Deepseek should be a wake-up call. It's time for Norway to develop its own AI.

The Chinese language model Deepseek has sent shockwaves through the artificial intelligence (AI) environment and the US stock market. Over $1,000 billion has been shaved off tech companies' stock values.

The incident, referred to as AI's “Sputnik moment”, increasing the risk of an unfortunate escalation of AI development and further politicization of the most important technology developments of our time. It is essential that Norway sets out a course that is in line with our values.

We need control

Deepseek is impressive. The company has created a language model on par with the best models from OpenAI, Anthropic and Google, despite having only a fraction of the crew and computing power. The Last thanks to US export controls.

However, the Chinese language model should be a wake-up call also for politicians in the West. It shows that top-level AI is possible with limited resources, but also that Norway should be wary of entrusting the development of IT to American or Chinese companies.

We need democratic control over AI systems that are going to permeate all basic infrastructure.

Energy-efficient models

The Chinese breakthrough indicates that AI development will accelerate from an already insane level. One factor is that the US technological, political and military elite are determined that the US must ensure dominance in the AI race.

More energy-efficient models also reduce an important bottleneck on the road to artificial general intelligence (KGI), i.e. AI systems that can perform almost all human economic activity that can be performed using a computer.

It is the power of calculation that is The biggest barrier towards further AI development.

The cheaper AI systems become, the faster they will also be adopted by profit-driven companies.

The next step in development are so-called AI agents, who can carry out larger projects over a longer period of time, as well as take initiative and carry out operations in the real world.

There is some uncertainty around how much more energy efficient Deepseek is compared to models like Anthropic's Sonnet. However, we see a clear trend towards more energy efficient models. The push towards more efficiency will only increase the better the AI agents become.

An important consequence of Deepseek, therefore, is that we are increasingly moving closer to super-intelligent AI systems like our society don't have a clue how to deal.

Can be abused

Already under President Joe Biden, there were few meaningful regulations of the most powerful AI models. With Donald Trump, we can only forget that the United States will put the brakes on its own development for the sake of world security. Answering how other states should be able to put brakes on American companies is not easy.

Perhaps our best hope is that the Oil Fund and other big investors Taking action where states fail.

Deepseek has unfortunately also made the road to KGI much more dangerous. That the company chooses to post parts of the recipe on the internet is a delight for many. This makes it possible for Norwegian and other researchers to create their own models without billions of dollars of investment in computing power. But it also allows all the world's villains to access the most powerful AI systems in existence.

The fear that powerful AI systems can abused by those with evil intentions, It has become more real.

There is some positive news, too. Although Norway is not China, the small environment behind Deepseek shows that it may be possible to create outstanding AI models here as well. The models also show the need for more democratic control over AI development.

Deepseek is surprisingly good, but unfortunately has built in the limitations of the Chinese censorship regime.

If you ask Deepseek for information about the Chinese student uprising in Tiananmen Square in 1989, the demonstrations in Hong Kong or Taiwan, you will not get any answers or propaganda responses following the ideology of the Communist Party.

You might think that we can trust the Americans. But it is, for one thing, unhealthy whether the infrastructure of tomorrow is to be dependent on a bundle of American companies.

Worse it could be now that we know the outlines of an unholy alliance between the undemocratic Trump regime and the tech sector in the United States. It is not inconceivable that both American and Chinese language models could be used to threaten democracy, in the United States and other countries.

A unique opportunity for Norway

Norway has for many years been a leading country in the world in terms of digitalization in the public sector. Not every country is privileged to be able to communicate (relatively) easily through services like Altinn, use BankID for most things and get your tax return done.

In Norway, we have a unique opportunity to implement language models and AI agents in a number of public services. Language models can provide effective guidance to your tax return, help understand and fill out public forms, start a sole proprietorship, or provide you with the right health information.

For this to happen, the Norwegian models must be controlled by the Norwegian democracy and open to the public. It is possible to imagine a Norwegian model that is integrated into public services and is under democratic control.

A good infrastructure for Norwegian AI will be able to contribute to an AI transformation that respects the values of Norwegian society while keeping pace with development.

It will give Norwegian citizens greater control over how AI replaces tasks in the labour market, how we should learn from and protect Norwegian citizens' data (which trains the model), and what values and tendencies it should be trained in (through human feedback). The flaws and shortcomings of the model (s) may be the subject of public discussion.

Norwegian AI will be a safer bottom plank in Norwegian public infrastructure than machinery developed and managed from Silicon Valley or Beijing. Deepseek shows that the resource constraints to make this happen are less than we think.

To the extent that Norway is too small to develop its own AI models, there are more possibilities if we look to the Nordic countries and Europe. A Nordic cooperation on the development of AI has already been initiated. In the EU they want to make a Cern for AI.

Norway and Europe have the money. But it requires that we prioritize what will be the most important technology of the 21st century. Such initiatives need to be followed up with money and ambitious political investment in AI infrastructure. Artificial intelligence is too important to be left to oligarchs and party dictatorships.

‍