Sovereign AI is a distraction
3 min read
Last year, the Tony Blair Institute advocated for the development of “Sovereign AI capabilities”.
To their credit, the idea has taken off in a big way. At a recent AI conference in Saudi Arabia, I met software company executives from every corner of the world, marketing “sovereign” AI solutions to other countries. There was nothing sovereign about it.
In reality, there can really be no such thing as truly sovereign AI, and it is a distraction that could undermine our technological capabilities.
No nation can create cutting-edge artificial intelligence in complete isolation. The ecosystem of AI innovation is global, dependent on international talent, research, shared technological frameworks, and complex supply chains.
Take for example, OpenAI, the makers of ChatGPT, whose efforts single-handedly put generative AI on the map. They are an American company backed by global investors. Their renowned former CTO, Mira Murati, is an Albanian-born, Canadian-educated, engineer. They have partnerships with Sanofi in France for biopharma, with G42 in the UAE for regional data, with the FT in the UK on content creation, and with the Economic Development Board of Singapore to build local ecosystems. OpenAI are truly global, and that is why they are so successful. If they were called ClosedAI they would likely be a complete flop.
Consider the basic components of AI – starting with semiconductors. Advanced chips are the product of complex global supply chains and involve sourcing raw materials, components, expertise, specialty chemicals and advanced manufacturing equipment from all over the world. To attempt to develop a fully contained domestic semiconductor industry is next to impossible.
Then consider the Large Language Models (LLMs), many which are trained on internet data. By its very nature, internet data is global and hosted on global servers, just as the most brilliant minds in AI are global, moving between research centres, universities, and tech companies worldwide.
What we need is not isolated tech development, but the development of strong guardrails, ethical and enforceable guidelines, and international partnerships that allow for AI innovation while protecting national interests. The UK led these efforts last year at the AI Safety Summit in Bletchley Park, which was a huge success, and which we must build on. The “Bletchley Declaration” was agreed to by the EU and 28 countries, including the US and China, demonstrating that international consensus on AI safety is achievable.
There are dangers in pursuing AI sovereignty - particularly reduced innovation and higher development costs. For most countries in the world, the costs of acquiring and maintaining sufficient compute power will be the first barrier to entry. Sovereign AI capabilities, if ever developed, could only ever benefit a small number of countries. The sad fact is that just as internet poverty is real, the inability to benefit from AI, or “AI poverty”, will soon become a reality too.
A better use of time is to focus on sovereign data policies, because data security is national security.
Our challenge is not to create sovereign AI, but to develop AI that serves sovereign interests through ethical, global collaboration.
PoliticsHome Newsletters
Get the inside track on what MPs and Peers are talking about. Sign up to The House's morning email for the latest insight and reaction from Parliamentarians, policy-makers and organisations.