sure, if it came with a disclaimer, "This is AI Generated and may be complete horse puckey," then yes, there is likely no defamation liability. But that does not seem to be the facts here.

[Archibald Tuttle]

That doesn't jibe with EV's comments in the post. Specifically, the use of I I I again and again. That is equating a person with a machine.

If I just quote something from a web site about Joe Schmoe the professor, I'm generally protected by § 230; likewise if I quote something from a web site about Joe Schmoe the criminal. But if I combine the two, in a way that falsely conveys that the two are the same person, I don't think I'd be immune from liability, since the libel would stem from what I added myself (the juxtaposition of the two items, together with words that make them falsely seem like they're about the same Schmoe).

[ReaderY]

How it was created does matter for Section 230 purposes. Google takes individual web site links and posts them verbatim. That stays within Section 230. Microsoft wrote its own summary tying information from the different sites together into a single paragraph, with language indicating everything in the paragraph is about the individual identified at the beginning of it. That takes it outside Section 230’s limitation of liability.

[lewis]

I agree with all of you