Big Tech wants your website

This week, I added my name as a signatory to an open letter addressed to the Australian Government. The letter is composed by Australia’s key creative bodies across publishing, music and film, and it urges the Government to hold firm against the pressure from Big Tech to amend our copyright laws to make data mining for AI training easier, cheaper and, in their ideal scenario, free.

Arts Minister Tony Burke has said the government has no plans to weaken copyright, but creative industries are rightly concerned. Globally, Big Tech is spending millions on political lobbying and high among its priorities are exemptions to copyright to allow for ‘text and data mining’ without the consent of the creator. Amid all this, the government’s independent think-tank, the Productivity Commission, released a report at the end of last year that was basically about keeping up with the Jones’s. The short take was that if content licensing markets between copyright holders and Big Tech failed to emerge organically, then the Australian government should see to what degree other countries eroded their copyright laws before eroding our own.

Meddling with copyright sets a dangerous and unnecessary precedent. But, you know where I stand on this. I signed the letter, after all. So I thought it would be interesting to jump the fence for a moment and look at one of the arguments Big Tech is putting forward, because it is interesting.

It’s transformative, stupid

Prime among the arguments is that AI training is ‘transformative’ and therefore doesn’t impinge on copyright. This is actually true if we examine it from this narrow perspective. Think of it like reading. You would have read an unquantifiable amount of literature in your lifetime, all of which forges the way you think and present information to others. Some things you’ve read stay with you, while others converge and meld with the vast resources of your mind. You are free to use your knowledge and present it in your own words, which is essentially what AI Large Language Models (LLMs) do, and this doesn’t infringe on copyright.

AI training involves feeding millions of resources into the AI model. The AI doesn’t retain this information in its original form. For instance, it doesn’t remember what it has read and it cannot reproduce the text. What it does is search for statistical patterns in the information and compress them into model weights. The weights determine how strongly a statistical pattern relates to another, whether it be about grammar or facts or human interaction. The stronger the statistical pattern, the greater emphasis the AI gives to that information. Which is why LLMs are prone to hallucinations, particularly on obscure topics where dominant patterns have not emerged. It’s also why they may present bias on topics where they have been fed a high degree of misinformation or propaganda. In many ways, it’s in our interest to ensure LLMs are trained on a broad range of high-quality literature spanning all educational disciplines, creative outputs and cultures.

Big Tech argues that because the way LLMs use knowledge is transformative, they should be able to feed the AI whatever information they like, just as humans are free to read whatever we like. Fair enough, perhaps, if you obtained the information lawfully. But any man and his dog could tell you downloading pirated books is illegal, which is exactly what the major tech companies did until legal action put a stop to it. Some writers will be compensated, but with millions of works unlawfully used, many never will.

Aside from this staggeringly huge copyright infringement, the whole ‘transformative use’ argument, while interesting, disintegrates the moment we get to the part where all this is for the creation of a commercial product deployed worldwide on a massive scale.

Commercial use

To avoid further legal action, tech companies have begun to deal directly with major publishers and media giants to legally access information, which is how it should be. But there are millions of smaller online publishers that are much harder to contact and negotiate with. This is where the political lobbying efforts to obtain ‘text and data mining’ exceptions to copyright laws come into play. And it’s here we come to the lazy point of the matter. Big Tech wants to be able to freely ‘scrape’ the internet for AI training purposes because it’s just too hard and time consuming to get permission from individual publishers. Time consuming, but not impossible. For companies that need the original works of others to train their AI models, that stand to make billions from AI’s commercial use, and are willing to spend millions to avoid paying for the required literature, that’s just not good enough. It’s lazy, and it’s purely taking advantage of what Big Tech perceives to be easy targets.

As the Productivity Commission’s report shows, there are those of influence who would trade away copyrights that aren’t theirs to trade in order to look attractive to Big Tech and their theoretical investment dollars. And Big Tech wants the cheapest and easiest way forward. It’s important that the creative industries maintain a presence in Canberra amid this intense lobbying because it’s easy to be overlooked when you’re silent.

If you’re in Australia and would like to add your name as a signatory to the open letter, you can do so here.

---

Last month’s podcast reached my biggest audience yet! If you missed it, catch up here. And if you’re rolling your eyes at all this AI, I’m sorry—next month I promise to give it a break and flex on something else.

---

Thank you for Mind Flexing with me. If you enjoyed this essay, please subscribe on Substack or your favourite podcast app, comment, click the ❤️ button, or share it with someone who would appreciate it. I’ll be back in a month. Until then, keep 💪