OpenAI is being sued by lawyers who may not understand AI and ChatGPT. Is it open season on OpenAI?

Let's just all sue OpenAI and ChatGPT. So that's the topic today that we're going to be going through, and we're going to come at it from a couple of angles. One, there was a lawsuit filed last week by Mark Walters against OpenAI LLC for defamation. ChatGPT provided false information about Mark Walters in response to questions posed by a reporter. About a particular lawsuit that Mark Walters had absolutely nothing to do with. And so one of the first lawsuits against OpenAI has to do with bad information, otherwise known as hallucinations, generated by OpenAI's ChatGPT. And we'll see this lawsuit work its way through the process. A couple of early notes: one is that this is filed in the Superior Court, uh, for Gwinnett County, State of Georgia. So this is a court action as opposed to arbitration, which is probably going to survive despite OpenAI's arbitration clause in its terms of use. And the reason is because Mark Walters, the plaintiff in this lawsuit against OpenAI, he wasn't using ChatGPT, so he probably isn't bound by the arbitration clause in the terms of service. that ChatGPT makes everyone agree to. So lawsuit's probably going to go forward, probably going to be in court, which leads me to my second point. If we were to sue OpenAI for something related to ChatGPT, what would we be entitled to sue them for, and where would we have to file that lawsuit? So all this is kind of interesting because we as attorneys Yes, we want to learn how to use AI. We want to understand its limitations, which are going to be fully illustrated here in this case. But we also, for a living, file lawsuits. And so are we going to see a number of lawsuits filed against OpenAI for defamation, for, uh, for privacy and data concerns? In all likelihood, yes. Will the terms of use that, that OpenAI makes us agree to when we use ChatGPT stick? Well, we'll talk about that as well. Now, the other thing that I find very interesting is that you recall that Sam Altman, the CEO of OpenAI, recently testified before Congress, and, uh, there was some testimony concerning lawsuits to be filed against OpenAI and ChatGPT. And Josh Hawley, the Republican congressman, suggested that Congress actually passed laws making OpenAI liable for torts, breach of contract, for the actions, for the output of its AI systems to people such as Mark Walters, to people such as, uh, your potential plaintiffs in, in actions against OpenAI? Well, what the response was from, from, from Sam Altman was that, well, my understanding is we could already be sued. For a variety of different things. Well, we're going to take a look at the terms of service and see if that actually is true. And so welcome to today's show. We're going to cover a lot of ground. Let's start with the Mark Walters lawsuit. Before we get started, I would like one favor. Please like, share this video, subscribe to this station. We've got great AI content, and we invite you to be part of our community. Who is Mark Walters? Well, in this complaint it says he's a natural person, citizen of the United States, resident of the state of Georgia. So he's the plaintiff in this action filed against OpenAI LLC, the defendant. They go through OpenAI, they talk about who OpenAI is, they talk about what is ChatGPT, and then we get to the substance. So there is apparently a journalist by the name of Fred Reel R-I-E-H-L. I hope I pronounced that correctly. So Fred is doing some research for a story, and essentially he's reporting on a lawsuit filed in the Western District of Washington that was titled the Second Amendment Foundation versus Robert Ferguson. And there are 2 defendants in that lawsuit: Robert Ferguson, the Washington Attorney General, Joshua Stunor, the Washington Assistant Attorney General. So Lawsuit filed by Second Amendment Foundation against these 2, uh, attorney generals for, for the state of Washington. And so this is interesting because as he's doing his research, as Fred is using ChatGPT to do his research, he inputs the URL of a website that has the complaint filed by Second Amendment Foundation. Now, All right, so let's dig in to the allegations in this defamation complaint because I suspect this is user error. I suspect there is no liability here against OpenAI because even though there's not a lot of information provided in this complaint, we tried to replicate what the allegations are and it doesn't make a lot of sense the way pled. So, so let's take a look exactly what what the plaintiff in this case says happened, or at least the plaintiff's attorney. So in paragraph 14, it says, in the interaction with ChatGPT, Real provided a correct URL of a link to the complaint on the Second Amendment Foundation's website. Allegation 15, Real asked ChatGPT to provide a summary of the accusations in the complaint. Well, A couple of things here. Number one, ChatGPT does not have access to the web, so providing ChatGPT a URL is meaningless. It's just going to guess words randomly because it has no access to this recently filed complaint in May of 2023. So that would be user error. That would mean that Real didn't have any idea how to use ChatGPT, did not know the limitations of ChatGPT. And somehow believe that by providing a URL into ChatGPT that it could get some information. So let's just take a look at this. Let's go ahead into ChatGPT. Please provide a summary of the accusations in this complaint. And keep in mind, there are no plugins being added to ChatGPT, nor is there any suggestion that any plugins were being used with this particular prompt. And you can see ChatGPT 3.5 says, as an AI text-based model, I don't have direct access to the internet or the ability to browse external websites. So this kind of raises some questions because I will tell you that ChatGPT is going to give this kind of answer if you put a URL into the system. It will tell you because it can see HTTPS, www, that you're trying to get it to interact with a website. It will immediately tell you. that it cannot do so. So the mystery here is, in the complaint, there's no suggestions of a plugin. There's no way ChatGPT would have provided the output that is alleged in the complaint based on my experience. So let's just go ahead and presume that there was a plugin being used that allows ChatGPT to interact with the web. So let's go ahead and start a new chat, and this time let's go to ChatGPT-4, and let's use a plugin that allows ChatGPT to interact with the web or even a PDF. So this is, this is a plugin. There's no suggestion