I find myself as an author in the midst of the AI chat revolution. I am one of the creators of several books that have been incorporated into LLaMa, the chatbot developed by Meta Platforms. This puts me in competition with OpenAI’s ChatGPT bots, which are more well-known in the market. Thanks to Alex Reisner of the Atlantic, there is now a search tool available for the database called Books3. This tool allows authors worldwide to search for their names and assess the impact of their books on LLaMa. Personally, I haven’t fully made up my mind about this situation. On one hand, I feel a bit annoyed that only three out of my seven books have been included in LLaMa’s training. On the other hand, I consider the value of my contribution and why I shouldn’t receive compensation for it.
The reactions from other authors, both prominent and lesser-known, have been quite diverse. Some have expressed genuine outrage. This includes well-known novelists such as John Grisham, George R.R. Martin, and Scott Turow, who are members of the Authors Guild and involved in a copyright infringement lawsuit against OpenAI. Sarah Silverman is also a plaintiff in a similar legal battle against Meta Platforms and has shared her frustration on social media. Margaret Atwood and Lauren Groff are among those who have used social media to express their irritation or even outright fury.
However, there is another camp of authors that doesn’t see this as a big deal. For instance, Ian Bogost, the author of ten books primarily centered around game-playing, wrote an article for the Atlantic titled “My Books Were Used to Train Meta’s Generative AI. Good – It Can Have My Next One Too.” Lastly, Stephen King seems to have a more resigned reaction. Despite seeing a database listing 87 of his works, he does not oppose the idea of his stories being taught to computers. He compares himself to King Canute, who famously attempted to command the tide not to come in.
Before delving further into the legal aspects, let’s take a detour to understand what the database and its usage mean in the context of “generative AI” – the technology category that these chatbots fall into. As I’ve written previously, these products are inaccurately labeled as “artificial intelligence.” They are not intelligent in the same way that humans or animals are. They are designed to appear intelligent to outsiders, while the actual electronic processes behind them are different. Using the term “artificial intelligence” distorts our perception of what they actually do. They don’t truly learn or create perceptions of the world based on the information they possess. They lack the sentience required for true creativity, as King himself remarks, although he adds a caveat that they may, in the future, become creative.
Chatbot developers “train” their systems by feeding them trillions of words and phrases from the internet or specialized databases. When a chatbot answers a question, it generates a string of responses based on probability, aimed at resembling a human’s answer. However, it remains an imitation of human thought rather than genuine thinking. What’s interesting about the revelation that Books3 was used to train LLaMa is that it highlights the fact that everything produced by these chatbots ultimately originates from human sources. Although Meta Platforms has not officially confirmed the use of Books3, its role was disclosed in a technical paper written by the developers of LLaMa.
Another important point to consider is that none of the training methods have fully resolved the significant problem faced by chatbots – their frequent and spectacular inaccuracies. When chatbots cannot find factual information to respond with, they often fabricate or provide irrelevant answers. The resemblance of their answers to human thought and speech can mislead users, resulting in embarrassing and costly consequences. This issue is prevalent in the broader field of AI. Notably, on September 20, the respected journal Nature retracted a paper by Google researchers that claimed an AI system could design computer chips in a matter of hours, which typically takes human designers months. The opposite conclusion was reached by the author of the paper. As for my personal experience, despite the rigorous training LLaMa received using my books, it didn’t seem to grasp much. Its responses to my questions revealed it to be just as clueless as other generative AI chatbots. When I asked what it knew about me, its answer consisted of a mix of information from a biobox published on latimes.com and the mention of three books. However, none of these books appeared in the Books3 database. Curiously, one of the books listed wasn’t even authored by me (though my name was mentioned in the endnotes), and two of the books seemingly don’t exist. Nonetheless, LLaMa did describe me as “a highly respected and accomplished journalist who has made significant contributions to the field of journalism,” indicating some semblance of wisdom and judgment.
Regarding the concerns raised within the literary world, the confusion arises from the complexities of copyright law. This is especially true when it comes to “fair use,” which allows the use of copyrighted material without permission. Fair use permits the quoting of snippets from published works in reviews, summaries, news reports, research papers, or for parody or transformative purposes. However, defining what qualifies as “transformative” has proven challenging. Courts have arrived at varied decisions, resulting in countless dollars spent on legal fees to establish guidelines. The emergence of new technologies, such as digital reproduction and chatbot training, further complicates matters. The lawsuit filed against OpenAI by the novelists and the Authors Guild argues that OpenAI copied their works without permission or proper compensation, amounting to widespread theft. According to the authors, AI training involves reproducing entire works or substantial portions thereof, deeming it a euphemism for copying and ingesting. They claim that the chatbots pose a threat to the livelihoods of fiction writers by allowing anyone to generate texts that would otherwise require payment to the original authors. Furthermore, these bots can produce derivative works that mimic, summarize, or paraphrase the authors’ original works, thus harming the market for their creations. This is a crucial point because any interference with the marketability of a copyrighted work weakens a fair-use defense in court.
It’s worth noting that AI encroaching on professional skills has been a significant factor in recent Hollywood writers’ strikes and continues to be a concern for actors currently on strike. The limitations on AI usage have been included in the contract that ended the writers’ strike and are likely to feature in any settlement with the actors.
Sarah Silverman and her fellow plaintiffs’ lawsuit against Meta closely mirrors the Authors Guild’s case against OpenAI. It may not work in Meta’s favor that Books3 itself is alleged to be the product of piracy, as some of the works in the database were taken from illicit sources circulating online. In fact, one database host removed Books3 following a complaint from a Danish anti-piracy organization. In its response to the Silverman lawsuit, Meta argues that… [the text ends abruptly here]
Denial of responsibility! Vigour Times is an automatic aggregator of Global media. In each content, the hyperlink to the primary source is specified. All trademarks belong to their rightful owners, and all materials to their authors. For any complaint, please reach us at – [email protected]. We will take necessary action within 24 hours.
