Can I Sue an Artificial Intelligence Company for AI Copyright Violations?

Although the issue hasn't been settled, you likely can sue a company that uses your copyrighted work to train its artificial intelligence software. Copyright gives you the power to decide the terms under which your work may be reproduced. Unless you grant the AI company permission to use your work for this purpose, you may have a copyright infringement claim (among other claims) against the AI company under federal law.

The interaction between generative AI and copyright law is a new area of the law. Lawyers may have different views about the legal uses of copyrighted works. If you're upset at the prospect of your life's work being used to help others generate imitations, you may want to consult an experienced intellectual property and copyright attorney. They can help you decide whether you have a valid lawsuit. They can also start the court process with a letter to the AI company demanding that they respect your copyright.

Artificial Intelligence

There is nothing new about people wanting to create something that can think for itself. For example, in Frankenstein, Mary Shelley wrote about a creature composed of human parts that learns to think and feel. Now, we have machines that are close to thinking and feeling for us. We call this artificial intelligence or AI.

The birth of artificial intelligence, as we think of it today, stems back to 1950. That year, Alan Turing, the father of computers whose story is told in the award-winning The Imitation Game, posed the question directly: “Can machines think?" Scientists have since spent decades developing software programs that can do things, with varying degrees of success, that historically have required human intelligence to do.

Current AI Models

The technology is developing rapidly, but so far, AI can do some amazing things through machine learning. For example, OpenAI, in partnership with Microsoft and its subsidiary GitHub, developed a system called Copilot that generates open-source code. OpenAI's DALL-E 2 can generate AI art from text that looks like masterpieces. Stability AI's Stable Diffusion and Midjourney are two other popular text-to-image AI art tools. And DoNotPay has announced the world's first “robot lawyer."

AI systems must be trained to function. AI programmers use a body of materials called a “corpus" that is processed by the AI and used to generate its future output. At the risk of oversimplifying, an AI model is only as good as the breadth of its corpus. The more it “studies," the more it can do, and the more human it seems to be.

ChatGPT

One AI model that has taken the public by storm is ChatGPT, another OpenAI creation. Released on November 30, 2022, ChatGPT (GPT stands for “generative pre-trained transformer") is a large language model that can be used for, among other things, natural language processing tasks, such as writing articles, essays, books, and even poetry. The architecture was developed by an OpenAI team of researchers led by Rewon Child, David Luan, Alec Radford, and Jeffrey Wu.

One of the key features of the ChatGPT architecture is its ability to learn from large datasets. OpenAI claims that the model has been trained on a “massive" amount of data. Although they have not publicly released the amount of data, what it is, or where it came from, OpenAI believes it to be “in the range of hundreds of billions of words."

AI raises a concern for anyone who creates — writes, draws, drafts, paints, programs, or codes — for a living. What if an AI company is using works you created to train their AI model? They didn't ask permission, they won't give you credit, and they certainly didn't pay you for it.

Isilmir's Hammer

Let's say you're a professional fiction writer. After years of supporting yourself as a barista in a local coffee shop, you come up with an idea for a new fantasy series. Your first book, Isilmir's Hammer, was a New York Times Best Seller and won you a Hugo award. Your second and third books, Isilmir's Bane and Kalabar's Doom, were even more successful. You have been approached by Hollywood about drafting a screenplay. Things are looking up; no more having to listen to people complain about too much foam in their lattés.

Unbeknownst to you, WriteAI, the developers of an AI natural language generation model of the same name, used your books as part of their enormous training corpus. They decide to make WriteAI available for free.

Benedict Arnold, a 12-year-old kid from Nebraska who loved your series so much that he named his dog “Isilmir," decides to make his own fan fiction based on your style. So, he creates a WriteAI account and tells it to produce a fantasy novel in your style about a dwarven lord, Kerilbow. A few minutes later, WriteAI spits out a roller coaster of an epic. Benedict calls it Kerilbow's Threat, uses Midjourney (an image generator) to create cover art, and self-publishes the story on Amazon.

What took you years to make takes him all of 40 minutes, and you're not even mentioned in the acknowledgments. WriteAI is so good that the story reads just like you wrote it. “This can't be right," you think. Well, it isn't.

Copyright Law

Since the 1790s, federal copyright law has protected original works of authorship fixed in a tangible form. A copyright gives the holder exclusive rights, including the right to:

• Reproduce the work
• Make "derivative works" (other works based on the original work)
• Distribute copies of the work
• Perform the work publicly
• Display the work publicly

Although you don't have to register your copyright with the U.S. Copyright Office, doing so has certain advantages. For example, the copyright owner needs to register their copyright in order to bring a copyright infringement suit.

