By
Josh TyrangielColumnistFebruary 4, 2024 at 7:00 a.m. EST(Ann Kiernan for The Washington Post) Listen8 minShareComment on this storyCommentAdd to your saved storiesSave
The original “Mothra vs. Godzilla” came out in 1964, which means that even the people who saw it then can’t remember it. So allow me to recap. Mothra — a moth goddess, so fertile and nurturing that her brown speckled-body could be a Goop caftan — lays a giant larvae-filled egg off the coast of Japan. Godzilla — coldblooded stomper of things — wants to smash the egg. A huge battle ensues between the forces of creation and destruction.
WpGet the full experience.Choose your planArrowRightThis is essentially the story the New York Times tells in its lawsuit against OpenAI. The Times (Mothra) is a life force of liberal democracy fighting off a Big Tech destroyer. OpenAI (Godzilla) has already pillaged the Times by using its articles to train ChatGPT without permission. The suit’s inclusion of ChatGPT results that are near word-for-word rip-offs of the New York Times’s stories does indeed make OpenAI look very stompy.
OpenAI rejects this premise entirely. Why, just recently it was negotiating rights fees with the Times! In a blog post written in the tone of a man unsure why his date left the restaurant, the company says, “Our discussions with the New York Times had appeared to be progressing constructively. … We regard the New York Times’ lawsuit to be without merit. Still, we are hopeful for a constructive partnership with the New York Times and respect its long history.”
Advertisement
There are several cases at the intersection of copyright and generative AI — Sarah Silverman vs. Meta, Getty Images vs. Stability AI — but this is the only one that could tempt Don King out of retirement. These are culturally dominant apex corporations, with well-earned reputations for innovation and arrogance. Yet it’s also an underdog story. The Times is worth about $8 billion. OpenAI checks in at about $100 billion, while its largest stakeholder, Microsoft, is worth 16 Jeff Bezoses.
As a journalist who writes about artificial intelligence, NYT vs. OpenAI is the only thing people in my professional life want to talk about. And they all have the same question: How does it end?
Josh Tyrangiel: A new AI predicts when we’ll die. It says even more about how we live.
Elon Musk famously declared that “The most entertaining outcome is the most likely.” This is obviously insane, but it’s solid column-writing advice. So let’s indulge Musk’s Razor and dive a little deeper into the three possibilities: (1) A verdict. (2) Congressional action that cleans up copyright law and obviates the need for a verdict. (3) A settlement.
Advertisement
The most entertaining outcome would clearly be the New York Times taking its case all the way to the Supreme Court where it proves it’s been the victim of such rapacious harm that the justices, emerging in Tom Friedman mustaches, rule unanimously in the Gray Lady’s favor.
Now the entertaining part: The court then agrees with the Times’s demand that OpenAI destroy any chatbot models and training data that use copyrighted Times material. This is known as algorithmic disgorgement — a legal concept to address unlawful gains from algorithmic sneakiness. An ordinary tech company might be forced, at significant cost, to remove infringing material from its product. But generative AI isn’t ordinary tech. It’s created through the combination of hundreds of billions of parameters of information and language. That makes extracting the New York Times from ChatGPT as impossible as removing the eggs from a cake. If subjected to disgorgement, OpenAI could theoretically be forced to eliminate everything it used to train ChatGPT and start over, essentially killing the company. The next day’s Wordle: Pwned.
I don’t think any of this is going to happen. First of all, OpenAI has a better-than-decent fair use argument based on ChatGPT’s ability to transform Times articles into something original. (OpenAI dismisses the Times’s plagiarism claims as a “rare bug.”) Consideration of social benefit is also part of fair use. That’s why news organizations are frequently allowed to use copyrighted materials, or a poetry scholar can use excerpts of a poem. Societally, we’ve decided those are things we want to encourage. You can debate whether ChatGPT use is a social good, but tens of millions of Americans are already using it. Disgorgement, or even an injunction, would harm those people as well as OpenAI. Lastly, the legal process can take years, during which ChatGPT is likely to become even more entrenched. Think of how Uber showed up in cities, nudged out taxis and never asked forgiveness. You don’t have to be righteous to win in court. You just have to be inevitable.