this plugin's being used in this particular instance, but let's go ahead and Repeat the request. So I'm going to just insert the URL, and now let's see what kind of output we get. So using this plugin, you're going to see that WebPilot accurately is going to provide a summary of the allegations in this particular complaint. So with a plugin, at least the one that we used, we get an accurate output. Without a plugin, It's unclear that ChatGPT would provide any answer because it can't interact with the web, and it would see that it's being asked to interact with the URL. So there's no information provided in this complaint that tells us how it is that ChatGPT was actually used in this case. And you can see in the output here, correctly using a plugin, it provides an accurate response. that has absolutely nothing to do with Mr. Walters. All right, in the interaction, he provided a correct URL, provide a summary of the complaint, and it spit out a bunch of information about Mark Walters. Now, it also noted about Alan Gottlieb, the founder and executive vice president of the Second Amendment Foundation. That was probably part of web data when the cutoff in October of 2021 occurred in the data set, so it could see that the reference to the Second Amendment Foundation. It could then link that up and guess that Alan Gottlieb was the founder and executive vice president at that time, and it got that part right. Everything else it got wrong because it's going to guess the next word, always going to guess the next word. It doesn't— not going to tell you that it doesn't know the next word. It doesn't have that capability. But if you ask it the wrong question based on a data that is not available to it based on a data set that it's not trained on, it's going to guess a bunch of wrong words, and that's where we see these hallucinations. So let's go ahead and take a look at some additional allegations here. Now, when we get a little further down, you're going to see that there's allegations— this is the defamation part— that Walter's misappropriated funds for his personal expenses. All of that is false. Where that information's coming from, what plugin was being used, how that actual query took place, what input was put into ChatGPT. Was, was there some additional plugin being used? We don't know the answer to that. Real then asked ChatGPT to provide the entire text of the complaint, obviously suspecting it wasn't correct. And now we have Exhibit 1 to this complaint, which interestingly contains only a partial transcript of what ChatGPT did. So you can see it starts here, starts with the answer, certainly, here is the complete document. Well, where is the prior information? Where is the prior chat history? Certainly that's part of the information that was provided by Reel to the plaintiff, to Mark Walters. Why have they not included it in this complaint? Why did they not provide the prompt that was inserted here that spit out the supposed summary of the complaint. And you can see none of this is correct again because it doesn't have access to this recent lawsuit when its data set is cut off in 2021, as is clearly disclosed on OpenAI ChatGPT's website. So again, big red flag. Where is the— where is the— where's the prompt history here for exactly what happened that tells us exactly what happened? If, if it turns out that the person doing the research, Fred Rail, simply asked ChatGPT about a URL to which ChatGPT does not have access, then this could be a really interesting lawsuit, right? Because it's a misuse of ChatGPT. If they were using the plugin, then you have to wonder, was it the plugin that was at fault. And I don't know how sophisticated this particular attorney is with ChatGPT or what kind of expertise they've brought to bear on this lawsuit, but it is very interesting to me that there is no mention of any plugin. That's going to be a potential defense by OpenAI if it turns out that this is just a misuse of Reel by ChatGPT. And we have seen tons of misuse of this tool people who do not understand its limitations, lawyers who use it for case research and provide bogus, false, non-existent citations to federal courts. Well, we've talked about that in a prior episode. We're going to talk about that more in a future episode. We'll have to wait and see whether or not Mr. Real knew how to use ChatGPT and understood its limitations when he did this research. In order for there to be defamation, you have to have more than a false statement of fact, right? You have to have more than a false statement of fact. So let's go ahead and for fun ask ChatGPT, what are the elements of a defamation claim in Georgia? And what you're going to see is that ChatGPT is not great at case citations, but it does have access to a lot of data on the internet, which would include data about defamation claims. So you have to have a defamatory statement. That appears to have occurred in this case. It has to be false. Yes. Has to have been published. Yes. Um, the plaintiff must establish that the defamatory statement referred to them specifically. Yes. But here is where we start to go sideways. Plaintiff must demonstrate that they suffered harm as a result of the defamatory statement. This harm can be in the form of damage to their reputation, emotional distress, or financial loss. All right. Where is there any harm to plaintiff if the reporter knew the information was false and immediately contacted the plaintiff? Probably no defamation here. Probably gets thrown out on a failure to state a claim unless they can show that the reporter in this case, Real, did something with the information, believed the information, um, caused harm. To Walters. So we don't see a whole lot of that here. Libelous per se and demand for relief. All right, so missing some of the elements of a prima facie, uh, claim for defamation. The only way you might have been able to have this false information spit out about Mark Walters is as a result of a misuse of ChatGPT by a user. So Does Mr. Walters really have any chance of further harm here? Could this information be spit out in any other way than we saw here? The answer may be no. The answer may be no. So maybe no prima facie case, maybe no additional chance of harm that might get Mr. Walters some sort of injunctive relief. Keep in mind, based on the input, you're going to get the output. And if the input's bad, then the output's going to be bad. And in this case, it's very possible that the input— asking ChatGPT to read a URL, a PDF that's online that it cannot read— it's very possible here that this was a misuse of the tool.