You have a copyright in your fantasy series that the law requires others to respect. The question is whether WriteAI violated it by training on your works to generate derivative works. At this time, the law isn't clear on whether you can sue an AI company for copyright infringement based on the work it produces. That's not to say you couldn't sue them for some other reason if warranted, such as:

• Defamation
• Consumer fraud
• Conversion
• Unfair competition
• Violation of federal data privacy laws

You would definitely want to consult an experienced attorney if you were planning to sue it for Benedict's Kerilbow's Threat.

Training AI and the Fair Use Doctrine

You can probably sue the AI company for using Isilmir's Hammer, Isilmir's Bane, and Kalabar's Doom to train WriteAI. Your copyright gives you the right to set the terms under which works based on your original work, called "derivative works," are created. Benedict certainly didn't get your permission to make a derivative work.

At least one AI company has told the U.S. Copyright Office that it can use copyrighted works to train its AI model under the fair use doctrine. Fair use lets you use limited portions of a copyrighted work, including quotes, for purposes of commentary, criticism, parody, news reports, teaching, or scholarly reports. To show that a use is “fair use," the AI company must persuade a judge that the following factors weigh in its favor:

• The purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes
• The nature of the copyrighted work
• The amount and substantiality of the portion used in relation to the copyrighted work as a whole
• The effect of the use upon the potential market for or value of the copyrighted work

This is a cutting-edge area of the law, so lawyers may have differing opinions. But if you look at these factors, you have a good argument that using someone's entire work to train an AI model is not fair use. Instead, you would argue each one supports protecting your copyrighted work.

Purpose and Character of the Use

The purpose and character of the use is to allow others to create similar works without time and effort. It's as simple as that. The AI developers may have noble reasons for training their model on your work, but Benedict isn't using WriteAI to save puppies. He's using it to make money. While some might use WriteAI for non-commercial reasons (such as getting out of having to write their own college essays), the commercial applications are undeniable. Training an AI model on copyrighted works allows users to create competing works.

“But," the AI company says, “this technology is really useful – transformative, even — and it's the way of the future!" So what? A technology being “useful" does not justify depriving copyright holders of the rights to their works. If the company wants to use copyrighted works to train its model, it can do what everyone else does—get the owner's permission.

“But, but," the AI company says, “that would be hard. We needed to use many works to make a really good model." That the AI company stole a lot of people's works and used them to its own ends doesn't mean it should get to keep its ill-gotten gains. They should fix their algorithm. You don't lose the rights to your property — and that's what a copyright is — because it may be easy for a talented thief to steal.

Nature of the Copyrighted Work

The second factor, the nature of the copyrighted work, tilts in your favor. You didn't write your series overnight. It took years and a lot of hard work. It took sacrifice — you barely made ends meet as a barista while you were writing it. Because of the time, effort, and sacrifice involved in creating your series, this factor of the fair use doctrine would weigh in favor of copyright protection.

Amount of the Work Used

This third factor is so easy it's obvious. WriteAI trained on your entire series. All three books. It processed and digested every single word you wrote. It analyzed your plot lines, themes, characters, imagery, and settings. Using an entire work without permission for your own ends cannot be “fair use."

Effect on the Market of the Copyrighted Work

Finally, if training AI on copyrighted works were considered “fair use," it would destroy the market for your books. Fantasy and science fiction books generate around $600 million in sales each year in the United States. Depending on the publishing deal, royalties can vary between 8% and 60% of every book sold. For a bestselling, award-winning author, that's a lot of money.

Kerilbow's Threat could crater your market. If someone can get a story in your style for cheap, there would be little or no reason for them to shell out the money to buy your books. Your books would become worthless.

And that's just your books. If other Benedicts can create and publish their own novels in a famous author's style in 40 minutes, they could flood the market with competing AI creations. This would hurt the sales of all authors' works.

A Lawyer Could Help

It seems that the fair use doctrine would not apply. Although courts haven't settled the issue, AI companies probably don't have the right to train their models using your copyrighted works. Expect a flood of lawsuits based on training data, including class-action lawsuits.

In the meantime, if you create for a living and don't want AI companies to use your works without your permission, you should consider contacting a capable copyright attorney. They would be able to give you their opinions about this cutting-edge area of the law. They can also provide legal advice about whether at least a cease-and-desist letter or maybe a lawsuit is your best option.