Why are AI companies destroying rare books?

As City AM reports that AI companies are destroying old and rare books whilst using them to train their models, Christian May sets out the pros and cons of the new, transformative technology. 

After court documents revealed Anthropic bought, scanned and destroyed millions of physical books as part of a controversial project to feed an insatiable demand for training data, is AI a net bonus or hindrance to humanity? 

Week in Business is also available in video format here. 

Transcript

SPEAKER_00 0:00

This is a special edition of the Week in Business because my children are here just off camera. Don't want to hear from you, don't want to hear from you. They're here because it's the summer holidays and childcare is a bit bit difficult at this time of year, and so they're in City AM. And I've just been having a conversation with them about AI, which I think is going to be the topic of the weekend business today. Because I keep as close an eye on it as I can and I oscillate almost day to day, certainly week to week, between the utopian and the dystopian idea of what our AI future could look like. And the case for the optimistic side of the ledger is quite clear. There'd be extraordinary productivity gains, incredible scientific breakthroughs, all sorts of great equalization of access to knowledge. It's all very uplifting. But unfortunately, the items on the negative side of the ledger do weigh a little bit heavier. And I'm just talking about a few things like civilizational collapse, uncontrollable cyber attacks, the emergence of extraordinary levels of inequality, possible hacking and nefarious use, AI-powered authoritarianism. So there's that to weigh up as well. Not very cheerful for my kids to sit through that. And the the the godfathers of AI themselves recognize this. Elon Musk still says there's between a 10 and 20% chance of AI leading to civilizational collapse. But his his response to that, now that he is running a publicly listed AI company, is that we should focus on the uh on the upside, uh, as he called it. Um and uh so that's what he chooses to do. Um the other godfathers, Sam Altman, Dario Modi of OpenAI and Anthropic, respectively, have been very candid um in recent months about their own fears of misuse of the technology or where it might lead or what are the societal and economic implications of it. Um OpenAI's case, of course, is very interesting given that they uh are still uh recovering, if you like, from uh what must have been quite a chastening experience of uh testing their latest model in a secure environment with absolutely no possible contact with the outside world, set this AI a series of tasks, and in order to complete the tasks, the AI found an impossible way out of its secure testing environment and hacked into all sorts of uh other businesses to try and complete the test, which Sam Altman described as unprecedented and extraordinary, uh, before saying he expects this sort of thing to happen a lot more in the future. Um, as for Anthropic, and this brings me right up to date, a really quite extraordinary development. In uh we get the information from recently unsealed court filings in the US that show the extent to which Anthropic, and they're not alone in this, have been buying up almost the world's supply of physical books in order to train their large language models. They don't want to train it on internet copy because, with some exceptions and great news organizations aside, it's very often itself slop. And so they've discovered that scanning the pages of books is the best way to train their language models. And to do this, they've set into motion the most extraordinary logistical operation. Court documents show their objective was to acquire a copy of every book in the world, including rare and out-of-print and antique books. They have acquired almost uh well over a million of them, uh, and they take them to a dedicated facility where specially constructed hydraulic machinery rips the spines off the books, sprays open the pages, scans them, and then discards and destroys what's left. And this is being done on an industrial scale, all in the name of progress. If you asked an AI system for a dystopian vision of an AI-powered future, it would probably include that sort of industrial scale setup of all the collected works and endeavours and written words of humanity on a conveyor belt into an AI-powered system that destroys the books, ingests the knowledge, and then spits out the remnants. I find that pretty distasteful. We've spoken to some rare book dealers and antique book dealers in London who say they are getting requests from anonymous companies for massive bulk orders of rare and unusual books to feed the beast. It's pretty dispiriting stuff. So I thought I'd just give you that to end the week. How finely balanced it is for me to one day I can be full of optimism for this great uh AI-powered future. Um, but actually, I suppose my sort of resting state is a slightly more pessimistic um analysis. Um, perhaps not full dystopian at this stage, despite some dystopian examples, but uh slightly pessimistic on occasion. Anyway, I'd like to know where you stand on this issue. Um, it is fraught with debate, it is fascinating, it is the most extraordinary technology powered by the most important companies any of us have ever seen come into existence. So we do try to keep a close eye on it, and in doing so, as I say, I sway between the optimist and the pessimist, the utopia and the dystopia. But that'll do for now. Speaking of utopia, my kids are here in the office, and uh I gotta do something now to keep them busy. Uh that's it. Join me again next week. In the meantime, stay up to date and in the know um on the City AM app and on cityam.com. I'll see you next week.