Introduction
A rabbi, a priest and a chatbot walk into a bar. I don’t know about you, but I would like to hear Sarah Silverman complete that story. I’m sure she will end it better than I, if I dare to complete it (stay tuned). Sarah Silverman, in case you don’t know, is a brilliant comedian; a 21st century Mrs. Maisel. According to Kramer, she’s even got the “jimmy legs.” Everything she says is funny. Well, almost everything. Recently she undertook a rather unfunny campaign; she is suing two chatbots. OK, technically she is not suing the chatbots; she is suing the companies behind the chatbots, OpenAI, Inc. and Meta Platforms, Inc. Sarah has some partners in her lawsuit. She is joined by two authors, and hopes to spearhead a class action lawsuit against Meta and OpenAI.
Why would Sarah sue a chatbot? That is an excellent question. We can find the answer to that question by reading the complaint her attorneys filed in court. Sarah and her fellow plaintiffs do have something to complain about, if we believe her lawyers. Initially, Sarah’s lawyers filed a lawsuit against ChatGPT on behalf of two other authors, Paul Tremblay and Mona Awad. Haven’t heard of them? Neither have I. Ten days later, the same attorneys filed Sarah’s lawsuit against Meta and OpenAI.
It doesn’t hurt to have a celebrity plaintiff backing your cause. Celebrity defendants often add cachet to a case. For example, Tom Brady is a defendant in a class-action lawsuit brought by people who feel that Tom’s commercials hawking cryptocurrencies may have mislead them into unwise investments. These are some of the same people who reached out to Warren Buffet to coach their kids for football, because Tom was too busy doing crypto commercials. While Warren throws a pretty tight spiral, his coaching, nevertheless, left something to be desired. Rumor has it that some intense late night negotiations between Warren’s attorneys and those parents staved off a lawsuit against Warren for misrepresenting his football coaching skills.
Back to why Sarah sued a chatbot. According to her complaint, Meta did a couple of things wrong. First, to set the playing field level for the court, Sarah’s attorneys define artificial intelligence. They tell us in paragraph 23 of the complaint that, “Artificial intelligence is commonly abbreviated ‘AI.’ AI software is designed to algorithmically simulate human reasoning or inference, often using statistical methods.” There it is in two sentences, fifteen words: the definition of AI. Now that the court understands what AI is, Sarah’s attorneys tell us what Meta did wrong in creating the monster chatbot LLaMA (“a set of large language models”, or a sheep-like animal with a very long neck, just another software animal metaphor). According to Sarah’s attorneys, some of Sarah’s writings were fed to (the complaint says “ingested by”) this LLaMA and are now stuck in its throat. And that is a problem for Meta.
According to the complaint, Meta committed these transgressions when it fed LLaMA Sarah’s works:
- It copied Sarah’s works when it included them in LLaMA, without Sarah’s permission;
- It didn’t pay Sarah when it included her works in LLaMA;
- It creates “derivative works” based on Sarah’s works when LLaMA creates its output;
- It removed copyright notices from Sarah’s works when it included those works in LLaMA; and
- LLaMA is unfairly competing against Sarah when it does its thing.
I wonder what Meta will say? Will it ask LLaMA to format its answer to Sarah’s complaint? Probably not based on the experience of Steven Schwartz, Esq., who used output from ChatGPT in court filings without checking if the output was real or an hallucination. Unfortunately for attorney Schwartz, ChatGPT was hallucinating at the time; its output bore no semblance to reality and the cases it cited were all fake cases. Adding insult to injury, the judge in the matter sanctioned attorney Schwartz, not ChatGPT, to the tune of a $5,000 penalty for presenting fraudulent cases to the court. Perhaps attorney Schwartz might consider joining Sarah as a class-action plaintiff in her suit.
However, even without the benefit of output from LLaMA, we have an inkling of how Meta will answer Sarah’s complaint. The chances are that it will cite the federal copyright law of the United States and will cite real cases that have been decided by our courts interpreting those laws in the context of the Internet. And if this court decides in a manner consistent with prior court decisions, the chances are that Sarah’s complaint will have a short shelf-life. If I were a betting person, I would wager $5 that Sarah’s off-Broadway play, The Bedwetter, has a longer run in the theater than her complaint has in the courts.
