Latest / 7am / AI is eating the world’s books
Transcript
- 0:06Speaker 1: That's the sound of a book being scanned to train AI.
- 0:09Speaker 1: This one remains intact. Others face a more brutal fate.
- 0:13Speaker 2: Cut the spine off it, rip the pages out, scan them,
- 0:18Speaker 2: throw it away.
- 0:19Speaker 1: Nobody knows how many rare old books have now been
- 0:21Speaker 1: bought and destroyed to feed the AI beast. But second-hand
- 0:25Speaker 1: booksellers around the world have reported anonymous orders ranging from
- 0:28Speaker 1: a thousand to a million books at a time. And
- 0:32Speaker 1: internal documents from AI giant Anthropic have revealed plans to, quote,
- 0:36Speaker 1: destructively scan all the books in the world. I'm Ruby Jones,
- 0:44Speaker 1: and you're listening to 7am. Today, author Sarah Wilson on
- 0:49Speaker 1: the destruction of knowledge to make artificial intelligence. It's Monday,
- 0:56Speaker 1: September 28th. Sarah, I thought we could begin with this
- 1:04Speaker 1: bookseller who became suspicious about who was actually putting in
- 1:09Speaker 1: orders for the rare books they owned and ultimately managed
- 1:13Speaker 1: to find out quite a lot about what was going on.
- 1:16Speaker 3: Tell me about what happened.
- 1:18Speaker 4: Yeah, so there's this kind of network of rare booksellers
- 1:21Speaker 4: and they'd been getting these large orders for rare and
- 1:25Speaker 4: unusual books, but specifically ones that were older or were
- 1:28Speaker 4: published pre-2022.
- 1:33Speaker 3: One particular bookseller.
- 1:34Speaker 4: Teamed up with a media organisation, an independent media organisation
- 1:38Speaker 4: called 404 Media. I had set up the AirTag, so
- 1:42Speaker 4: it was linked to my phone and I could see
- 1:45Speaker 4: where it was going. And they put an AirTag in
- 1:49Speaker 4: one of the books and tracked this book all across
- 1:51Speaker 4: the US.
- 1:52Speaker 3: But it left California from an airport. Next up, it
- 1:58Speaker 3: appeared in an airport in Wisconsin.
- 2:02Speaker 4: It was in.
- 2:03Speaker 3: A warehouse near the airport.
- 2:04Speaker 4: Until it finally arrived at an Amazon warehouse in Las
- 2:09Speaker 4: Vegas or the Nevada desert. And this particular warehouse has
- 2:14Speaker 4: a logo. It's a picture of a Tyrannosaurus rex attacking
- 2:17Speaker 4: a book. And what they were able to find is
- 2:20Speaker 4: that Amazon workers had been sharing information about what was
- 2:24Speaker 4: going on in this particular warehouse. They'd been charged with
- 2:28Speaker 4: removing the spine of each one of these books... so
- 2:31Speaker 4: that the pages could be easily photocopied, scanned and then
- 2:35Speaker 4: digitally uploaded. And the allegation is that these pages were
- 2:39Speaker 4: being used to feed the LLMs, the large language models
- 2:44Speaker 4: for Amazon.
- 2:50Speaker 1: And we've since learned that this isn't isolated to one
- 2:53Speaker 1: bookseller or a few booksellers or even one tech company.
- 2:56Speaker 1: It turns out that Anthropic has had this secret plan
- 3:00Speaker 1: as well to buy and then destroy large amounts of books.
- 3:04Speaker 1: So tell me about that project that they were calling
- 3:07Speaker 1: Project Panama.
- 3:09Speaker 4: Yeah, once again, it's secretive. It was only discovered through
- 3:12Speaker 4: court filings for another court case, a related court case
- 3:17Speaker 4: that Anthropic was going through for piracy. And so there
- 3:21Speaker 4: were a bunch of authors who did a class action
- 3:24Speaker 4: seeking compensation for piracy. And in the process of that
- 3:28Speaker 4: court case, it was revealed that they were also engaged
- 3:31Speaker 4: in this, what's called destructive scanning.
