🔗 Share this article

Around the globe, nations are pouring hundreds of billions into what's termed “sovereign AI” – creating their own AI technologies. From Singapore to Malaysia and the Swiss Confederation, states are vying to create AI that grasps native tongues and cultural nuances.

The International AI Competition

This trend is an element in a broader worldwide race spearheaded by major corporations from the US and China. While firms like OpenAI and Meta pour massive funds, mid-sized nations are additionally making independent investments in the artificial intelligence domain.

However amid such huge amounts at stake, can developing nations achieve significant benefits? As noted by an expert from a prominent thinktank, “Unless you’re a affluent state or a major firm, it’s quite a challenge to develop an LLM from the ground up.”

Security Considerations

Numerous countries are reluctant to rely on external AI models. In India, for example, US-built AI solutions have at times fallen short. A particular case saw an AI assistant employed to teach pupils in a remote community – it interacted in English with a pronounced Western inflection that was nearly-incomprehensible for regional students.

Then there’s the state security dimension. In the Indian defence ministry, relying on certain international models is viewed not permissible. Per an developer commented, It's possible it contains some unvetted learning material that may state that, oh, a certain region is not part of India … Employing that certain model in a defence setup is a big no-no.”

He added, “I have spoken to individuals who are in defence. They want to use AI, but, forget about specific systems, they prefer not to rely on US technologies because data could travel abroad, and that is completely unacceptable with them.”

Domestic Efforts

In response, some states are backing national projects. An example such project is in progress in the Indian market, in which a company is working to create a domestic LLM with public backing. This project has allocated approximately $1.25bn to AI development.

The expert imagines a AI that is less resource-intensive than top-tier models from US and Chinese firms. He states that India will have to compensate for the funding gap with talent. “Being in India, we lack the advantage of investing huge sums into it,” he says. “How do we vie with say the enormous investments that the United States is investing? I think that is the point at which the fundamental knowledge and the strategic thinking is essential.”

Local Focus

In Singapore, a public project is funding language models developed in local regional languages. Such dialects – for example the Malay language, Thai, the Lao language, Indonesian, the Khmer language and more – are frequently poorly represented in American and Asian LLMs.

I wish the experts who are creating these sovereign AI models were aware of the extent to which and how quickly the frontier is moving.

An executive involved in the initiative explains that these models are designed to supplement bigger models, rather than displacing them. Systems such as a popular AI tool and another major AI system, he states, commonly find it challenging to handle regional languages and cultural aspects – speaking in stilted Khmer, as an example, or recommending non-vegetarian recipes to Malaysian consumers.

Developing native-tongue LLMs allows national authorities to incorporate cultural sensitivity – and at least be “smart consumers” of a sophisticated system created in other countries.

He adds, I am cautious with the term independent. I think what we’re trying to say is we want to be more adequately included and we wish to grasp the abilities” of AI technologies.

Multinational Collaboration

Regarding nations seeking to carve out a role in an escalating international arena, there’s an alternative: collaborate. Researchers connected to a prominent university have suggested a government-backed AI initiative shared among a alliance of emerging nations.

They term the proposal “Airbus for AI”, drawing inspiration from Europe’s effective strategy to create a competitor to Boeing in the 1960s. Their proposal would see the formation of a government-supported AI organization that would pool the capabilities of several states’ AI programs – such as the United Kingdom, Spain, Canada, the Federal Republic of Germany, the nation of Japan, Singapore, the Republic of Korea, the French Republic, the Swiss Confederation and the Kingdom of Sweden – to establish a competitive rival to the Western and Eastern giants.

The lead author of a report outlining the proposal says that the concept has drawn the attention of AI leaders of at least three countries up to now, in addition to multiple sovereign AI organizations. While it is presently focused on “middle powers”, emerging economies – the nation of Mongolia and the Republic of Rwanda among them – have likewise shown curiosity.

He comments, “Nowadays, I think it’s just a fact there’s less trust in the promises of the present White House. People are asking such as, can I still depend on these technologies? Suppose they decide to