AI and Books: Literacy’s Bookends

“There must be something in books, something we can’t imagine, to make a woman stay in a burning house; there must be something there. You don’t stay for nothing.” — Ray Bradbury, Fahrenheit 451

The rapid rise of artificial intelligence forces us to question the value of essays, books, and deep reading. Is reading still necessary? Does engaging directly with long-form literature still hold real value? Why spend hours working through a complete text when an AI tool can summarize it in seconds?

As Ray Bradbury illustrated in Fahrenheit 451, literature possesses a unique, transformative power. Books spark critical thinking and imaginative depth in human minds. Conversely, over-relying on automated summaries risks systematically eroding our cognitive skills. As argued in Blueprint for Disaster’s essay AI Summaries Are a Threat to Our Cognitive Sovereignty, outsourcing our reading to generative summaries acts like a quiet lobotomy, gradually stripping away our critical thinking, memory retention, and mental sovereignty without our noticing.

Beyond changing how we process information, the AI industry directly opposes literacy by how it acquires data. Because large language models require massive amounts of human text, we must ask two fundamental questions: Are AI companies like Meta, OpenAI, Anthropic, and Grok relying exclusively on copyright-free material? Second, are physical books being destructively processed to feed these systems, and why?

The Atlantic answered the first question by reviewing court documents from recent copyright litigation against tech firms. These records revealed that Meta senior management turned to LibGen, a massive shadow library of pirated books, to train Llama 3 because long-form books provide vastly superior training data compared to standard web scrapes. The same legal filings indicated that OpenAI also drew from LibGen in its earlier development. Because these internal datasets remain proprietary, the public cannot verify which specific works were used or excluded.

However, pirated digital files were not enough for every company, leading directly to the second question. In The Guardian, Kathryn James analyzed internal documents exposed during the northern California lawsuit Bartz v. Anthropic PBC. These filings revealed "Project Panama," Anthropic's internal initiative to scan published books on a global scale.

Rather than navigating what Anthropic's CEO called the "legal/practice/business slog" of licensing e-books legally, the company initially turned to pirated sources, leading to a $1.5 billion settlement with affected authors. To circumvent further licensing barriers, Anthropic pivoted to destructive scanning, turning a legal issue into a logistical one under the defense of "fair use." Under this process, vendors sliced the spines and edges off physical books, scanned the individual pages into digital formats, and destroyed the paper originals afterward.

As these documented cases demonstrate, generative AI acts as an enemy of literacy and literature. It simultaneously destroys physical works of human ingenuity and weakens the neural networks required to appreciate them.