- 3:33Speaker 2: It's the largest copyright settlement in American history. The AI
- 3:38Speaker 2: company Anthropic forced to pay out a whopping $ 1. 5 billion
- 3:42Speaker 2: after a judge said it illegally pirated more than 7
- 3:45Speaker 2: million digitized books to train its AI model.
- 3:48Speaker 4: And the court documents that were revealed actually found that
- 3:52Speaker 4: the aim was to destructively scan all books in the
- 3:56Speaker 4: world for this particular project. But once again, Ruby, we
- 3:59Speaker 4: only found out about it by accident via another court
- 4:04Speaker 4: case with a bunch of authors, which, by the way,
- 4:06Speaker 4: I was implicated in. I am one of the authors
- 4:09Speaker 4: in this class action against anthropics. So this now seems
- 4:13Speaker 4: to be a widespread way to be able to scan
- 4:17Speaker 4: books without actually having to pay authors copyright. It's far
- 4:21Speaker 4: cheaper to buy a copy of a book, get it
- 4:23Speaker 4: shipped to some big warehouse somewhere and scan and destroy them.
- 4:28Speaker 3: And so why destroy them though?
- 4:30Speaker 1: So the books are being used to train AI models,
- 4:33Speaker 1: but why then go ahead and get rid of them afterwards?
- 4:37Speaker 4: Well, it's simply easier for them. That's the real answer.
- 4:42Speaker 4: So there is some speculation that in this court case
- 4:45Speaker 4: with Anthropic, that the fact that they destroyed them and
- 4:48Speaker 4: then replaced them with a digital copy, lended things to
- 4:51Speaker 4: a fair use defense. But I should just point out
- 4:54Speaker 4: that the court in the US did declare it fair
- 4:58Speaker 4: use to buy a book, scan it for training purposes,
- 5:02Speaker 4: and then to destroy it. The destruction is because they've
- 5:05Speaker 4: already destroyed it. They've taken the spines off. And what
- 5:07Speaker 4: can you do with a whole heap of leaves of
- 5:09Speaker 4: a book?
- 5:10Speaker 1: And you mentioned that all of these books that were
- 5:12Speaker 1: being bought and destroyed in this way, they were written
- 5:16Speaker 1: before 2022.
- 5:18Speaker 3: So tell me about why that is.
- 5:20Speaker 4: Yeah, it's significant because that was when books and materials
- 5:24Speaker 4: started to have AI content in them. And so for
- 5:28Speaker 4: AI to keep on learning, it needs to avoid recursive learning.
- 5:32Speaker 4: This is the idea where it's actually sort of like
- 5:34Speaker 4: the idea of a photocopy of a photocopy of a photocopy,
- 5:38Speaker 4: you know, 10 photocopies in. the picture is deranged and unusable.
- 5:42Speaker 4: That's what happens if AI feeds on its own content.
- 5:46Speaker 4: So to guarantee that it's not getting its own recursive content,
- 5:50Speaker 4: it has to get material from before AI was actually
- 5:53Speaker 4: introduced into the world, which is 2022.
- 5:55Speaker 2: Okay.
- 5:55Speaker 1: And so what happens then when these companies run out
- 6:00Speaker 1: of books.
- 6:00Speaker 3: To use in this way?
- 6:02Speaker 4: Yeah. Well, this is the interesting question, right? So They've
- 6:05Speaker 4: essentially scraped the internet of most of the content that
- 6:08Speaker 4: they're able to use. They're now working their way through books.
- 6:12Speaker 4: They have actually committed at this stage, I should say,
- 6:15Speaker 4: to not using books more than a hundred years old.
- 6:17Speaker 4: They're actually not that much interested in it. They actually
- 6:19Speaker 4: want modern books. And it's not even something to be
- 6:22Speaker 4: flattered about that my books were used because they're actually
- 6:25Speaker 4: really kind of trying to get hold of weird like pamphlets,
- 6:28Speaker 4: like the sewerage techniques of greater Chicago region in the 1980s.
- 6:31Speaker 4: Like They're wanting really obscure stuff because they're wanting to
- 6:35Speaker 4: extend the learning capacity of their technology. But when they
- 6:40Speaker 4: run out, that is a very good question. And it
- 6:42Speaker 4: looks like they're going to run out in about 2028.
