.
The author is a Professor and the Director of Center on Forced Displacement at Boston University
Perhaps I am old fashioned. I never got a Kindle or any other gadget to read books on a tablet. Every book that I read, I want to experience the joy of holding the physical copy, of turning a physical page, of smelling the book. I love libraries and bookstores and am most at home in the presence of books. I also grew up in a family where the written word was revered, regardless of the publication. Putting the newspaper on the floor was considered highly disrespectful, objectionable and counter to the values of learning. Coming from this world the idea that physical books will be ripped apart, fed to big automated machines, destroyed and then ‘recycled’ is deeply disturbing and hurtful. But that is exactly the world we are in.
Reports from several newspapers, independent organisations and court documents suggest that AI companies (including Anthropic through its project Panama), in order to train their large language models (LLMs) are buying old and sometimes rare books in large numbers, breaking the spine of the books, taking out each page, and then feeding them to automated scanners that scan the text in high resolution. These pages are then recycled; and to make us all feel better, the companies are saying that they recycle the paper responsibly. What a relief! I was really concerned that our heritage and history would be recycled irresponsibly.
This new strategy, of course, comes after the effort to simply take the material without permission from authors, an effort that resulted in copyright infringement lawsuits and a settlement. Some would argue that this book destruction-scanning-training of LLMs enterprise is all legal, and hence OK. That may be so, but the history of things that are legal but deeply problematic is long. Ask communities that were on the wrong end of exploitation for decades, perhaps centuries. All was legal at the time. Destroying pristine environments or dumping toxic materials in waterways may not be illegal in certain places, but the act remains indefensible regardless.
Some argue that these are books that no one cares about – old, dusty volumes that were on the back shelves in used bookstores. But since when did we decide to protect only the best sellers? Is that the criteria of preserving literature and knowledge? What if someone actually does benefit from the insight, wisdom and creativity of that book. What if something that is destroyed was in fact the last remaining copy of a book that captured the ideas of a particular time. What if it was the only dictionary of a language that is at the risk of extinction? And even if this does not happen in our lifetimes, books are a cultural heritage, just as art is. At some point if the new models need art to train, will we be OK feeding original paintings from centuries ago, and be OK since no one was queuing to look at those paintings?
The internal memos and planning documents of Anthropic make the intentions quite clear. Unsealed court documents from earlier this year include internal company memos that state “Project Panama is our effort to destructively scan all the books in the world” and “we don’t want it to be known that we are working on this”. I want to emphasise the term ‘destructively scan’ here – this is hardly the reassuring language for anyone who cares about books, language, culture or heritage.
The question we must confront ourselves with is what is this for? Why must we destroy our heritage, our connection with the past, our link to creative pursuits, for newer models? Should we leave the future generations with fewer books, including those that are now rare, so that new AI models could be trained? What if this destruction does not create a better model? What if all of this is in vain? Is our fixation with a newer AI model such that we are willing to purge our history and heritage? And perhaps the most basic question we should ask, is this fixation worth it?
