When I read that artificial intelligence (AI) companies are destroying rare books as they feed their contents into what are called their large language models (LLMs) which power their AI applications, I was reminded of a conversation I had many years ago that seems to illustrate what is at stake. (For the record, the companies cut off the spine to give them better access to all the pages and then after scanning the pages, dispose of those as well.)
Regarding that conversation, the town where I used to live decided to give its public library a thorough rework. After it was completed, I was lamenting the aesthetic result during a conversation with someone who was a local hero for having meticulously restored historic homes and other buildings in and near the downtown area, restorations that helped propel a revival in the downtown and adjacent neighborhoods.
I observed that the exterior facing of the renovated library changed color and materials at strange intervals. The result gave the impression that the contractor more than once had run out of one type of facing and then gone to the hardware store and picked up whatever was on sale to continue the job, but still didn't have enough to finish the job and so repeated the process. My take was that the cartoonish modernism of the building exterior would quickly move beyond its sell-by date.
To my surprise, my aesthetically minded friend said he liked the result and praised the new library design. This struck me as a very odd response from someone who had dedicated his life to historic preservation. I asked him what he thought the purpose of a library is. He said without hesitation, "An information center." I responded saying that to me, a library is a "repository of memory" and that the architecture ought to reflect that.
Regarding our current topic, the destruction of rare books, the AI companies are definitely thinking of themselves as information centers and not repositories of memory. Now, part of what makes libraries "repositories of memory," of course, are the books in them. Collective transgenerational memory is critical for the cohesiveness and functioning of any society. Collective transgenerational memory includes such things as the law, religious practices and especially any religious calendar which re-enacts the key principles of any religion, literature, art, music, history, the sciences and many other manifestations of culture which preserve things learned through the generations. What we are as humans and what we can become relies heavily on what we can remember about who we've been.
For the AI companies the purpose is decidedly NOT preserving these books digitally. In fact, the companies are not at all interested in sharing the digital copies of the rare books they scan with either the public or, of course, their competitors.
Digital copies of rare books for public consumption are currently being produced by such organizations as the Internet Archive in conjunction with major libraries and are done one page at time without destroying the books. For AI companies, however, the idea is to get more "information" as cheaply as possible to put into the word blender they call AI, the output of which neither preserves the text of the originals nor presents the context provided by them. So yes, AI can act as an "information center" as my friend suggested libraries should. But it cannot preserve memory.
Memory is more delicate. It is highly contextual and cannot be randomly merged with other "information" taken from other sources and other centuries. AI, after all, does NOT think. It acts probabilistically to determine what the next word should be regarding any particular topic. It does not therefore think about the past or the context. And, in fairness, AI purveyors don't claim that it does because for them the product is the amorphous substance called "information."
But information can be mistaken, intentionally false or misleading, incomplete and downright wrong. All this can be true of books and other forms of communication, but AI is facing special problems as various state and non-state actors try to influence its output using "biased and manipulated data," and bots to "flood platforms with engineered content" and "[amplify] false narratives and [silence] opposing ones." For example, the so-called "Pravda network" is flooding the internet with Russian propaganda, hoping to influence LLMs with the sheer volume of articles and reports.
By definition the text of an existing book cannot be influenced in this manner. A book, of course, needs to be understood in its historical and social context and should be interrogated like any other piece of information. But it does not typically present a moving target the way an AI application does.
The loss of books, of memory, is a well-documented historical tragedy. Perhaps the most famous was the burning of the Library of Alexandria, thought to be the most comprehensive library in the Ancient Roman world containing the works of philosophers and scientists across the known world past and present. The library is thought to have consisted of several buildings which were destroyed chiefly by fire at disparate times from 48 BCE to 391 CE.
As we digitize more and more publications and discard the physical substrate from which they come, we should ask ourselves what is more likely to exist for our descendants centuries from now, electronically stored written materials such as ebooks or actual books themselves?
The destruction of rare books to feed AI models is on an altogether different level than the unintentional burning of much of the Library of Alexandria in 48 BCE by the forces of Julius Caesar. Today, AI companies are intentionally seeking out rare books and destroying them in the process of devouring their words to help them operate systems of doubtful veracity. It's true that even if a book is rare, there may still be more copies of it left on bookseller and library shelves. But as the whole world moves toward adoption of AI, how many more copies of the rare books that are left will be consumed in an orgy of competitive book destruction to satisfy the maw of the LLMs?
AI will not maintain the transgenerational memory contained in books. It's not designed to. The least the companies could do is to let rare books survive the ordeal of assimilation into the companies' LLMs and be returned to shelves where they can perform their age-old task of keeping our collective memories available through the generations.
Kurt Cobb is a freelance writer and communications consultant who writes frequently about energy and environment. His work has appeared in The Christian Science Monitor, Resilience, Common Dreams, Naked Capitalism, Le Monde Diplomatique, Oilprice.com, OilVoice, TalkMarkets, Investing.com, Business Insider and many other places. He is the author of an oil-themed novel entitled Prelude and has a widely followed blog called Resource Insights. He can be contacted at kurtcobb2001@yahoo.com.
No comments:
Post a Comment