Ask HN: Will Microsoft be liable for defamatory Bing statements?

39 points by graeme ↗ HN
Got access to new Bing, tried some chats. It got a few things factually wrong. Then I asked it about a minor business leader I knew who had a scandal.

The ai replied to the effect that:

“Yes, so and so had a scandal. They even had complaints filed with the better business bureau and the california bureau of consumer protection”

First part was accurate and sourced. Second part was invented out of whole cloth. The ai admitted as such when I pressed it.

But, isn’t that a false statement of fact and legally actionable? And it is Microsoft saying it.

49 comments

[ 2.8 ms ] story [ 103 ms ] thread
No one knows.

People thought this about general internet postings then 230 saved the companies. Governments seem a lot less likely to “help” the companies today though.

It’ll likely be decided similarly to copyrights: with messy and impactful trials. If the prompter owns a copyright, then you the searcher may end up responsible for the AIs statements (probs not though).

Section 230 was about hosting user comments though. A company is still liable for what they publish.

Microsoft, for example, can’t publish an article on Microsoft.com defaming people and cite section 230.

However, they can link to defamatory sites and excerpt snippets from them, for that isn’t considered publishing.

Bing chat seems an awful lot more like publishing.

The point is that pre-230, no one knew who was liable for comments, and 230 clarified. The tech industry needs a law detailing who is liable, the way 230 clarified.

The AI-Generated art world seems to want the prompter to be the "artist", but if you prompt bing for search... are you the "creator"? It seems unlikely. On the other hand, Microsoft is "simply" using an algorithm to rehash the existing web.. just like search results page did before, so this is no different.

ChatGPT is the publisher by the fact it has restrictions on your prompts. "As an AI model I can not provide..." etc.
Moderation of content does not imply publishing.
Someone is the publisher. It can’t be nobody
Does Word moderate what I type into it? No because it's a tool. ChatGPT publishes
The web sites in search results were always liable for defamation though. And Microsoft/Google could have been had they rewritten snippets to falsely suggest a page had defamatory information (they don’t)
> The point is that pre-230, no one knew who was liable for comments, and 230 clarified.

That's not really true; pre-230 there waa a pretty clear idea of how, absent additional legislation, traditional publisher and distributor liability would apply to online services, and what incentives that created with regard to offering, moderating, and engaging with such services. Section 230 was very much intended to alter the economic incentives, not just to “clarify” liability.

> Section 230 was about hosting user comments though.

No, 230 is not about “hosting”, its about (what would, absent Section 230, legally be) publishing.

> Microsoft, for example, can’t publish an article on Microsoft.com defaming people and cite section 230.

It absolutely can, if some other “information content provider” submitted the information in it, because they are not legally it’s publisher in that case; that is the central point of § 230. [0]

The question here seems to be what happens when the way interactive computer services relay information from other internet content providers is by farther from relaying distinct externally provided content units that may be mixed, ordered, promoted, demoted, or suppressed by an algorithm, and instead synthesizing new content units from externally provided and actively sourced information.

But the 230 safe harbor doesn’t focus on content units (by any name), it focuses on information, so as long as the defamation comea from information provided by another “information content provider”, no matter what algorithmic means and content structure goes into presenting it, it would seem squarely within the safe harbor.

When defamation arises not from a particular third-party source material, but only from the algorithmic synthesis, things get trickier, but if its all third partt information (even separate pieces of it that the algorithm presents together) it still seems within the safe harbor.

When its (in whole or in part) information Microsoft actively sourced or that the MS-operated algorithm created ex nihilo (hallucination) that makes it defamation, though, that would seem to be Microsoft’s liability.

[0] 47 USC § 230(c)(1): 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.

how about getting song lyrics out of bing or ChatGPT? those are covered under copyright, aren't they?
Google already does this so I assume it's in line with the law. If they get the lyrics wrong, maybe that's under fair use?
Google pays for the right to display lyrics, but if they display them incorrectly that might be an issue. Especially if the wrong version somehow goes against the morals of the artist.
Aha, I didn't know they were paying for it.
Chat with Bing has found to be given explicit instructions to not give out book passages and music lyrics. This came out the other day when someone was about to get it to say its code name (Sydney) along with it default prompt set.
> People thought this about general internet postings then 230 saved the companies.

Not just the companies, also other users whose actions (e.g., reply with quote) would be “publication” under the law without Section 230.

Pretty much every EULA I've ever seen exempts the developer from liability from how the software is used. Is there reason to believe GPT is any different?
GPT is irrelevant, the issue is somebody publishing the content. If I (or Microsoft) publish the claim "Joe Bloggs is a pedophile", and he isn't, it's defamation regardless of how the statement was generated. And yes, there's ample legal precedent that this applies to search engines too.
Yeah, this is different.

Search is just snippets of other peoples websites. Search companies would like to see themselves as neutral. A common carrier, like the power companies.

GPT is something different. Its original speech. Not just a snippet of somebody else's speech.

Playing devil’s advocate: I guess the legal argument here could be that you wouldn’t be able to come after Logitech for making a keyboard used to publish lies. If microsoft could manage to convince a court that you could only get defamatory content out of bing by asking it to defame someone, I could see how that argument could work.
The analogy would be of Logitech would’ve released a keyboard that types for you and that keyboard then decided to type incorrect and possibly liable statements and not only that but that keyboard also decided to publish it.

For Microsoft to be liable all you need to prove is that when asking a question without any guiding prompts the output contains libel.

