Summary of Key Points
A mysterious large customer has been purchasing a large number of obscure and rare books through second-hand book dealers. Media tracking has revealed that these books are eventually shipped to an Amazon warehouse in Las Vegas, where workers cut the spines from the books, scan their contents, and then shred them for destruction—the purpose of providing “exclusive training data” for large AI models. This incident has been likened by the media to the American version of the “Burning of Books and Execution of Scholars” (a historical event), sparking a debate about the acquisition of AI data and the protection of cultural heritage.
Detailed Analysis
1. Why Do AI Models Prefer Obscure and Rare Books? – The “Picky” Nature of AI Training
Training large AI models is similar to how humans learn; they need to consume a vast amount of “unseen knowledge” to become intelligent. The content of popular books (such as classic novels and textbooks) has already been digitized, and many AI models have already been trained with this data, so using it again doesn’t significantly improve their performance. In contrast, obscure and rare books (such as niche academic works from decades ago, out-of-print local chronicles, and old magazines) are largely not available in digital form, making them a “new source of information” for AI models. Training with these unique materials allows the models to gain knowledge that others don’t have, enabling them to answer questions more comprehensively. Additionally, the copyright on these books may have expired (e.g., more than 50 years after the author’s death), or the copyright holders may be unreachable, reducing the legal risks for AI companies.
2. Is the Metaphor of “Burning of Books and Execution of Scholars” Exaggerated? – The Death of Physical Books vs. the Preservation of Knowledge
The media’s use of this metaphor is to highlight the destruction of physical cultural artifacts, but it differs from the historical event in essence:
- Historically, the burning of books was aimed at eliminating ideas and suppressing scholars; here, the physical books are destroyed, but the content is scanned and preserved in AI systems (which is like removing the “soul” of the books while keeping the “shell”).
- However, the value of rare books goes beyond their text; some contain author’s handnotes, old bookplates, and unique binding designs that cannot be captured in digital scans. There are also unique copies that exist only in one copy in the world, and if the physical books are lost due to scanning errors, the content could be permanently lost. For cultural researchers, this represents a significant loss—similar to destroying the artifacts in the Forbidden City and only retaining photographs.
3. Who Are the Participants in This Process, and Who Is In the Dark?
- Second-hand Book Dealers: Most are unaware of the ultimate fate of the books; they see a large customer willing to buy their inventory of obscure titles and make a profit. They may be curious but are not informed by the customer.
- Amazon Warehouse Workers: They perform their tasks according to procedures: cutting book spines for easier scanning, scanning the content, and shredding the books. They may know that the books are being processed, but they may not realize they are for AI training or that they are rare editions.
- The Mysterious Customer: It is likely an AI company or a service provider specializing in AI data. They operate in secrecy to avoid criticism of “cultural destruction” or to prevent competitors from learning about their acquisition of these unique data.
4. The Central Debate: Does AI Progress Require the Sacrifice of Cultural Heritage?
This incident highlights a contradiction: AI needs massive amounts of data to develop, but the methods used to acquire it may harm cultural heritage.
- Supporters argue that AI represents the future, and scanning these books makes knowledge more accessible to everyone through AI, which is better than letting them gather dust in warehouses.
- Opponents argue that rare books are part of cultural heritage, and their physical existence has intrinsic value. For example, a 19th-century plant atlas contains not only text but also hand-drawn illustrations and scientists’ notes, which AI data cannot replace. If everyone follows this practice, it will become increasingly difficult to find rare physical books in the future.
- A middle ground suggests exploring alternative approaches, such as collaborating with libraries to digitize rare books while preserving the physical copies, or having AI companies fund the restoration of these books before scanning them, thus achieving both data acquisition and cultural preservation.
5. What Are the Impacts on Ordinary People?
- Finding Rare Books Becomes More Difficult: If you are a collector of rare books or need a specific obscure title, you may find that they have been acquired by AI companies, driving up their prices in the market.
- AI Answers Become More Inclusive: AI models may provide more accurate answers to rare or obscure questions due to their exposure to these rare books.
- Loss of Cultural Memory: As more rare physical books are destroyed, we risk losing tangible connections to the past. For example, if you want to see the old books that your grandparents read, you may only be able to access digital versions in AI systems.
This incident reflects a clash between technological advancement and cultural preservation: AI needs data to grow, but we must ensure that cultural heritage is not lost in the process. In the future, we may need to establish more balanced rules to allow both AI and culture to coexist peacefully.