Artificial Intelligence is transforming how people create, consume and preserve information. But as AI companies build systems capable of processing billions of words, a more unsettling question is emerging. What happens to the books and human creativity that feed these systems?
One of the most important ways AI companies develop their systems is by providing them access to enormous amounts of written material. Books, articles, websites and other forms of human-created text provide the data needed to train and improve AI models.
Anthropic, the company behind Claude, faced this question in a federal copyright case after authors sued the company over its use of copyrighted books.
The details of the project became especially significant because they suggested that Anthropic was not simply collecting digital texts that were already online. Through Project Panama, Anthropic planned to “destructively scan all the books in the world,” according to an internal memo included in court exhibit 21. The memo also advised discretion: “Why use a code name? … [B]ecause we don’t want it to be known that we are working on this.”
The court found that training AI on lawfully acquired books was fair use, while creating a permanent library from pirated copies was not. The court also compared AI training to the way humans learn from books, stating:
“Like any reader aspiring to be a writer, Anthropic’s LLMs trained upon works not to race ahead and replicate or supplant them,” the court wrote. “But to turn a hard corner and create something different.”
But is an AI system really learning from a book in the same way a human does?
Ben Cater, professor of history and director of the Honors Program at Point Loma Nazarene University, doesn’t think so.
“When a machine absorbs information, they absorb the text, not the context,” Cater said. “If a human reads something and absorbs information, they absorb the text, but they are also absorbing the context. There is a wealth of perspective that humans bring to make sense of the text.”
“Machines may have some sentience … but without question the things that make humanity marvelous is that we have a will, we have the ability to create and to reflect,” Cater said. “The will to create and reflect makes us completely different learners of text, compared to a machine.”
The scale of AI training further complicates the comparison. A person can read a book and learn from it, while an AI company can process billions of words to build a commercial system.
For Cater, however, a digital copy is not necessarily equivalent to physical artifacts.
“They are not the same things,” Cater said. “Physical artifacts matter, they are a fuller and more comprehensive way to learn about a text when you hold a book in your hands.”
A rare book can contain information beyond its words. Its physical condition, markings, binding and other characteristics can provide historians and scholars with evidence about the time and people connected to it.
This raises a larger question about cultural memory. If a rare physical book is destroyed and its digital version becomes the only surviving copy, who controls that record?
“What’s to stop them,” Cater asked, “from deleting entire books, or rewriting chapters, or even modifying a single word to suit their ends?”
The concern is what could happen when important cultural and historical material becomes concentrated in digital systems controlled by powerful organizations.
Charles Graeber, an author involved in the Anthropic lawsuit, wrote in The New York Times that AI cannot continue improving by relying only on AI-generated material.
“A.I. can’t keep improving by training only on A.I. output,” Graeber wrote.
The contradiction is this: AI systems depend on high-quality human writing while the growth of AI could change the economic environment in which that writing is produced.
For Cater, the larger concern is what society gives up in exchange for technological convenience.
“The convenience is the poison,” Cater said.
AI may be able to process billions of words and may continue to transform literature, but technology cannot change the fact that its most valuable source material began with human beings creating, reflecting and telling their stories.
The question is not simply what AI is legally allowed to learn from books. It is what happens to the human culture that created those books in the first place.
Nathan Gibbs