AI binge

I just read an article from the Helsinki Times.  Second hand book sellers  across Europe are experiencing large  orders of books that are unrelated to one another. It is suspected that AI companies are buying up the books .

Of course, this leads to all sorts of questions as to how the information is being used, modified, and what is happening to the old books once  the information is processed.

A quick search provided this information on Snopes: the idea that AI companies are buying up old, hard-to-find books and then shredding them so they can scan the pages for their programs is currently rated as: "mostly true."

Snopes: Anthropic and other AI companies "destructively scanning" old books.

Recently, the one political blog I read had a commenter who likened this process to "burning the Library of Alexandria, again."

There are organizations such as The Gutenberg Project and Archive dot org that are trying to scan and preserve old (AKA, public domain) books, magazines, periodicals, and other media so that their contents become available at no cost to everyone, everywhere, for all time. I suggest you visit and support them when you can.

The article did not draw any hard conclusions. It did express suspicions. Of course, if true, many questions can be raised. If AI is not behind it, one must still ask, Why?

In my observation, Snopes is rather conservative in their speculations when attempting to confirm a claim, unless the evidence is rather overwhelming and/or substantiated by actual documentation (of course). They have a reputation that's nearly as old as public-access Internet itself. Personally, I've followed-up on crazy-ass Internet claims via Snopes for decades. I'm convinced they are reputable.

That said, I did a bit more digging. Seems like Futurism has researched this phenomenon, and the Washington Post has, as well. Apparently, Anthropic was forced to pay a $1.5 million settlement to authors, though a judge ruled that buying, scanning, and then shredding the books (ostensibly to train their AI models) was "transformative" and so a protected activity. It appears the AI companies were still plagiarizing the works of the authors to add content to their LLMs, regardless of the physical end-state of the books.

The WaPo article is hidden behind a paywall and I'm not a subscriber, so I can't read their article. But it's referenced by Futurism in theirs.

And the New York Post reported earlier this month that Amazon was training its own book-writing AI the same way, citing other companies such as Anthropic and Meta.

Stephen B. Thomas wrote:Apparently, Anthropic was forced to pay a $1.5 million settlement to authors ...

Actually that's billion, with a B. It's a token amount.

I find it hard to trust any system that is has, as its bedrock foundation, the wholesale theft of intellectual property.

You can't teach a leopard to change its spots. My 2c.

If a book collector spent their life obtaining old and rare books and then kept them locked up in their private library and then the books molded when they got old and couldn't care for the collection -- and the books ended up lost to time, how would you feel?

In general, in our collective western culture, we believe people have the right to spend their money on objects, particularly when available to the open marketplace. And that they have the right to do whatever they want to with them. In the scenario above, I think most people would either shrug it off, not really caring, or think it's a shame, and also move on.

I would rather Project Gutenberg get, scan, and present the books for free to all, for all time...but if they can't do that, the projects described here seem better than the scenario I painted at the top and maybe better than the books sitting in old shops where they're not doing anyone any good. At least their contents are being preserved even if they're currently in private hands.

*I* would personally support giving the library of congress ten billion dollars a year to scan and serve all the written works on the planet, but it's apparently not a popular priority.

The problem is the completely opposing use cases for Project Gutenberg, which scans the books and makes the text freely available, and the LLM trainers, who are only interested in ingesting the text for training their models. The source is preserved in the first example but turned into the semantic equivalent of Pink_slime in the second...all the meaning and context reduced to token representations that merely assign a statistical weight to the the order of combinations of letters.

It's not just the physical destruction of the rare books. It's the annihilation of the very meaning of the texts by feeding them into the models and reducing everything to numbers in a predictive algorithm.

i just came across the AI researcher [specializing in critical thinking and the importance of humanity in the age of AI] who has talked about this at length (wanted to share the other day but couldn't remember her name. She's got a substack, youtube, etc but I only have accessed her instagram content. https://www.youtube.com/@iamshaeo ).
this is an instagram post, the comments are really much more informative than the actual post because they get right into the discussion about the books. https://www.instagram.com/p/DbYO4HGsurg/

and also https://www.instagram.com/p/DbYBT3AshmI/, where they discuss this happening as early as 2023 (we're only hearing about it now) and some reasons why.

I agree with her general takeaway: the big overarching challenge is that DO books need to be digitized, but for everyone's use. Gutemberg and academic collections. A book that is rotting in a private collection or never gets accessed on a stack is not going to be of use to humanity. The more that get scanned and put into a publically accessible format, the better. Supporting libraries and researchers who are doing this work makes sure it stays in the public domain.