AI companies destroy physical books – let's scan rare books before it's too late

Based on 15 sampled comments from this thread. Analyzed 2026-08-21.

The thread spent less time mourning individual volumes than interrogating the legal and economic machinery behind AI training corpora. Many commenters argued that book destruction is a sensational symptom of copyright rules: companies can cheaply acquire a physical copy, scan it for model training, and then discard it while avoiding the far murkier rights attached to digital editions. They stressed that trainers need only one copy of any title, whereas publishers, libraries, and used-book channels routinely pulp far larger numbers of unwanted books. Several also doubted that genuinely unique artifacts are involved, suspecting the targets are scarce but still replaceable works. Even so, readers saw an ugly concentration of power in private firms converting hard-to-access human knowledge into proprietary training assets. The appeal of pre-2022, human-authored text was understood as both a source of cleaner training material and a way to distinguish machine-generated content. Some saw LLM access as an improvement over rare books sitting unread in collections, while others insisted that model-mediated access is not public preservation. The proposed remedy was not simply outrage at AI, but durable public digitization: copyright reform, archival exemptions, and eventual release of scanned materials through institutions such as the Internet Archive. A smaller group rejected the premise that destruction is legally required, calling it a cost-saving business choice rather than an unavoidable compliance measure.

The sharpest fight was over whether AI-driven book disposal is a serious cultural loss or a misleading anti-AI symbol. Critics argued that private model builders are swallowing inaccessible knowledge while discarding physical copies, making preservation and public access more urgent. Defenders replied that one scanned copy is negligible beside routine industrial pulping, likely does not mean one-of-a-kind books are vanishing, and reflects restrictive copyright markets more than an AI-specific assault on literature.

• Praise: Pre-2022 human-written books offer comparatively clean training material, and that provenance can help models or filters detect synthetic text rather than recursively learning from AI sludge.
• Praise: Training on a single scarce title can expose its knowledge through AI systems to far more people than a copy locked in a library or private collection.
• Praise: The controversy is pushing concrete preservation ideas, including nonprofit digitization rights and eventual public release of AI-scanned works after copyright expires.
• Praise: Several commenters see the episode as evidence that copyright reform could unlock broader, lawful access to training data and cultural archives instead of leaving them behind private legal walls.
• Concern: Destroying physical copies after digitization feels like private AI firms converting cultural memory into proprietary model fuel while the public receives neither the scans nor archival guarantees.
• Concern: Copyright restrictions on electronic editions may perversely steer model builders toward buying, scanning, and discarding print books instead of using durable licensed digital copies.
• Concern: Claims that disposal is legally necessary drew skepticism; if shredding merely trims processing costs, AI companies are choosing convenience over preservation.
• Concern: Even if most titles have surviving copies, scarce works can become harder to independently verify or access when their digitized form is captured inside closed training datasets.