Amazon Destroys Rare Books to Train AI Models

In a striking turn for a company that began its journey as an online bookseller, recent findings indicate that Amazon is now purchasing large quantities of rare books, then cutting off their covers and separating their pages to scan them and convert them into text data used to train artificial intelligence models. This practice, uncovered by an investigative report, highlights the growing hunger among tech companies for authentic textual content.

How Was the Story Uncovered?

According to a report published by "404 Media," the journalism team placed a tracking device inside one of the rare books to learn its fate, and it ended up at an Amazon facility in the city of Las Vegas, United States. This facility is known internally as "VGT3" and bears a whimsical logo depicting a dinosaur holding a book between its claws.

In response to inquiries, Amazon explained that it "purchases books through standard commercial channels with the aim of improving the products and services its customers use," without denying the essence of what was revealed.

Why Do Companies Specifically Need Rare Books?

Large language models rely in their development on vast amounts of text that are difficult to fathom in scale. These companies have already exhausted most of what is freely available on the internet in terms of articles, forums, and websites, prompting them to search for new sources of high-quality data.

Here is where the value of rare books emerges, especially those whose editions have gone out of print or that are impossible to find in digital copies. These works represent reservoirs of knowledge not yet consumed in training processes, and they provide linguistic and stylistic diversity that is hard to find in repetitive digital content.

The Risk of "Model Collapse" and the Importance of Human Content

There is a deep technical reason behind this interest in old texts. Books published before 2022 carry a practical guarantee that they were written by human hands and were not generated by AI models. This is a highly sensitive point in the world of language model development.

When models are trained on texts produced by AI itself, they become susceptible to a phenomenon known as "Model Collapse," a condition in which output quality gradually deteriorates as synthetic content accumulates in the training data. The problem can be summarized in the following points:

• Repetition of linguistic patterns and loss of diversity in phrasing.
• Amplification of errors and biases generation after generation of training.
• Decline in the model's ability to represent rare or precise knowledge.
• Outputs gradually drifting away from natural human language.

For this reason, "pure" texts that predate the spread of automated generation tools have become akin to a scarce resource over which major companies compete.

An Ethical and Cultural Dilemma

The destruction of rare books raises questions that extend beyond the technical aspect to the cultural and ethical dimension. Some of these works carry historical value or physical rarity that is difficult to replace, and turning them into mere text data by tearing them apart may mean the loss of physical copies that cannot be recovered.

The issue is not limited to Amazon alone; other companies in the field of artificial intelligence have previously faced criticism regarding their data sources, including accusations of using pirated books without a license, reflecting a deeper crisis in how the data that feeds these systems is obtained.

Conclusion

This incident reveals the scale of the challenge facing the artificial intelligence industry in securing reliable and authentic training data. As companies race to feed their models with pure human content, questions remain open about the required balance between advancing technology on one hand, and preserving intellectual heritage and respecting property rights on the other. It is a delicate equation that will shape the features of the next phase in the evolution of these models.

✦ بقلم فريق دروب أيديا