Advertisement
Timely congressional action also seems unlikely. If you need this explained further, I’d like to direct you to The Post’s circulation department, where a staffer has a few questions about why you subscribe.
Share this articleShare
That leaves a settlement, which sounds anticlimactic. But given the parties’ current positions there’s more drama to be found in a negotiation than in slow motion legal combat.
According to several people familiar with his thinking, A.G. Sulzberger, chairman of the New York Times Company and publisher of the New York Times, believes he has been robbed of billions of dollars of intellectual property. He knows that OpenAI has already ingested the Times’s content and can’t take it out. A one-time payment is a nonstarter, because even if OpenAI made a Godfather offer, there’s no guarantee that today’s windfall won’t one day lead to AI swallowing his family business. Having almost perished in the first wave of AI-powered social media algorithms, the New York Times sees a chance to soak a tech company for a massive licensing fee on a long-term contract — billions of dollars in recurring revenue. And if it can’t get its number, the company has a proud history of going to court to defend its principles.
Advertisement
On the other side, OpenAI is at the peak of its “let’s incinerate cash on the way to market dominance” phase. Everyone is giddy with possibility, which makes it a perfect moment to pay the vig and drop a one-time sum on the New York Times. But: No large recurring payments. The last thing OpenAI wants is to become a royalty administrator, constantly arguing over percentages with every little publisher who can claim to have been a part of ChatGPT’s training. Not only is it a hassle and a drain on future profits, it’s unglamorous by Silicon Valley standards.
Since filing suit, the Times is doing what it does best — enjoying its impact in silent sanctimony. (To be fair, it’s also keeping its ethical wall high; none of the Times reporters I’ve spoken with have any clue what management is thinking about the case, much to their annoyance.) Meanwhile OpenAI has been Jay Gatsby at the White House correspondents’ dinner, buying drinks and striking deals with all the other news organizations it can. The value of those deals is in the very low millions, but it’s a great show of magnanimity toward the Fourth Estate, old sport. A perfect way to demonstrate to the public and the courts that OpenAI isn’t the stubborn one.
What might ultimately drive a settlement — and I’d bet one happens in the next few months — is that both sides have a sliver of existential risk. The Times is one the best financial stories there is in the wasteland of American journalism, but AI has brought huge new unknowns. Securing its sustainability now seems smart. OpenAI has already had one near-death experience, and while a disgorgement ruling is hard to imagine, it’s been forced to admit that ChatGPT is addicted to copyright. “It would be impossible to train today’s leading AI models without using copyrighted materials,” the company wrote in a submission to the United Kingdom’s House of Lords.
Advertisement
It’s that last bit that leads me to think OpenAI is more likely to bend. Now’s the time to fix a price for its drug of choice. When the big record labels battled Spotify, they ended up reaching an agreement to license music to the platform while also getting 18 percent of the company. Spotify didn’t want to do it (and OpenAI would never go anywhere near that high) but a music service can’t exactly live without music. Neither side has been thrilled with the relationship, but it’s a form of mutually assured preservation. Would a 5 percent stake in OpenAI, divided proportionally among the largest news organizations and book publishers, and rebalanced once a year to avoid too much accounting, do the trick?
In the movie, Godzilla kills Mothra with his fiery atomic breath, while Mothra’s children wrap Godzilla in silk and get their vengeance when he lizard-shuffles into the ocean and dies. These would indeed be thrilling outcomes. But with so much at stake, and apologies to Musk, the least entertaining outcome is the most likely.
