In an effort to decrease its reliance on overseas hyperscalers, Europe has set its sights on data independence.
This was a challenging issue from the get-go but has been further complicated by the rise of AI. Countries want to capitalise on its potential but, to do that, they need access to the world’s best minds and technology to collaborate and develop the groundbreaking AI solutions that will have the desired impact. Therein is the challenge. How to create the technical landscape to enable AI to thrive whilst not compromising sovereignty.
Governments and the AI goldrush
Let’s not beat around the bush. This is something Europe needs to get ‘right first time’ because of the speed at which AI is moving. Nvidia CEO Jensen Huang recently underlined the importance of Sovereign AI. Huang stressed the criticality of countries retaining control over their AI infrastructure to preserve their cultural identity.
It’s why it is an issue at the top of every government agenda. For instance, in the UK, Baroness Stowell of Beeston, Chairman of the House of Lords Communications and Digital Committee, recently said, “We must avoid the UK missing out on a potential AI goldrush”. It’s also why countries like the Netherlands have developed an open LLM called GPT-NL. Nations want to build AI with the goal of promoting their nation’s values and interests. The Netherlands is also jointly promoting a European sovereign AI plan to become a world leader in AI. There are many other instances of European countries doing or saying something similar.
A new class of accelerated, AI-enabled infrastructure
The WEF has a well-publicised list of seven pillars needed to unlock the capabilities of AI – talent, infrastructure, operating environment, research, development, government strategy and commercial. However, this framework is as impractical as it is admirable. For such a rapidly moving issue, governments need something more pragmatic. They need a simple directive focused at the technological level to make the dream of AI sovereignty a reality.
This will involve a new class of accelerated, AI-enabled infrastructure that feeds enormous amounts of data to incredibly powerful compute engines. Directed by sophisticated software, this new infrastructure could create a neural network capable of learning faster and applying information faster than ever before. So, how best to bring this to life?
A fundamental element of openness
For a start, for governments to achieve AI sovereignty, they must think about a solid, secure and compliant data foundation. It is imperative that the data they are working with has been subject to the highest levels of hygiene. Beyond this, they need the capabilities to scale. AI involves training and retraining data while regulation is also likely to evolve in the coming years. Therefore, without the ability to scale, innovation will be stifled. That means it is imperative to have an infrastructure with a fundamental element of openness on several levels.
Open data models
Achieving sovereignty for each state will be impossible without collaboration and alliances. It will simply be too expensive and some countries do not have pockets as deep as hyperscalers. This means a strategy for Europe must not only have open data models that countries can share, but also involve clever ways of using the available funding. For instance, instead of creating a fund that many disconnected private companies can access, invest it in building a company that is specifically focused on one aspect of AI sovereignty that can be distributed Europe-wide for nations to adapt.
Open data formats
When it comes to sovereignty, it’s not as arbitrary as having open or closed data. Some data, like national security, is sensitive and should never be exposed to anybody outside a nation’s borders. However, there are other types of data that could be open and accessible for everyone which would cost-effectively allow nations to train models within with that data and create appropriate sovereign AI products and protocols as a result.
Open data verification
One of the challenges with AI is data provenance. Without standardised and established methods for verifying where data came from, there are no guarantees that available data is what it claims to be. There is no reason that a European-wide standard for data provenance cannot be agreed upon in much the same way as the sourced footnotes in Wikipedia.
Open technology
In the context of sovereignty, this might seem counterintuitive but it has been done successfully and recently with the Covid tracking app. The software ensured that personal data was protected at a national and individual level but that the required information was shared for the greater good. This should be the model for achieving AI sovereignty in Europe.
Transformative impact of open source
This is where open source (OSS) technology can be transformative. For a start, it’s the most cost-effective approach. What’s more, realistically, it’s the only way nations will be able to build the programmes they need. Beyond the money, one of the founding principles of OSS was that it was open to study and utilise with no restrictions or discrimination of use. It can be adopted and built upon in a way that suits nations while not compromising on security or data sovereignty. This ability to understand and modify software, hardware and systems independently and free from corporate or top-down control gives countries the ability to run things on their own terms.
Finally, and perhaps most importantly, it can scale. Countries can always be on the latest version without depending on a foreign country or private enterprise for licensing requirements. It allows countries to benefit from a local model but, at the same time, have boundaries on the data.
A debate we don’t want to continue
When it comes to AI sovereignty, openness could be considered antithetical. However, the reality is that sovereignty will not be achieved without it. If nations persist in being closed books, we’ll still be having this debate in years to come – by which point it may be too late.
The fact is, nations need AI to be open so they can build on it, improve it, and ensure privacy. Surely that is what being sovereign is all about?
- Data & AI
