Big-Name Authors Sue OpenAI For Copyright Infringement

[Stevicus]

George R.R. Martin And Other Big-Name Authors Sue OpenAI For Copyright Infringement

Writers like George R.R. Martin and John Grisham say their work has been used without permission to train OpenAI’s ChatGPT—a popular AI chatbot.

www.forbes.com

The writers and the Authors Guild—a professional association for authors—accused OpenAI of “systematic theft on a mass scale,” according to a filing in federal court in New York, alleging the company copied the authors’ works “without permission or consideration” and used the content to train their language models, such as ChatGPT.

The guild said the language models jeopardize fiction writers’ ability to make a living, arguing OpenAI could have instead opted to train their language models on public domain works or paid a licensing fee to use the copyrighted works.

The lawsuit argues OpenAI should be required to receive permission from authors to use their works for language model training and that authors should be compensated for such a transaction.

The guild is seeking unspecified actual damages but is alternatively open to statutory damages up to $150,000 for each infringed work paid out to the plaintiffs and class members.

The guild alleged in a statement that ChatGPT is used to generate books mimicking authors’ works, citing a recent attempt to generate two unreleased volumes of George R.R. Martin’s Game of Thrones series, A Song of Ice and Fire.

OpenAI did not immediately respond to Forbes’ request for comment.

The article quoted OpenAI's response to similar suits:

Responding to two similar lawsuits, OpenAI said in an August copyright filing that such claims “misconceive the scope of copyright, failing to take into account the limitations and exceptions (including fair use) that properly leave room for innovations like the large language models now at the forefront of artificial intelligence.”

I'm not sure what that means, other than it seems they're contending fair use of copyrighted material. Although I'm not sure what it means to train AI using these works. Is it analogous to buying a book and using it to help a child learn to read?

 

[The Hammer]

They should sue. AI has no business creating works of art or language.

 

[Quintessence]

They're not wrong.

If we believe "intellectual property" is a thing - and America certainly believes it is a thing - machine learning language models are trained with data. In this case, that data absolutely includes the "intellectual property" of countless numbers of humans who have gotten neither credit nor compensation for any of their contributions. If this was a case of "for research purposes only" and "not for profit or personal or corporate gain" that would be one thing. But that is not the case here.

Now, one can argue on a philosophical level that all future products are derivative of the present and the past. That in some sense, literally everything is plagiarized and the concept of "intellectual property" is a modern and ridiculous notion. I actually agree with that notion, but it also isn't how American society is set up. It believes in ownership, it believes in property, it believes in giving credit to individuals over recognizing the interconnected and interdependent nature of all existence (aka, individualism > collectivism). So failing to recognize these models are plagiarism and theft will be hypocrisy.

It's a hypocrisy I fully expect to occur, because there's another American philosophy that seems to always trump all others - profit and progress above all else. The allure of profit and progress by exploiting the works of others will be far, far too great for America to bother to abide by its own ethical maxims regarding property rights. And if America doesn't do it, some other nation with a looser idea of property rights - specifically I'm thinking China - will do it instead.

So what will come of all this? We'll see.

 

[Brickjectivity]

I'm not sure what that means, other than it seems they're contending fair use of copyrighted material. Although I'm not sure what it means to train AI using these works. Is it analogous to buying a book and using it to help a child learn to read?

What differentiates the AI from the human is the work involved. Maybe new legislation is required. We could differentiate AI products from 'Works' since AI does not not labor. We could also forbid their products being considered 'Art'. Art takes effort.

A child must use patience to learn to read. One might say that they earn their ability to read, but this is not true for an AI. An AI that is trained is not using patience and is not doing anything. It is being fashioned and is itself art. A child's learning process is creative, so that is not the same.

Which leads us to the question of who owns the AI. If it is trained on my book, then do I partially own the AI and all of its outputs? Maybe.

 

[McBell]

I am interested in how the courts are going to rule in these cases.

As I strongly suspect they will be taken all the way to to top court before it is all said and done, I can not help but wonder if I will still be around when it reaches its final verdict...

 

[Heyo]

I'm not sure what that means, other than it seems they're contending fair use of copyrighted material. Although I'm not sure what it means to train AI using these works. Is it analogous to buying a book and using it to help a child learn to read?

It is analogous to a child learning - and as the child (or their parents) have to buy the book, it is only fair that OpenAI should also buy the book.

 

[Viker]

AI has the ability to learn law. Focus should be put on that.

 

[Debater Slayer]

AI has no business creating works of art or language.

If the copyright issues are resolved fairly, why not?

 

[Debater Slayer]

Although I'm not sure what it means to train AI using these works. Is it analogous to buying a book and using it to help a child learn to read?

These AI models are trained on existing data and select features to pick up from them depending on what the subject of training is and what features the algorithms the models use focus on. For example, if an AI is trained on George R. R. Martin's works, it could, depending on the algorithms it uses, pick up features of his writing style, choice of vocabulary, and sentence structure. If an art-generating AI is fed data containing Renaissance-era art and trained on them, it could pick up features of the art style, shading patterns, color palettes, and sometimes even the smaller details like proportions of each artistic element relative to the others in a painting.

This is the main reason some of these AI models sound so human-like: They're trained on data containing features of how we talk, how we structure our sentences, and what vocabulary we typically use. They technically don't just imitate us either; the algorithms enable them to absorb those features and generate output based on them well beyond superficial imitation.

Trained on huge amounts of data from the Internet, including conversations, it was also trained using a machine learning technique called Reinforcement Learning from Human Feedback (RLHF), in which human trainers provided the model with conversations in which they played both the AI chatbot and the user.