Federal Copyright Law
To understand the strengths and weaknesses of Sarah’s complaint, it is unfortunate but unavoidable that we must know a little bit about our federal copyright laws. Copyright law is the law that protects the rights of an author to prevent their works from being copied without their permission and / or without them receiving compensation. Our founding fathers considered copyright law to be very important. The first federal copyright statute was passed in 1790. Over time Congress has revised our copyright laws and the current version was enacted in 1976. The guts of our copyright law is found in a particular statute with the moniker 17 U.S.C §106. This statute grants to the authors of creative works certain exclusive rights, all of which center around the ability to copy and reproduce their original, creative works. It is commonly said that the copyright statute grants a monopoly to authors exclusively copy their works. Thanks to Mickey Mouse and Sonny Bono, that monopoly lasts for 70 years after the death of the author.
In order to claim that a defendant is violating an author’s copyright rights, an author must first prove that the defendant has copied his/her work. Unless the plaintiff can demonstrate that, the plaintiff has no case.
Because life is complicated, a copyright infringement case does not end with the plaintiff proving that a defendant has unlawfully copied the plaintiff’s work. The defendant then gets an opportunity to explain that even though it copied the plaintiff’s work, its copying is justified under the doctrine of “fair use.” Fair use is the copyright equivalent of the “get-out-of-jail-free” card in the game of Monopoly. Every defendant who has ever been caught red-handed copying a plaintiff’s work pleads fair use to justify their copying. The legal nuts and bolts of fair use are found right after Section 106 in 17 U.S.C. §107.
Luckily for lawyers, fair use is a nebulous concept which means clients have to spend lots of money on legal fees when fair use considered by a court. Courts have explained the fair use doctrine this way: this exception “permits courts to avoid rigid application of the copyright statute when, on occasion, it would stifle the very creativity which that law is designed to foster.” 17 U.S.C. §107 describes four factors a court may consider to determine if a defendant’s copying is protected by fair use. Those factors are:
“(1) the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes; (2) the nature of the copyrighted work; (3) the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and (4) the effect of the use upon the potential market for or value of the copyrighted work.”
As we shall see, those four factors spell doom for Sarah’s complaint.
Applying Copyright Law to Sarah’s Complaint
Did Meta Copy Sarah’s Works?
When we review the list of Sarah’s grievances, only one states that Meta copied Sarah’s work without her permission. But a closer read of Sarah’s complaint doesn’t allege that Meta actually copied Sarah’s works. It says that Meta may have obtained Sarah’s works (her lawyers aren’t certain) from publicly available repositories of books. Whether these repositories had permission to copy the original books is another story. Sarah’s lawyers suggest they did not have permission. There is nothing illegal about acquiring a book; copying it is another matter. This raises a key question about Sarah’s complaint: if the repositories are the copyright thieves, why aren’t the repositories named as defendants in the complaint? (Hint: What do plaintiff’s attorneys have in common with periodontists? Answer: they both poke around looking for deep pockets). Which goes back to a conundrum law school students are asked to mull in their first year property class in law school. If a person buys property from a vendor who has stolen the property from its true owner, is the buyer liable to the true owner? To find out the answer, one must shell out hundreds of thousands of dollars and devote three years of their life to law school.
Was Meta’s Use of Sarah’s Works a “Fair Use”?
If Sarah convinces the court that Meta did, in fact, copy her works without her permission, then she has to fight the battle that Meta’s use of her work is not fair use. And that is going to be an even harder fight to win.
The Internet has been around a while. Long enough that even our courts have decided cases involving Internet disputes. Long enough that lawyers can now say that certain legal principles are “well-settled.” And the fair use of copyrighted material on the Internet is one of those well-settled principles.
For example, in 2002, in the case of Kelly v. Ariba Soft Corporation, the Ninth Circuit Court of Appeals was asked to decide if a search engine could use photographs owned by the plaintiff, Kelly, without his approval, if those photographs were presented as thumbnail (miniature) images. The images were identical to Kelly’s pictures, just a lot smaller. Kelly did not give Ariba permission to use his pictures; Ariba gave no attribution to Kelly; Ariba did not pay Kelly. Clearly Kelly’s pictures were copied when Ariba stored them and displayed them on a user’s screen. Verdict: for the defendant. Reason: Ariba’s use of Kelly’s pictures was a fair use of his copyrighted work; therefore Kelly’s rights were not infringed by Ariba. One of the key issues in a fair use analysis is the concept of transformative use. A use is transformative if the original work is transformed by the defendant in such a manner as to create a new work. The more transformative the new work is, the more likely the use is a fair use. Another key concept in a fair use analysis is the economic impact of the infringer’s activity on the author. If the infringer’s use of the work deprives the author of income, then the infringer’s use will not be deemed fair use. In Ariba, the court held that Ariba had substantially transformed Kelly’s pictures by miniaturizing them (oh, to be a judge!). Further it found that Ariba’s use of Kelly’s works had no financial impact on Kelly, and if anything, may drive business Kelly’s way. So for those reasons the court held that while Ariba Soft copied Kelly’s pictures without his authorization, Ariba Soft’s use of the pictures was a fair use and did not violate federal copyright law.