- 6:44Speaker 4: That's what the modelling is suggesting. After that, they can
- 6:47Speaker 4: actually have to turn to real life authors and writers
- 6:51Speaker 4: to create more content, presumably. So it's a very real
- 6:55Speaker 4: risk because when they do run out, their AI actually
- 6:58Speaker 4: kind of starts to grind to a halt. And I
- 7:00Speaker 4: think it's quite ironic that they're going to need writers. Now,
- 7:03Speaker 4: this has also happened with philosophers. Philosophers are now earning
- 7:07Speaker 4: sort of six-figure incomes working for these tech companies because
- 7:12Speaker 4: they're needed to navigate the moral maze that the technology
- 7:16Speaker 4: has created for humankind. And it's quite interesting, isn't it,
- 7:21Speaker 4: that it's actually these thinkers, writers, artists, creatives... The people
- 7:27Speaker 4: are the very thing that AI can't create and is
- 7:29Speaker 4: going to have to rely on going forward.
- 7:36Speaker 1: Still to come, why Sarah accepted money from Microsoft to
- 7:39Speaker 1: use her books for AI training.
- 7:49Speaker 3: So you mentioned... that you were part of this case, Sarah.
- 7:52Speaker 1: So before Anthropic was buying up these books, it was
- 7:55Speaker 1: using these pirated.
- 7:56Speaker 3: Online libraries to train its models.
- 7:59Speaker 1: Your books were part of those libraries, and then you
- 8:01Speaker 1: were part of this group of authors who ended up
- 8:03Speaker 1: settling with Anthropic for, it was $ 1.
- 8:05Speaker 3: 5 billion.
- 8:07Speaker 1: I mean, when you first found out that Anthropic was
- 8:10Speaker 1: using your books in that way, how was that realization
- 8:13Speaker 1: for you?
- 8:14Speaker 4: I wasn't shocked, to be honest, because I'd been witnessing
- 8:16Speaker 4: this kind of thing happening over the years. My content
- 8:20Speaker 4: had been used illegally without my authorization for all kinds
- 8:24Speaker 4: of things. There's publishers in Iran, for instance, that take
- 8:28Speaker 4: my books, scan them, translate them and produce them as
- 8:31Speaker 4: physical books and just put them out to market and
- 8:33Speaker 4: make money from them. There's nothing I can do because
- 8:35Speaker 4: they're not part of international copyright laws. I know that
- 8:40Speaker 4: my work has been all across various LLMs. for many
- 8:44Speaker 4: years because people can go in there and say, please
- 8:47Speaker 4: write this in the style of Sarah Wilson, and it
- 8:49Speaker 4: spits out literally the kind of phraseology that I've been
- 8:52Speaker 4: using for years.
- 8:53Speaker 3: So it's in there.
- 8:55Speaker 4: Authors can also go into some of these pirate libraries
- 8:58Speaker 4: and do a search and can find their work in there.
- 9:00Speaker 3: So it wasn't a surprise.
- 9:02Speaker 4: There's also about 75 cases that are actually out there
- 9:06Speaker 4: at the moment, various authors suing other companies. So this
- 9:11Speaker 4: is just the first one to come through with a settlement.
- 9:14Speaker 4: And it's also the largest in history. Okay.
- 9:18Speaker 1: And after that class action, you were invited to opt
- 9:21Speaker 1: into a licensing deal with Microsoft, which would allow them
- 9:25Speaker 1: to use your books.
- 9:26Speaker 3: To train AI.
- 9:27Speaker 1: There were some authors who refused to do that on principle.
- 9:32Speaker 3: You decided to do it, to take the fee, though.
- 9:34Speaker 3: Tell me about that decision.
- 9:35Speaker 4: Yeah, I did think about it for quite some time
- 9:38Speaker 4: and the bulk of authors did take the fee. And
- 9:40Speaker 4: I share their thoughts, but this issue is massive. There
- 9:45Speaker 4: are multiple LLMs out there who are taking our content anyway.
- 9:50Speaker 4: So we should take whatever money is being put forward.