Because of this method of learning, ChatGPT answers can come across as natural-sounding and human-like. And the bot is not just parroting text it has learned. According to Professor Mike Sharples, a professor of Educational Technology at the Open University in the UK, OpenAI’s language model is “creating an internal representation, not just at the surface text, but of the ideas and concepts behind it”.

ChatGPT: Why everyone is talking about the most advanced AI chatbot

It took just five days for ChatGPT to reach 1 million users. We take a look at why it's so popular, how it works and what else it can do.

www.euronews.com

I can see a solid case for either prohibiting the usage of copyrighted material as training data for AI (although that could be a nightmare to enforce, if it were possible at all to enforce in practice) or imposing a requirement to pay authors whose works are being fed to AI models for training, since the AI models use said works to generate commercially profitable output that may also be quite similar to those works in many ways. If a human did the same thing manually, I suspect they would be sued and forced to compensate the authors. I don't see why doing it using AI should be treated any differently.

 

[McBell]

I can see a solid case for either prohibiting the usage of copyrighted material as training data for AI (although that could be a nightmare to enforce, if it were possible at all to enforce in practice) or imposing a requirement to pay authors whose works are being fed to AI models for training, since the AI models use said works to generate commercially profitable output that may also be quite similar to those works in many ways. If a human did the same thing manually, I suspect they would be sued and forced to compensate the authors. I don't see why doing it using AI should be treated any differently.

Can a person copyright their style?
And if so, why haven't they done so?

 

[Stevicus]

These AI models are trained on existing data and select features to pick up from them depending on what the subject of training is and what features the algorithms the models use focus on. For example, if an AI is trained on George R. R. Martin's works, it could, depending on the algorithms it uses, pick up features of his writing style, choice of vocabulary, and sentence structure. If an art-generating AI is fed data containing Renaissance-era art and trained on them, it could pick up features of the art style, shading patterns, color palettes, and sometimes even the smaller details like proportions of each artistic element relative to the others in a painting.

This is the main reason some of these AI models sound so human-like: They're trained on data containing features of how we talk, how we structure our sentences, and what vocabulary we typically use. They technically don't just imitate us either; the algorithms enable them to absorb those features and generate output based on them well beyond superficial imitation.

ChatGPT: Why everyone is talking about the most advanced AI chatbot

It took just five days for ChatGPT to reach 1 million users. We take a look at why it's so popular, how it works and what else it can do.

www.euronews.com

I can see a solid case for either prohibiting the usage of copyrighted material as training data for AI (although that could be a nightmare to enforce, if it were possible at all to enforce in practice) or imposing a requirement to pay authors whose works are being fed to AI models for training, since the AI models use said works to generate commercially profitable output that may also be quite similar to those works in many ways. If a human did the same thing manually, I suspect they would be sued and forced to compensate the authors. I don't see why doing it using AI should be treated any differently.

You make some excellent points here, DS, and while reading it, I was struck by how much of the focus is in trying to teach them to emulate humans so that they sound as human-like as possible. But I have to wonder if they're actually trying to replicate or replace humans, or if it's meant to enhance the human condition or do things that humans really can't do?

For example, @Viker mentions above that AI can be used to learn law. It's conceivable that an AI lawyer could be far more objective and be able to practice law without passion or prejudice far better than emotional, illogical humans. I have nothing against AI in principle, although I resist the idea of using AI to imitate or act like humans, nor do we really need AI to create works of art or write stories of fiction. But we could use AI to do the things that humans can't do or have a difficult time doing.

 

[F1fan]

If the copyright issues are resolved fairly, why not?

I don't think we can call it art. It is a machine mimicking art.

I don't mean to say all art is good or valuable, but art is defined as human expression, and it has meaning and value on that as a core intention. Still we won't see AI do pottery, so will be limited to 2 dimentional mimickry.

There is an ad that advocates for the advantages of AI and one example they gove is AI helping a person design clothes. Well my impression is that it's no longer humans doing the designing, just humans programming a machine to invent ideas. To my mind what is the point of creativity if it's just going to be machines? Will we be impressed by Hollywood stars walking down the red carpet wearing a dress designed by computer Fg549hlMZ?

I suspect many will want the organic human creativity over what machines can put together. To my mind an ideal use of AI is things like drawing congressional maps in a fair way that represents the citizens of the state. The gerrymandering we see in many states is a sort of creativity that we can do without.

 

[Quintessence]

Can a person copyright their style?
And if so, why haven't they done so?

The notion that this would even be necessary is so ridiculous it's genuinely depressing.

All hail the new era of identity theft - REAL identity theft, not this "identity theft" that's just about taking a name and some numbers! All hail the new era of identity theft - where deeply personal means of expression and self can be stolen from you without rhyme or reason! Oh boy, oh boy, I'm so very excited about this!

*insert vomit emote*

 

[McBell]

The notion that this would even be necessary is so ridiculous it's genuinely depressing.

All hail the new era of identity theft - REAL identity theft, not this "identity theft" that's just about taking a name and some numbers! All hail the new era of identity theft - where deeply personal means of expression and self can be stolen from you without rhyme or reason! Oh boy, oh boy, I'm so very excited about this!

*insert vomit emote*

So i should hurry up and copyright the "horror" genre, right?

 

[Quintessence]

So i should hurry up and copyright the "horror" genre, right?

Go for it. It's probably my least favorite genre anyway.

 

[ADigitalArtist]

Go for it. It's probably my least favorite genre anyway.

Aw, really? I love horror. Not because it's all good, definitely not a fan of teen slasher style movies, but I love psychological horrors that have something to say about the nature of humanity, consciousness, fear, control, etc.

Some of it can have fantastic prose, too. Pretty sure AI has a field say with 'write X in the style of Lovecraft.'