The recent revelation that prominent Silicon Valley tech companies are buying vast quantities of second-hand books from British dealers only to destroy them after scanning has sparked widespread concern. The primary focus of this destruction is to harvest data to train AI chatbots, a process that involves physically dismantling the books to digitize their contents. This practice, described by insiders as ‘destructive scanning', has stirred a debate around the ethical and legal implications of such actions.
Lucy Hingston, owner of Compass Books in Dartmouth, Devon, inadvertently stumbled upon this practice when she received a baffling large order from a company linked to a Heathrow-based shipping address. Conversations with fellow booksellers revealed a pattern: numerous shops across the UK were selling books in bulk, which were then shipped to the US, their pages removed, scanned, and subsequently discarded. Books on niche topics and potentially rare volumes were part of these bulk purchases, hinting at an indiscriminate approach to sourcing data.
Dan Conway, CEO of the Publishers Association, criticized the tech giants not only for their destructive handling of books but also for wider intellectual property theft, emphasizing the harm to the creative industries and professionals relying on royalties and copyright protections. This new chapter of digital exploitation follows a ‘Make It Fair' campaign spearheaded by influential figures like Sir Elton John and Simon Cowell, aiming to curb the exploitation of creative content by large tech corporations.
Anthropic, a key player in the AI industry, has been openly engaged in ‘destructive scanning' since early 2024 as part of an initiative dubbed ‘Project Panama'. Although their activities were intended to be discreet, court documents related to a substantial lawsuit revealed extensive details about their operations. Anthropic, like other AI companies, has rationalized the practice by purchasing books legally, thus bypassing copyright infringement issues under the American ‘first sale doctrine'. This legal doctrine allows the owner of a physical book to dispose of it as they see fit, although this legality doesn't address the broader ethical questions raised by their actions.
AI models like Claude from Anthropic and OpenAI's ChatGPT rely on vast datasets, which include texts from books, to respond accurately to queries. These models are trained on a mixture of modern internet content and historical texts not available online, making books that have not been digitized particularly valuable. Anthropic previously used digitized books from pirate sites for training purposes but shifted strategies following legal challenges and a hefty settlement.
The practice has not gone without its detractors within the tech industry itself. Elon Musk, for example, advocated for more preservation-sensitive methods of book scanning used by his team at SpaceXAI. Similarly, specialized scanning companies argue that removing book spines is not only faster but also provides higher quality scans devoid of shadows that normally occur from traditional scanning techniques.
Local booksellers experiencing these bulk purchases express a mix of consternation and resignation. While the sales are financially beneficial, the thought of books being destroyed, particularly rare or out-of-print editions, is troubling. Booksellers like Finn Windle at The Minster Gate Bookshop in York and Helen Bott of Treasure Chest Books in Felixstowe shared concerns about contributing unwittingly to a process that ultimately leads to the physical destruction of books.
The scenario underscores not only a technological transformation in how AI companies source data but also a cultural shift in the value placed on physical books and intellectual property. The ethical debate is particularly poignant given historical instances of book destruction for ideological reasons, contrasting sharply with the present motivations driven purely by commercial gain.
In conclusion, the widespread purchase and subsequent destruction of books by tech companies for AI training purposes reveal a contentious crossroads between technological advancement and cultural preservation. Debate continues over the legality, ethics, and impact of such practices, with stakeholders from various sectors calling for reconsideration and, potentially, regulation of how data for AI development is sourced and utilized.
Read the full post on dailymail.com