Share Comments
Popular opinions articles
HAND CURATED
- Opinion|Doctors have fatphobia, too — which does serious harm to patientsJanuary 31, 2024Opinion|Doctors have fatphobia, too — which does serious harm to patientsJanuary 31, 2024
- Opinion|I stopped trying to sleep — and found a different self in the nightFebruary 1, 2024Opinion|I stopped trying to sleep — and found a different self in the nightFebruary 1, 2024
- Opinion|The GOP’s blunders take their tollFebruary 1, 2024Opinion|The GOP’s blunders take their tollFebruary 1, 2024
View 3 more stories
The original “Mothra vs. Godzilla” came out in 1964, which means that even the people who saw it then can’t remember it. So allow me to recap. Mothra — a moth goddess, so fertile and nurturing that her brown speckled-body could be a Goop caftan — lays a giant larvae-filled egg off the coast of Japan. Godzilla — coldblooded stomper of things — wants to smash the egg. A huge battle ensues between the forces of creation and destruction.
This is essentially the story the New York Times tells in its lawsuit against OpenAI. The Times (Mothra) is a life force of liberal democracy fighting off a Big Tech destroyer. OpenAI (Godzilla) has already pillaged the Times by using its articles to train ChatGPT without permission. The suit’s inclusion of ChatGPT results that are near word-for-word rip-offs of the New York Times’s stories does indeed make OpenAI look very stompy.
OpenAI rejects this premise entirely. Why, just recently it was negotiating rights fees with the Times! In a blog post written in the tone of a man unsure why his date left the restaurant, the company says, “Our discussions with the New York Times had appeared to be progressing constructively. … We regard the New York Times’ lawsuit to be without merit. Still, we are hopeful for a constructive partnership with the New York Times and respect its long history.”
There are several cases at the intersection of copyright and generative AI — Sarah Silverman vs. Meta, Getty Images vs. Stability AI — but this is the only one that could tempt Don King out of retirement. These are culturally dominant apex corporations, with well-earned reputations for innovation and arrogance. Yet it’s also an underdog story. The Times is worth about $8 billion. OpenAI checks in at about $100 billion, while its largest stakeholder, Microsoft, is worth 16 Jeff Bezoses.
As a journalist who writes about artificial intelligence, NYT vs. OpenAI is the only thing people in my professional life want to talk about. And they all have the same question: How does it end?
Josh Tyrangiel: A new AI predicts when we’ll die. It says even more about how we live.
Elon Musk famously declared that “The most entertaining outcome is the most likely.” This is obviously insane, but it’s solid column-writing advice. So let’s indulge Musk’s Razor and dive a little deeper into the three possibilities: (1) A verdict. (2) Congressional action that cleans up copyright law and obviates the need for a verdict. (3) A settlement.
The most entertaining outcome would clearly be the New York Times taking its case all the way to the Supreme Court where it proves it’s been the victim of such rapacious harm that the justices, emerging in Tom Friedman mustaches, rule unanimously in the Gray Lady’s favor.
Now the entertaining part: The court then agrees with the Times’s demand that OpenAI destroy any chatbot models and training data that use copyrighted Times material. This is known as algorithmic disgorgement — a legal concept to address unlawful gains from algorithmic sneakiness. An ordinary tech company might be forced, at significant cost, to remove infringing material from its product. But generative AI isn’t ordinary tech. It’s created through the combination of hundreds of billions of parameters of information and language. That makes extracting the New York Times from ChatGPT as impossible as removing the eggs from a cake. If subjected to disgorgement, OpenAI could theoretically be forced to eliminate everything it used to train ChatGPT and start over, essentially killing the company. The next day’s Wordle: Pwned.
I don’t think any of this is going to happen. First of all, OpenAI has a better-than-decent fair use argument based on ChatGPT’s ability to transform Times articles into something original. (OpenAI dismisses the Times’s plagiarism claims as a “rare bug.”) Consideration of social benefit is also part of fair use. That’s why news organizations are frequently allowed to use copyrighted materials, or a poetry scholar can use excerpts of a poem. Societally, we’ve decided those are things we want to encourage. You can debate whether ChatGPT use is a social good, but tens of millions of Americans are already using it. Disgorgement, or even an injunction, would harm those people as well as OpenAI. Lastly, the legal process can take years, during which ChatGPT is likely to become even more entrenched. Think of how Uber showed up in cities, nudged out taxis and never asked forgiveness. You don’t have to be righteous to win in court. You just have to be inevitable.