- 9:53Speaker 4: People should be paid. We get paid licensing fees if
- 9:56Speaker 4: somebody wants to use our work for a PowerPoint presentation
- 10:01Speaker 4: or in schools for lectures and this kind of thing. Now,
- 10:04Speaker 4: of course, the devastating issue is that the books aren't
- 10:07Speaker 4: just being used by university students in a lecture. They're
- 10:11Speaker 4: being used to train students a system that could replace writers, artists, creators.
- 10:18Speaker 4: That is the bigger issue here. Now, is it going
- 10:20Speaker 4: to be fixed by authors refusing to take the money
- 10:23Speaker 4: that they can? You know, there are principled decisions that
- 10:27Speaker 4: could be made here.
- 10:28Speaker 3: Ultimately, I decided to.
- 10:29Speaker 4: Take the money because I sort of saw it and went, okay,
- 10:32Speaker 4: I'm taking money while I can, and I'm going to
- 10:35Speaker 4: put it into funding my ability to go and fight
- 10:37Speaker 4: the bastards.
- 10:38Speaker 3: And so how do you do that then? Once AI
- 10:40Speaker 3: has gone.
- 10:41Speaker 1: Through all of these books, used all this knowledge, its
- 10:44Speaker 1: language skills have improved as a result, how do you
- 10:48Speaker 1: fight back to make sure that writers aren't made obsolete
- 10:52Speaker 1: by AI?
- 10:53Speaker 4: Yeah, there's a couple of things. I mean, I'm talking
- 10:55Speaker 4: to you right now and exposing this. We're having a
- 10:58Speaker 4: conversation about it so more and more people know about it. Remember,
- 11:01Speaker 4: we became aware of this because some rare booksellers started
- 11:05Speaker 4: talking to each other and decided to expose it, right?
- 11:08Speaker 4: This topic is now being talked about around the world.
- 11:11Speaker 4: the European Commission has had discussions about copyright laws. Here
- 11:15Speaker 4: in Australia, the Australian Society of Authors have also got
- 11:18Speaker 4: together and are lobbying the government and have done so successfully.
- 11:22Speaker 4: They've managed to ensure that there are copyright protections in place,
- 11:26Speaker 4: that they're not destroyed. And they're also pushing for ways
- 11:29Speaker 4: to get compensation for authors going forward, but also for
- 11:31Speaker 4: past compensation. So I think that is part of what
- 11:36Speaker 4: needs to be done right now. We should keep in
- 11:38Speaker 4: mind that, you know, These tech leaders love to talk
- 11:42Speaker 4: in very, very dramatic language. We've seen a lot of
- 11:45Speaker 4: that happen in the last week or so with the
- 11:47Speaker 4: whistleblowers coming out. You know, so when they say that
- 11:50Speaker 4: they're destructively scanning all books in the world, they're in
- 11:54Speaker 4: fact destructively scanning one copy of each book. It's not
- 11:57Speaker 4: like they're taking all copies of my books and putting
- 12:00Speaker 4: through an incinerator, but they are actually taking the content,
- 12:04Speaker 4: the creative content. The other thing I would say is
- 12:07Speaker 4: this is actually a sector of the broader debate that
- 12:11Speaker 4: can be legislated.
- 12:13Speaker 2: We can actually control this.
- 12:14Speaker 4: We've seen it with the music industry, the Napster case
- 12:18Speaker 4: over 20 years ago. That actually revolutionised the music industry
- 12:22Speaker 4: and provided copyright provisions for artists. We can do the
- 12:26Speaker 4: same with books. It's not difficult to do. It can
- 12:30Speaker 4: be legislated against. So the more that people know about this,
- 12:33Speaker 4: the more pressure is put onto governments around the world
- 12:36Speaker 4: I do believe something will come out of this exposure.
- 12:40Speaker 1: And as you've noted, AI companies are now hiring philosophers
- 12:44Speaker 1: to help them navigate this problem that they've created.
- 12:48Speaker 3: You've raised the prospect.
- 12:49Speaker 1: Of writers being hired when all of the books run out.
