The video addresses claims circulating on social media that AI companies are destroying antique books by burning them, focusing specifically on Anthropic, a major AI firm. Internal documents reveal that Anthropic has been secretly scanning vast numbers of physical books, raising concerns about the destruction of unique cultural treasures. However, the situation is more nuanced than the dramatic headlines suggest.
AI models like ChatGPT and Claude require enormous amounts of high-quality text data for training. Initially, companies scraped the internet for this data, but the internet’s content is now largely saturated and increasingly generated by AI itself, which leads to a decline in quality known as model collapse. Consequently, older books, especially those written before the AI era, have become valuable sources of curated, reviewed, and structured information, making them ideal for training AI systems.
In recent months, book sellers and researchers noticed large-scale automated purchases of physical books by companies like Anthropic. This practice is partly driven by a recent $1.5 billion copyright settlement involving Anthropic, where the court ruled that training AI on books is legal, but downloading pirated copies is not. To comply, Anthropic buys physical copies, digitizes them, and then destroys the physical pages, a process deemed legal under US law as it only changes the format of the content.
The digitization process involves buying books, removing their spines to feed loose pages into high-speed scanners, and then discarding the physical pages. Importantly, Anthropic primarily targets out-of-print, technical, or niche books from the late 20th century, which are often not digitized or available online. These books are typically non-fiction and academic, and most are not rare since many copies exist, reducing the risk of losing unique cultural artifacts.
While some book sellers have stopped supplying books for this purpose, others welcome the sales of old stock. Critics worry that AI companies might eventually target rare books, but experts argue this is unlikely due to the difficulty and limited added value of rare texts compared to more accessible 20th-century publications. Ultimately, the video invites viewers to consider whether this practice is a form of smart recycling that transforms forgotten paper into accessible knowledge or a quiet destruction of historical materials.