Timely congressional action also seems unlikely. If you need this explained further, I’d like to direct you to The Post’s circulation department, where a staffer has a few questions about why you subscribe.
That leaves a settlement, which sounds anticlimactic. But given the parties’ current positions there’s more drama to be found in a negotiation than in slow motion legal combat.
According to several people familiar with his thinking, A.G. Sulzberger, chairman of the New York Times Company and publisher of the New York Times, believes he has been robbed of billions of dollars of intellectual property. He knows that OpenAI has already ingested the Times’s content and can’t take it out. A one-time payment is a nonstarter, because even if OpenAI made a Godfather offer, there’s no guarantee that today’s windfall won’t one day lead to AI swallowing his family business. Having almost perished in the first wave of AI-powered social media algorithms, the New York Times sees a chance to soak a tech company for a massive licensing fee on a long-term contract — billions of dollars in recurring revenue. And if it can’t get its number, the company has a proud history of going to court to defend its principles.
On the other side, OpenAI is at the peak of its “let’s incinerate cash on the way to market dominance” phase. Everyone is giddy with possibility, which makes it a perfect moment to pay the vig and drop a one-time sum on the New York Times. But: No large recurring payments. The last thing OpenAI wants is to become a royalty administrator, constantly arguing over percentages with every little publisher who can claim to have been a part of ChatGPT’s training. Not only is it a hassle and a drain on future profits, it’s unglamorous by Silicon Valley standards.
Since filing suit, the Times is doing what it does best — enjoying its impact in silent sanctimony. (To be fair, it’s also keeping its ethical wall high; none of the Times reporters I’ve spoken with have any clue what management is thinking about the case, much to their annoyance.) Meanwhile OpenAI has been Jay Gatsby at the White House correspondents’ dinner, buying drinks and striking deals with all the other news organizations it can. The value of those deals is in the very low millions, but it’s a great show of magnanimity toward the Fourth Estate, old sport. A perfect way to demonstrate to the public and the courts that OpenAI isn’t the stubborn one.
What might ultimately drive a settlement — and I’d bet one happens in the next few months — is that both sides have a sliver of existential risk. The Times is one the best financial stories there is in the wasteland of American journalism, but AI has brought huge new unknowns. Securing its sustainability now seems smart. OpenAI has already had one near-death experience, and while a disgorgement ruling is hard to imagine, it’s been forced to admit that ChatGPT is addicted to copyright. “It would be impossible to train today’s leading AI models without using copyrighted materials,” the company wrote in a submission to the United Kingdom’s House of Lords.
It’s that last bit that leads me to think OpenAI is more likely to bend. Now’s the time to fix a price for its drug of choice. When the big record labels battled Spotify, they ended up reaching an agreement to license music to the platform while also getting 18 percent of the company. Spotify didn’t want to do it (and OpenAI would never go anywhere near that high) but a music service can’t exactly live without music. Neither side has been thrilled with the relationship, but it’s a form of mutually assured preservation. Would a 5 percent stake in OpenAI, divided proportionally among the largest news organizations and book publishers, and rebalanced once a year to avoid too much accounting, do the trick?
In the movie, Godzilla kills Mothra with his fiery atomic breath, while Mothra’s children wrap Godzilla in silk and get their vengeance when he lizard-shuffles into the ocean and dies. These would indeed be thrilling outcomes. But with so much at stake, and apologies to Musk, the least entertaining outcome is the most likely.
login
who are we?
contact us
qosheapp
Toi Staff
Gideon Levy
Belen Fernandez
Rami G Khouri
Donald Low
Mort Laitner
Ali Fathollah-Nejad
Nikkei Editorial
Michael Kwet
visit website