- 12:52Speaker 1: It seems back to front though, doesn't it? To create
- 12:55Speaker 1: the problem by pursuing AI and then try and get
- 12:57Speaker 1: human thinkers back in to solve it. What kind of
- 13:01Speaker 1: future do you think we are hurtling towards here?
- 13:04Speaker 4: We are hurtling to a very uncertain future. And I
- 13:08Speaker 4: think what's been exposed in the last couple of weeks,
- 13:10Speaker 4: what a lot of people are becoming aware of is
- 13:12Speaker 4: how little the tech bros themselves know about what that
- 13:17Speaker 4: future looks like. You know, they're very much stabbing in
- 13:19Speaker 4: the dark. They're asking to be regulated for pauses to
- 13:23Speaker 4: be put on the development of their technology. And the
- 13:26Speaker 4: best thing that we can all do is be very,
- 13:27Speaker 4: very aware of what's happening and And to use whatever
- 13:30Speaker 4: democratic processes still remain in place to fight for, I guess,
- 13:36Speaker 4: retaining things that are so, so, so important to sort
- 13:40Speaker 4: of human life. And I guess the worst outcome for
- 13:44Speaker 4: us going forward is something that's totally Orwellian. George Orwell's 1984,
- 13:50Speaker 4: The Mystery of Truth, wound up controlling the one document
- 13:55Speaker 4: of recorded history and was able to change it at
- 13:57Speaker 4: will to suit its party lines. Now, I don't think
- 14:00Speaker 4: we're close to that yet. There's a lot of books
- 14:03Speaker 4: out there in the world. And so far, as I say,
- 14:06Speaker 4: the tech companies are actually committing not to buying up
- 14:09Speaker 4: books that are over 100 years old. We've got to
- 14:11Speaker 4: ensure that stays in place. So yes, there's a dystopian
- 14:15Speaker 4: future ahead of us if we don't act now I
- 14:18Speaker 4: mean how many times have we heard that but I
- 14:19Speaker 4: think actually this is actually viscerally disturbing anybody who's seen
- 14:25Speaker 4: the imagery of these books with their spines being cut
- 14:28Speaker 4: off it's actually really upsetting because books are so precious
- 14:32Speaker 4: to us people preserve books we create massive monuments to
- 14:36Speaker 4: books for preserving our history, 5,000 years of mindful recording
- 14:42Speaker 4: of the human moral endeavour. And so I do think
- 14:45Speaker 4: that this particular corner of the AI travesty, we probably
- 14:50Speaker 4: will be able to ensure that it doesn't go too far.
- 14:53Speaker 1: Yeah, books are text, but they're also symbols in a
- 14:56Speaker 1: lot of different ways.
- 14:57Speaker 3: Absolutely. Sarah, thank you so much for speaking with me.
- 15:02Speaker 3: My pleasure, Ruby.
- 15:02Speaker 1: Thank you. Also in the news, the federal government says
- 15:22Speaker 1: it's working closely with OpenAI after one of its AI
- 15:25Speaker 1: agents gained unauthorised access to Australian government websites. OpenAI boss
- 15:30Speaker 1: Sam Altman and Anthropic CEO Dario Amodei have now been
- 15:34Speaker 1: called to appear before a Senate inquiry following similar incidents
- 15:38Speaker 1: involving government sites in Australia and the United States. Prime
- 15:41Speaker 1: Minister Anthony Albanese says OpenAI has questions to answer after
- 15:46Speaker 1: it took almost three months to notify the government about
- 15:49Speaker 1: the Medicare breach. And Iran says diplomacy is the only
- 15:53Speaker 1: way forward with the United States after President Donald Trump
- 15:57Speaker 1: rejected its latest proposal to reopen the Strait of Hormuz.
- 16:01Speaker 1: Tehran had offered to reopen the shipping route within seven
- 16:03Speaker 1: days if the US eased military pressure and lifted its
- 16:06Speaker 1: blockade on Iranian ports. Iran says it's still waiting for
- 16:10Speaker 1: a formal response through mediators, while Trump has called for
- 16:13Speaker 1: Iran's continued economic isolation.
- 16:16Speaker 3: I'm Ruby Jones. This is 7am. Thanks for listening.