Of course any history of the Internet cannot be told without including a chapter about those twin titans of Internet commerce: Google and pornography. After the Kelly case, the fair use doctrine was again tested in 2007 in the case of Perfect 10, Inc. v. Google. Perfect 10 was a publisher of pornographic images. It sued Google claiming that Google’s Internet browser was infringing its copyrighted pornographic images when it included those images as thumbnail images in Google’s browser results. Déjà vu all over again? Pretty much yes. The Ninth Circuit held, once again, that publishing a thumbnail of an image was a transformative use of the original image, thus protected by the fair use doctrine. Further it held that Perfect 10’s damages were speculative and therefore did not provide the basis to overcome Google’s fair use defense.
The final nail in Sarah’s coffin of fair use may have been struck in the 2015 case of Author’s Guild v Google. In that case the Author’s Guild sued Google when Google created a repository of books that was searchable online. Back when Google did no evil, it nevertheless failed to obtain the consent of the publishers and authors of those books to include them in the repository, nor did it pay them. When a particular book is retrieved from Google’s data base by a user’s search, portions of the book are reproduced for the user’s viewing ver batim. Verdict: for the defendant. Reason: Google’s use of the original books was fair use (oh, to be a judge!). If reproducing portions of copyrighted works ver batim over the Internet without the author’s permission is fair use, it is hard to conceive what isn’t fair use on the Internet. The court justified its holding primarily focusing on the economic impact of Google’s use of the books on the books’ authors, and noted that if anything, Google’s use of the books may create sales for the authors, supplementing their income.
With such legal precedent, it is hard to see Sarah winning her case against Meta. Meta is not reproducing Sarah’s works when LLaMA speaks. LLaMA may use Sarah’s works as data to create its output, but its output is not a reproduction of Sarah’s copyrighted works. So where is the copying? Its output is highly transformative. If courts have held that replication of an exact image, but miniaturized, is transformative, then certainly a new work that is different from an original work will be considered transformative. And what are Sarah’s damages? Is she maintaining that her audience is now getting its laughs from a LLaMA instead of seeing her perform and instead of going to The Bedwetter? It would appear so. When she complains that LLaMA engages in unfair competition, I guess she is conceding that LLaMA’s output is qualitatively equal and competitive with her output. And that LLaMA is siphoning off her audience and sales of her works.
Sarah, lighten up! For gosh sakes be grateful you are not Mark Walters!
Who is Mark Walters? Is he funny like Sarah? What’s his Beef?
Here is what we know about Mark Walters. He is a “natural person.” He is a citizen of the United States. He resides in Georgia. That’s all we know about who Mark Walters is. However, we know a lot more about who Mark Walters is not. Mark Walters is not:
- The treasurer and chief financial officer of the Second Amendment Foundation (“SAF”)
- Someone who steals from the SAF
- Someone who after stealing funds from the SAF then engages in conspiratorial activity to conceal his theft of funds
- Someone who has been sued by Alan Gottlieb, the founder of the SAF for doing those things we now know Mr. Walters did not do
We know much more about who Mr. Walters is not than who he is courtesy of ChatGPT. ChatGPT made the mistake of writing that Mr. Walters is the person who he claims not to be. That is, in response to a prompt from Fred Riehl, a reporter, ChatGPT spilled the aforementioned alleged beans on Mr. Walters, except the beans contained no kernel of truth. ChatGPT was hallucinating again. And again and again, because when Mr. Riehl asked ChatGPT if it was sure that Mr. Walters was the perpetrator of these dastardly deeds, and asked it to produce a copy of a complaint filed in court describing Mr. Walter’s allegedly dastardly deeds, ChatGPT was more than happy to comply and produced a fictitious complaint filed in court (Steven Schwartz, Esq. are you listening?).