And this is the problem because Microsoft isn’t just pointing you to incorrect or libel information but they actively creating and promoting it.

(comment deleted)
It's defamation if Joe Bloggs wants to sue. He can certainly try, but first he has to prove that Bing was actively defaming him. Alright, let's see:

- Was this content published or promoted to other people? (No)

- Was it served in accordance with US law and the license agreement (Yes)

Proving either of those wrong would take a village. Assuming you get past that, you then have to prove that it was done with malicious or defamatory itent. How do you intend on doing that? You could subpeona OpenAI for their training data, but that only works if the training data was manipulated with dishonest intent.

Sure, there are AI-specific unknowns in the legal world. Entertaining this is no more relevant than suing Microsoft for a Swastika you saw on an online Minecraft world though - you knew what you were getting into when you signed up.

That’s just for the US though. I’m in Canada and you absolutely don’t need to prove malicious intent for a civil suit. And new Bing is being rolled out here.

  Was this content published or promoted to other people? (No)
It's not difficult to run the search on another computer.

  you then have to prove that it was done with malicious
  or defamatory itent
Microsoft is showing a reckless disregard for the truth, which is enough. [1] Microsoft has a history with this. Microsoft shut down within a single day a previous chat bot, Tay. [2] Tay was replaced with Zo, which was eventually shut-down. [3]

They are disregarding their own prior experiences with these sorts of AI chat bots. They are also disregarding that chatgpt was designed to work as a text generator that has zero regard for the truth. [4]

Anyone who recalls history or who works in the field knows that Microsoft is ignoring the expected and actual results of Bing's AI.

Plus, that's just for the US. The rules are different elsewhere.

[1] https://www.nolo.com/legal-encyclopedia/defamation-libel-sla...

[2] https://en.wikipedia.org/wiki/Tay_(bot)

[3] https://en.wikipedia.org/wiki/Zo_(bot)

[4] https://towardsdatascience.com/understanding-gpt-3-in-5-minu...

EULA are pretty irrelevant in court…

Microsoft can no longer claim to be just a search engine that connects between a user and information published on the web since their ai search assistant generates bespoke answers to user questions.

There likely is some liability buried there that can be successfully argued in court in favor of the plaintiff.

EULAs only can bind the people who are parties to them.

If say Bing were to tell you something false and defamatory about me, like say that I write Luca fanfiction that ships Alberto and Ercole, their EULA might protect them from liability for the psychological trauma being exposed to that pairing caused you (Alberto and Luca is clearly the OTP for that fandom...) because the EULA is between you and them, but it would not protect them from liability for defaming me because an agreement between you and them cannot waive my rights.

Here is a one-liner:

    while(1) console.log('abcdefghijklmnopqrstuvwxyz '[Math.floor(Math.random()*27)])
It will create all the same defamatory statements as Bing's algorithm does. Am I liable for those because I gave you this tool?
you're not presenting them to the public when they're asking something relating to that fact though
So I am liable when I put it on a GitHub page and add a button?

    <script>
        function reply() {
            while(1) {
                c = Math.floor(Math.random()*27));
                console.log('abcdefghijklmnopqrstuvwxyz '[c])
            }
        }
    <script>
    Your question <input></input>
    <button onclick=reply()>Your Answer</button>
    Disclaimer: The answer is AI generated and might be wrong.
your tool would not be thought of as a reasonable source of information by anyone, including the court, who are not mindless automatons but understand context & your disclaimer is a lot more visible than it is on bing etc.
That’s going to involve a series of court decisions that have yet to come.

Microsoft is a huge target and filing a suit while this has a lot of buzz is a big PR opportunity, so those court cases will start coming soon enough.

Clearly, Microsoft felt confident they could navigate whatever legal headaches and must believe that seizing this opportunity for Bing to outmaneuver Google was more valuable than whatever that might cost. What we can be sure they didn’t do is go in blind.

The end of capitalism and the beginning of fascism is when an invention hasn't even been released to the public yet and people are asking "What if somebody doesn't like it?"

Fuck AI though ask an amish farmer how to update your antiquated jquery to angular and just realize how much happier than you he is.

I've never had finer furniture than that made from the Amish community near where I grew up. The craftsmanship and attention to detail is unreal. Also, this particular community made some of the best cheese I've ever tasted.

There's something to be admired about their way of life.

Which area is that from? I've been looking for some quality furniture.
I’m in the public and I used it and it was pretty shocking how quickly and confidently it asserted a false and damaging fact about a non-public figure.

How would you feel about BingChat being convinced that “LinkLink’s real name” has been fired from multiple jobs for fraud and it kept repeating it until pressed? That’s basically what it did in the convo I had

Sure man it can say what it wants its just BingChat, anyone who's opinion I actually respect can't have their worldview altered by a collection of tensor GPU's.

increasingly as influence and MKultra esque government and social media programs continue to rollout, it is becoming blatantly obvious who is easily influenced and who is not. Which is exhausting, it shows just how few people really have any autonomy at all.

>innocuous thing is facism

>disparaging comments about the amish

Why?

">disparaging comments about the amish" >How much happier he is than you
Let me look into my crystal ball...
> First part was accurate and sourced. Second part was invented out of whole cloth. The ai admitted as such when I pressed it.

Has anybody had success in getting ChatGPT to admit that true things it said was wrong?

(comment deleted)
> Ask HN: Will Microsoft be liable for defamatory Bing statements?

Companies are never held liable.

No, the same way that they dodge responsibility when you Google for Rufus and they show you an ad for a fake Rufus with a trojan.