Each week we bring you one of the best pieces from CapX in full, in text and audio. For more writing like this, go to capx.co. For interviews and discussion, listen to The Capitalist every Wednesday.
Don’t hand China the future of AI
Damian Green
If you spend enough time in politics you learn to spot a familiar mistake dressed up as a novel dilemma. In the 1980s, Xerox forbade the American programmer Richard Stallman from so much as fixing a glitch in his own printer, on the grounds that the code inside it was proprietary. It was this small, almost comic grievance that lit the fuse for the open software movement. By the late 1990s, developers on both sides of the Atlantic had embraced the idea that code shared, copied and improved by many hands beats code hoarded by one company. The ‘open-source’ movement took hold even while closed-source code continued to thrive.
Britain and the US have in the decades since embraced both proprietary tech and their open alternatives. In allowing open- and closed-source tech to function side by side, we’ve reaped the rewards in innovation, in accessibility and in security. Just as important, in allowing both, governments avoided the kind of regulatory capture that would have let a handful of dominant firms lock the world’s code behind a paywall. Governments face the same issue today, and I fear they are approaching it with far less clarity than their predecessors managed.
Proprietary and open AI models are advancing at a pace few governments can properly track. History offers a clear lesson for this moment: the policies that made open code the engine of Western innovation are precisely the policies Britain and America need now to out-compete a resurgent China. It would allow Western designers to build AI systems the world can actually trust, and stop a handful of dominant firms from locking the planet’s AI behind paywalls of their own making.
Clearly China already understands the importance of open source, though it puts it to rather less generous use. It opens its models not out of commitment to the free exchange of ideas, but because doing so could help wire the world’s developers into a system of its own design. Provided, that is, the West makes the mistake of doubling down on closed AI models.
The evidence is no longer in dispute. The Chinese firm Z.ai has recently released a model that competes almost on equal terms with Anthropic’s most advanced ones. Alibaba’s freely released models have already spawned real and powerful capabilities of their own. Chinese builders are now bound to this ecosystem, creating an army of loyal developers. Meanwhile the closed American model earns its owners a subscription fee and precious little else.
The nations whose standards the world adopts win far more than the nations that merely rent their software by the month.
So let us be honest about the choice in front of us. The question is no longer whether artificial intelligence will be primarily built with open or closed models. China has settled that question for us. Its ecosystem will keep growing regardless of anything Westminster or Washington decrees. The only question left is whose open ecosystem the world builds upon. Britain has already made a start through the SOFAIR lab, a collaborative project bringing together top research talent from University College London, Oxford, Cambridge and Edinburgh in a national effort to develop next-generation open-source AI technology. In combination with America’s own open-source technology projects, it could help build a partnership among our two nations strong enough to outpace China.
The UK is currently one of the main hubs of AI technology development, with more than a dozen unicorn companies offering cutting-edge services. Some of these developed their own closed models. Others, like Stability AI and Fetch.ai, used open source to develop successful business concepts. And of course Google DeepMind, which is both founded and based in London, regularly releases open-source AI models, most famously its AlphaFold model which serves as an open-source research tool for the development of new medicines.
It should be obvious therefore that the most productive approach would be to let closed- and open-source models compete on an equal footing. Instead we find ourselves discussing whether open source should even be allowed. The debate is most intense on the other side of the Atlantic, but Britain has also started feeling its impact. This spring, NHS England announced it was restricting access to open-source code, ordering repositories to be made private by default. It was a classic example of government overreaction tackling a genuine problem in the wrong way: researchers using Anthropic’s Mythos had found serious vulnerabilities in open-source software used across the NHS, and alerted NHS England. Yet the very researcher who found them warned that locking down the code would not improve security. The guidance had already gone out, though, by the time that warning was heard.
It is a familiar pattern that the loudest warnings against open models come from those with the most to lose from them: the AI mega-firms whose engineering is world-class, but whose commercial instinct is to caution against the very openness that first made Western technology the envy of the world. A rented black box must be taken on trust, full stop. An open model can be examined, tested to its limits, and run inside a secure facility where not a single byte of data need leave the room.
My advice, therefore, is straightforward. Our government, including the NHS, should lean hard into open-source technology, and persuade the American government to do the same. Thereby allowing developers to build upon it freely.
The deeper danger isn’t losing to China, though that would create its own problems. It’s dependency that blocks the kind of creative development that would allow us to reap the full potential of this new technology. The nations whose standards the world adopts win far more than the nations that merely rent their software by the month. An AI ecosystem that is growing and allows any developer, small manufacturer, or university lab to examine, adapt, and build upon is not merely safer. It is an engine of prosperity that a handful of proprietary platforms could never match.
Open standards helped put the personal computer on every desk and the internet in every pocket. They did not impoverish the firms that built the underlying platforms, they multiplied the industries built on top of them a thousandfold.
Set AI free in the same way, and the dividend will not accrue just to London or Washington alone, but to every economy with the wit and grit to build on it.
Damian Green is a former First Secretary of State and Acting Chair of the Commons Culture, Media and Sport Select Committee.
