Rare Book Monthly

Articles - February - 2025 Issue

Your Chatbot May Be Using Illegally Pirated Books to Answer Your Questions

A battle is brewing between an ancient source of information, the book and its authors, versus a new invention, the chatbot and its developers. The chatbot is a program that can answer whatever questions you throw at it. The grandaddy (all of three years old) and most famous chatbot is ChatGPT. It uses artificial intelligence (AI) to quickly sort through reams of information to answer your every question. But, where does it get that information? One of the major sources is books, copyrighted books. When the chatbot uses that information to answer your questions, the authors and publishers of those books get nothing. That makes them sad (perhaps a better word is “angry” or “POed”).

 

Some authors are angry enough to go to court. There are various cases floating around out there but a notable one pits comedian and writer Sarah Silverman against Meta, operator of Facebook, headed by Mark Zuckerberg. Meta's chatbot, Llama, is the culprit here.

 

It is alleged that Meta used the LibGen (Library Genesis) dataset to train its Llama chatbot. LibGen is a notorious, shadowy entity, possibly operating out of Russia. It's dataset contains over 196,000 pirated books. LibGen has been in the news before for “lending” its pirated books free of charge without compensating the authors. LibGen infringes on authors' copyrights and operates illegally but it doesn't matter. They can't be shut down or forced to pay because they can't be found. They regularly change their urls to avoid being shut down. LibGen is no small operation, receiving an estimated 9 million visits per month from the U.S. to “borrow” books. It is supported by donations (accepted in untraceable bitcoin only).

 

What Meta has been accused of doing is using this large pirated database of books to supply Llama with much of the information it needs to answer users' questions. The plaintiffs have alleged that approval to do so came from the top, Mr. Zuckerberg himself. This claim has focused on the use of pirated (illegally obtained) books, but that perhaps is not the biggest issue here. What if the books were legally obtained, purchased, borrowed from a physical library, or received as gifts. Would that be any better from a copyright standpoint? Probably not.

 

In Meta's opinion, this use of the authors' work fits under the “Fair Use” exception to copyrights. “Fair Use” is what lets you quote from a book, write a review or book report, use information you found therein to write something of your own, without violating its copyright. Generally speaking, if you change what you read, add your own twist, copy only a small portion, and such, you are not guilty of copyright infringement. What Meta is doing, leaving aside the issue of using LibGen's pirated texts, is both copying the entire book, but then only sharing a small, rewritten portion such as might be expected to pass the Fair Use text.

 

This will have to play out in court but the Judge seems less than impressed with the arguments made by the authors. The reality is that chatbots provide very useful information. You probably use one to answer your questions. It's sort of like speaking to a very learned individual. Practically speaking, paying 196,000 authors some small pittance each would be an absolute nightmare, and they might not agree to such an arrangement anyway. It's not that they don't deserve anything, but it probably isn't a lot, and making such demands might force the shutting down of this very new and useful technology altogether. Progress is hard to stop, even if some people feel hurt by it, and my guess is the courts will not do so here.


Posted On: 2025-02-04 01:22
User Name: keeline

I don't know that an AI LLM (large Language Model) like ChatGPT and others fits the usual definition of a "chatbot." That word is usually used for software that is in a chat session and may represent itself as a human. They have been around for a long time.

There is an allegation that the LLMs have been trained on collections of books, copyright and public domain, along with crawled websites, posts in groups, and more. But a court case will have to determine what was really used. We may only know if it is decided IN COURT. Often, an out-of-court settlement will leave many important details confidential.

Although LLMs create an illusion of intelligence, they are not. The best you can say is that they can write a word salad that is passable. Sometimes even this is after several tries and refinements of the prompt that it responds to.

Attempts to get ChatGPT to write fiction in a certain style usually fails. At the very least there are major plot holes that a human editor/reader would spot. I've seen several examples of this.

LLMs are better with words than they are basic arithmetic. This is part of the reason why the AI-generated art often has logical errors and too many fingers or not enough keys on a typewriter. It's a bit like looking at a toy train where the designer has a vague notion of a train but not an understanding of the real parts of a steam locomotive.

When people ask for book IDs from ChatGPT, I have seen it make up books that do not and never did exist. One example invented a volume in the Judy Bolton series which never existed and was not even a proposed story. This leads people down false trails and spending energy to look for books that do not exist. Then there is the energy required to complete the LLM query in the first place.


Rare Book Monthly

  • DOYLE
    Rare Books, Autographs & Maps
    July 23, 2025
    DOYLE, July 23: WALL, BERNHARDT. Greenwich Village. Types, Tenements & Temples. Estimate $300-500
    DOYLE, July 23: STOKES, I. N. PHELPS. The Iconography of Manhattan Island, 1498-1909. New York: Robert H. Dodd, 1915-28. Estimate: $3,000-5,000
    DOYLE, July 23: [AUTOGRAPH - US PRESIDENT]FRANKLIN D. ROOSEVELT. A signed photograph of Franklin D. Roosevelt. Estimate $500-800
    DOYLE, July 23: [ARION PRESS]. ABBOTT, EDWIN A. Flatland. A Romance of Many Dimensions. San Francisco, 1980. Estimate $2,000-3,000.
    DOYLE, July 23: TOLSTOY, LYOF N. and NATHAN HASKELL DOLE, translator. Anna Karénina ... in eight parts. New York: Thomas Y. Crowell & Co., [1886]. Estimate: $400-600
    DOYLE, July 23: ROWLING, J.K. Harry Potter and the Goblet of Fire. London: Bloomsbury, 2000. Estimate $1,200-1,800
  • Sotheby’s
    Books, Manuscripts and Music from Medieval to Modern
    Now through July 10, 2025
    Sotheby’s, Ending July 10: Book of Hours by the Masters of Otto van Moerdrecht, Use of Sarum, in Latin, Southern Netherlands (Bruges), c.1450. £20,000 to £30,000.
    Sotheby’s, Ending July 10: Albert Einstein. Autograph letter signed, to Attilio Palatino, on his research into General Relativity, 12 May 1929. £12,000 to £18,000.
    Sotheby’s, Ending July 10: John Gould. The Birds of Europe, [1832-] 1837, 5 volumes, contemporary half morocco, subscriber’s copy. £40,000 to £60,000.
    Sotheby’s
    Books, Manuscripts and Music from Medieval to Modern
    Now through July 10, 2025
    Sotheby’s, Ending July 10: Ian Fleming. A collection of James Bond first editions, 8 volumes in all. £8,000 to £12,000.
    Sotheby’s, Ending July 10: J.K. Rowling. Harry Potter and the Philosopher's Stone, 1997, first edition, hardback issue. £50,000 to £70,000.
    Sotheby’s, Ending July 10: J.R.R. Tolkien. Autograph letter signed, to Amy Ronald, on Pauline Baynes's map of Middle Earth, 1970. £7,000 to £10,000.
  • Rare Book Hub is now mobile-friendly!

Article Search

Archived Articles