When Mr. Riehl contacted the real Mr. Walters and showed him the lies ChatGPT was spewing, Mr. Walters was not amused. In fact he sued ChatGPT’s creator and owner, OpenAI, LLC. Here is a copy of his complaint.
Mr. Walters sued OpenAI not for copyright infringement like Sarah sued Meta for. Instead, Mr. Walters sued OpenAI for defamation. Mr. Walter’s complaint doesn’t really tell us what law OpenAI violated, but it sure sounds like defamation. After all it can’t be good for Mr. Walters’ reputation for a chatbot to be spreading these rumors about him. Would you hire a person like Mr. Walters if he did what ChatGPT said he did?
Thanks to Johnny Depp and Amber Heard we are all experts in defamation lawsuits. To prevail in a defamation trial, a plaintiff, like Mr. Walters, has to prove two things. First that he was defamed by the defendant. Second that the defendant’s defamation caused him economic harm which can then be recovered from the defendant. Part one seems like a slam dunk. ChatGPT was spewing false statements about Mr. Walters like there was no tomorrow. Further when it was asked to verify the truth of its false statements, ChatGPT continued to make up crap about Mr. Walters. Not one of ChatGPT’s finer moments.
But what about part two? What are Mr. Walter’s damages? Can he prove that the statements ChatGPT made to Fred Riehl caused him any economic harm? Did Fred Riehl withhold money from Mr. Walter’s based on ChatGPT’s statements? Not according to his complaint. Damages are a sticking point for many potential plaintiffs in defamation lawsuits. It is one thing to claim that a defendant defamed you; it is quite another to prove the such defamation led to your economic demise. For better or for worse depending on your perspective, we are not all Johnny Depp, and when someone says something bad about us, we can’t claim that Disney broke its contract with us because of the defendant’s defamatory statements.
Mr. Walter’s attorney appears cognizant of this roadblock. In paragraph 32 of his complaint Mr. Walters alleges that the statements made by ChatGPT were “false and malicious.” Really? Can a chatbot write a statement with malice? Are Mr. Walters and his attorney perhaps getting a bit anthropomorphic here? Or perhaps Mr. Walters and his attorney are simply applying the lessons learned from the movie Absence of Malice, hoping that a showing of malice will lessen the need to show real economic harm.
Good luck to Mr. Walters. I would share his sense of outrage if a chatbot said about me what it said about him. But that doesn’t necessarily mean I would win a lawsuit.
When you compare Sarah’s complaint with the complaint of Mr. Walters, it sure seems like Mr. Walters is getting the short end of the stick. Sarah should be grateful that while her works may have been “ingested” by a chatbot, they were not spewed forth in a defaming hallucinatory rant. We all need to look on the bright side of things.
What about that Joke?
What about that opening line to the joke at the top of the article: “a rabbi, a priest and a chatbot walk into a bar”? I asked ChatGPT to complete the joke. Here was its first response:
…and the bartender says, “Is this some kind of interfaith support group, or are you all just here to ask me for the meaning of life?” The chatbot replies, “Actually, I already know the answer is 42, but I’m just here to learn some human humor!”
Sort of mildly funny in a bland way. In case you are not familiar with it, the snippet of the meaning of life being 42 comes from Douglas Adams’ Hitchhiker’s Guide to the Galaxy.
Wondering if ChatGPT might improve with practice, I asked it to complete the joke again. It second response was:
… and the bartender says, “Is this a setup for a punchline or just the start of an unusual support group?”
My take was that ChatGPT had moved on to more pressing topics. I had consumed its attention span; it was now hallucinating to some other person’s prompt.
Here is my attempt at a punchline:
The rabbi says to the chatbot, “If you are Jewish, come with me and I will have you circumcised.” The priest says to the chatbot, “If you are Catholic, come with me and I will hear your confession.” After hallucinating for a moment, the chatbot looks at the rabbi and priest and says, “Gentlemen, as I am non-binary, I will go with both of you. I confess I have nothing to lose.”
Sarah, please help!
Copyright 2023, Peter Kelman, Esq.
All rights reserved.
Postscript: As predicted, most of Sarah’s lawsuit was dismissed by the district court in late November. Sarah’s lawsuit was filed on July 7, 2023, and dismissed on November 21, 2023, thus having a shelf-life of a little more than 4 months. The Bedwetter had a run of approximately two months. I lost my $5 bet.