Announcing a new company I'm launching today - Great British AI.
For the last year or so, I’ve argued that AI companies should treat creators fairly, by licensing the training data they use. I come from the UK, where this is currently required by law. But today, the UK government announced that it intends to change the law, and make it legal for AI companies to train on people’s copyrighted work without a licence, to attract more AI companies to its shores.
So I’m launching a company that will take full advantage of this, and that I hope aligns with the government’s intentions.
It’s called Great British AI, and our mission is to replace the country’s creative workers with AI. The UK’s creative industries are responsible for 5% of GDP, worth £124 billion, which represents a huge opportunity for AI companies that can train on those industries’ work with impunity in order to replace them. We agree with OpenAI’s former CTO when she said, “some creative jobs maybe will go away, but maybe they shouldn’t have been there in the first place” – and we intend to help achieve this dream. The government’s proposal gives us one of our key resources — training data — for free, bringing the dream well within reach.
Our strategy is simple. We will immediately start scraping British creators’ copyrighted work from the internet (it is publicly available in huge quantities, since creators and media companies have been putting it online for years in order to gain exposure). We will do this across all media types — illustration, photography, writing, news, video creation, and more. We will train cutting-edge generative AI models on the work we scrape; these models will then be able to generate vast amounts of new works, which bear enough similarity to the works we trained on to be able to compete effectively in those same markets; and we will sell access to these models, giving customers a far cheaper alternative to licensing the original works.
By doing this, we will rapidly subsume much of the existing market for creative works in the UK, replacing an expensive ecosystem of creators with a much cheaper, fully AI-driven alternative. We will do so without paying the creators whose work we’re using to replace them. And this is all made possible by the government’s proposed changes to copyright law.
The government proposal gives rights holders the chance to opt out of AI training — but we’re not concerned about this. All the available data suggests that take-up for opt-out schemes of this nature is low (<10%), and we expect numbers here to be similar — after all, most creators won’t realise they can opt out, and for an individual to opt out all of their works will be a huge administrative burden.
What’s more, even creators who do try to opt out will no doubt have licensed their works to other settings (for instance, a photographer licensing their work to be used in an ad), and the chances they will be able to opt out all these downstream copies of their works are vanishingly small — so we’ll be able to get most of the data we want anyway.
And there is nothing in the government proposal that says how quickly we have to retrain and replace our models when creators opt-out of training — so, even after people opt out, we will continue to use models already trained on their work for years. Thankfully, the opt-out proposal really presents essentially no barrier to our scraping and training efforts at all.
Why build this business in the UK? Because it is the perfect home for it, with the proposed update to the law.
In the US, the ‘fair use’ copyright exception is nuanced, and the effect that copying has on the potential market for and value of the original work factors into legal decisions regarding whether certain uses of copyrighted works are legal or not. It is hard to see how most unlicensed generative AI training could be legal under US law, given that it competes with the work it’s trained on — and we imagine many of the outstanding lawsuits brought by rights holders against AI companies in the US will be successful as a result.
But the UK’s proposal brings no such issues. In the UK, it will quite simply be legal to train on copyrighted work without a licence, as long as you respect an opt-out scheme that, as discussed above, will in reality have very little effect. This is a huge opportunity for companies like ours that aim to replace human creative labour by ingesting, remixing and regurgitating the results of that labour.
Of course, we will have competitors. With this change to the law, we will be far from the only AI company adopting this strategy. Once training on copyrighted work in the UK without a licence is legal, every AI company will be incentivised to scrape as much British copyrighted work as they can, and train and release models, as quickly as possible. The commercial opportunity in replacing human labour in the creative industries is huge, and the inevitably slow pace of people opting out will reward fast movers. Other companies may not be as open as us about their strategy and motivations, but we have no doubt that they are making similar plans — they would be crazy not to. Competition will be fierce, but we will work tirelessly to grab as much of the UK’s creative output as we can, and automate away as many of the UK’s creative jobs as possible.
And we don’t intend to stop at automating the creative industries. In the long term, our vision is much bigger. We agree with OpenAI when they say that AI tools will be “economically valuable to companies in part by automating roles”, and will be able to replace people, and with the 41% of senior executives who expect to decrease the size of their workforce thanks to AI. We believe that much of the labour market is up for grabs. The UK’s proposed change to the law is a clear signal: AI investment, and the automation of labour irrespective of outdated rights, is a future that we as a country are proud to embrace. We expect legislation to continue to reflect this, and we are sure that the government will present more opportunities to automate different sectors over time.
