Ending an abusive relationship with Big Tech

My response to the Anthropic settlement

Photo by PMATAS - CC BY-SA 4.0

While working as a computer magazine editor in the 2000’s, I would sometimes be offered guest articles by vice-president of something or other at this-or-that corporation. My least favourite cliché in these articles was “harnessing the power of the open source community”, and not only because of its agricultural analogy. It was the implication that a pool of free labour was available for the corporation in question to profit from, if only it could be restrained and directed, that particularly rankled.

Some twenty years later, an old-fashioned paper letter dropped onto my doormat, from lawyers acting in the class action suit against Anthropic, developers of the ‘Claude’ large language models of so-called ‘artificial intelligence’. It transpired that Claude models had been trained using a corpus of hundreds of thousands of books, among which was my 2009 work ‘Crafting Digital Media’, published in New York by Apress, an imprint acquired by German media giant Springer. This letter invited me to join the lawsuit, with the offer of sharing in a $1.5 billion US dollar settlement. Because there were so many books used to train Claude, this headline figure works out at about $3,000 per book before unspecified legal expenses.

I fully understand, as the author of non-fiction works, that this type of writing usually does not pay very well. It is embarked upon for prestige, reputation and sometimes the genuine desire to share knowledge that the author has acquired with the world. I cannot begrudge those authors who have decided to join this financial settlement in order to be compensated by the meagre amount offered. I may join them, in the hope of obtaining a small contribution towards my next vacation once the lawyers’ bills have been paid. However, I object to the settlement on principle, not just because the sums offered are derisory in return for scraping the knowledge of many years’ work by each author.

Upon reading the terms of the settlement arranged with Anthropic, which may not be a household name but has industry backers which are, I feel like I’m in an abusive relationship with Big Tech. The fact that Anthropic has passed off the work of real human authors as ‘artificial intelligence’, in exchange for ‘Pro’, ‘Max’ or ‘Team’ subscriptions and ‘Enterprise’ volume pricing, does not form part of the financial settlement offered. Instead, it seems the settlement is purely for the copyright infringement involved when Anthropic, searching for books to scrape the content of, used a book pirate site rather than paying for access via the publishers in question.

I’m all for free and open knowledge. If a book of mine was offered on a site with gratis, free-as-in-beer access, and readers with no money, or in countries with weak currencies, were benefitting from my writing, I would not be considering legal action. I might politely ask that a pirate site copy be taken down, to respect the terms under which it was originally published, but I understand that when a copy of a work exists, other copies can and will be made, both verbatim and interpreted. Digital Rights Management and the legal protection of ‘Intellectual Property’ is and always was a confidence trick, because knowledge, being immaterial, doesn’t follow the laws of physical objects. A book can be borrowed without permission, sold without compensation, or destroyed, but knowledge can never truly be ‘stolen’ while it remains available.

My objection is that I never consented for my writing to be used to train large language models which would pass off my work as their own for profit. In effect, this is cyber-plagiarism, and I hold all ‘AI’ companies, not just Anthropic and its corporate sponsors, to be guilty as charged. Nor has my former publisher had anything to say to me on the matter, not even replying to my email requesting advice on the proposed Anthropic settlement.

In using the analogy of an abusive relationship with tech, I do not intend to belittle the real harm caused to victims of intimate partner violence. Having said that, after thirty years of working in and around the tech industry, I feel like I’ve been recruited without my consent into a power play of dominance and submission where I’ve been harnessed, slapped around and left with a few dollars on the bedside table. Not only my principal abuser, but all of their friends, were in the room to watch my violation.

There is a positive outcome to the Anthropic lawsuit reminding me of what is at stake. I always had concerns that Creative Commons licensing was far too complex, particularly with its multiple, shifting licence terms that almost no-one ever reads. Decades ago, I predicted that Creative Commons would prove a goldmine for the lawyers that devised the system. I did not anticipate that the real gold would be mined by for-profit ‘artificial intelligence’ plagiarists without any compensation for creators, working around copyright by re-writing content in a massively-parallel Turing test with ever-increasing energy demands contributing to global instability. As I write, the United States has been capturing foreign oil tankers on the high seas, and bombing unfriendly regimes in the Middle East once again.

Going forward, I intend to use copyright licensing for my work which permits human readers to use, store and share this content on the devices of their choice, but explicitly forbids its regurgitation as ‘artificial intelligence’, since it is nothing of the sort. The distinction between commercial and non-commercial use is misleading, perhaps intentionally so.

If the very wealthy but supposedly ‘non profit’ Wikimedia Foundation uses my writing as a source, which it makes available to Big Tech free of charge under a Creative Commons licence, and then accepts another round of donations from the very same tech companies, is this a non-commercial use, or an elaborate work-around of the principle of author compensation? Since we are the product, is the contemporary tech industry even viable without our acquiescence to being harnessed and spanked?

I will also research whether the HTML metadata standard can be expanded to make clear when an author does not wish to make their content available for machine-learning appropriation. Whatever happened to the robots.txt file? And should noindex, nofollow be augmented with a noscrape option to make explicit when content may be permitted for indexing, but not for ripping off? Since the tech industry controls online standards through various committees, metadata which goes against its interests may never become official, but it could become a de facto standard if enough humans use it.

I thank the authors who have attempted to obtain compensation from Anthropic for its blatant use of pirated works. This is just one battle in an information war of tech against people which is far from over.