Military Artificial Intelligence in Norway

Norway looks for niche opportunities within military AI in order to project its standing on the international stage.

King Oscar's church in Norway, right next to the russian border. Photo: colourbox.

Summary: Norway’s strategy for military artificial intelligence was published in October 2023 and complements the country’s wider efforts within digitalization and AI. Unable to compete with larger powers in the AI race, Norway is looking to find niche opportunities within military development where AI can be decisive and, in so doing, increase the country’s standing with key global allies. The Norwegians wish to build on their previous success with collaborations between the state, research and industry. However, key barriers remain including the fact that the data that modern AI systems are largely based on can prove to be unreliable and difficult to get hold of legally and in large enough quantities.

Global race for AI

Artificial intelligence is one of the world’s most hyped technologies as of 2024, and the technology behind it is predicted to have almost infinite possibilities for a variety of sectors. Although artificial intelligence did not fully break into the public’s consciousness before the release of ChatGPT in late 2022, the potential of AI has been clear for much longer. Certainly its military potential was epitomized by Vladimir Putin’s infamous statement in 2017 that whoever rules AI, rules the world.

The United States and China are widely regarded as the leaders in the AI field, with Russia eager to catch up. AI-systems such as OpenAI and Microsoft’s GPT 4o or Baidu’s Ernie Bot originated in the US and China respectively, both developed by private technology companies. The hype around such private sector innovation is also present within defense, which has led to planning for future military AI. Such planning has resulted in military AI strategies from countries such as the United States, the United Kingdom and France. Norway’s first foray into AI planning was through the North Atlantic Treaty Organization’s own AI strategy, which was revised in July 2024 from its original version in 2021.

Norway’s strategy for military artificial intelligence (Strategi for kunstig intelligens for forsvarssektoren) was published in October 2023 and is the first strategy wholly dedicated to military artificial intelligence in Scandinavia. It is part of a wider AI push in Norway, which includes the commitment that the country will be the world’s most digitalized country by 2030 and to build national infrastructure for AI.

Recreating human intelligence

The Dartmouth Summer Research Conference on Artificial Intelligence in 1956 was one of the first academic gatherings on AI and marked the beginning of the research field itself. Photo: Dartmouth / Chris Johnson.

There are several “types” of artificial intelligence, meaning that there are several technologies and ways to try to solve the challenge of recreating human intelligence – including symbolic AI, expert systems, machine learning, and neural networks/deep learning. Here we will mostly focus on symbolic AI and the breakthrough of deep learning. The idea of creating artificial intelligence was originally rooted in the idea of recreating the brain, but there were also other techniques explored to reach this goal. The “Perceptron” by Frank Rosenblatt from 1958 is an example of the first “mechanical brain” as he termed it. He used a “connectionist” approach with a type of “pattern recognition model” to illustrate some of the fundamental properties of intelligent systems in general, and this is one of the first examples of a so-called neural network (see Metz 2021, page 19).

This approach was criticized by some, including Marvin Minsky another early pioneer of the field, and subsequently “symbolic” approaches gained favor. Symbolic AI can be explained as humans defining symbols for a computer, then making explicit rules for them. Although symbolic AI (or GOFAI (Good Old Fashioned AI)) dominated much of the second half of the 20th century, neural networks or “deep learning” won out in the end; Geoffrey Hinton, now known as the “godfather” of AI, was one of the few proponents of “connectionist” approaches throughout his academic career. His company DNNresearch was acquired by Google in 2013 for 44 million USD validating the deep learning/neural network/“connectionist” approach to artificial intelligence. This approach to artificial intelligence constructs a so-called neural network made up of several neurons or nodes that are each their own algorithm, which then adjust in relation to each other according to the data that is passed through them. If you would like to learn more about the basic workings of connectionist AI which forms the basis of much of the AI systems discussed in this article and that we know today, then there are a number of free online tutorials you can take a look at.

Military Artificial Intelligence

Training an AI-system using deep learning or machine learning has – and is predicted to have – many uses in a military context. Different uses noted in the academic literature include military logistics and intelligence, both in cyber defense and attacks, as part of weapons systems, as well as several other uses (See Yde, Nielsen & Dahlberg 2021 and Skeie 2024 for more on this). These predictions on future use often disregard the fact that most AI development (even for military use) is happening in the private sector, which is another area of planning referred to in the military AI strategies of the countries referred to above. However, the focus is often on current use and implementation. There have been reports in the media of military forces using artificial intelligence systems, including AI-enabled “suicide” drones (classified as loitering munitions), AI-systems that process data for target recognition – both human and infrastructure – as well as various uses of facial recognition technology. However, this latter use of artificial intelligence in warfare is in its infancy, and quite recent. One particular example relevant to the military of Norway – and no doubt other states – is where an AI-system is trained to recognize Russian military aircraft. This is just one example of a narrow task of image recognition that is easier for an AI-system to handle accurately than a human (see picture below from the Norwegian Defense Commission’s report from 2023.)

Image of machine/deep learning algorithm being trained to recognize Russian military aircrafts from ‘Norges offentlige utredninger’ (NOU) in 2023 or the Norwegian Defense Commission, the first defense commission in Norway since 1990 in fact.

Planning for military AI: A niche strategy for Norway?

Norway’s dedicated strategy for military artificial intelligence was published in October 2023 and is the first strategy wholly dedicated to military artificial intelligence in Scandinavia. (Sweden and Denmark’s planning for military AI is mostly contained in their general AI strategies, for example.) The strategy lays out Norway’s goals, which are threefold, namely, to:

  1. identify needs and opportunities to develop, implement and utilize AI;
  2. prioritize AI where it has the largest operational impact; and,
  3. find niches of AI-use where Norway can excel and thus become an attractive partner to allies and other collaborators.

Such a strategy can be seen as a "wish list" of what a country could achieve if everything is implemented and utilized smoothly, and the first and second goals are similar to other military AI strategies, including those of the US and UK. The third goal, however, is one that stands out as Norway is a small – though highly digitalized – country on the world stage. It indicates that Norway’s technical competency could be one way of navigating the delicate balance of implementing artificial intelligence in the Norwegian military on the one hand, while simultaneously being an attractive ally and collaborator to the US as well as the rest of NATO. Norway’s size inevitably means that some degree of collaboration is necessary, making being an attractive ally that much more important. This “niche” strategy can be seen as one way of managing this balance.

Norwegian troops in 2017. Photo: colourbox.

Applying the Norwegian military AI strategy

The mere existence of this niche strategy should be seen as a response to and perception of the increasing competition between China and the US in terms of technology and economics, including military artificial intelligence. These two countries are acknowledged as the leaders in the AI race, and the AI-hype that promises endless possibilities has many states, including Norway, scrambling to catch up. Norway’s military AI strategy is one aspect of these efforts, but the three goals outlined above may not be so easily achieved. One of the more specific ways of reaching the country’s goals is to establish better data management practices in the Norwegian military.

Data drives the development of deep learning systems, and therefore the collection, sharing, labeling, storage and management of military data is a crucial step in implementing artificial intelligence where it can be useful. However, this is a lofty goal. In November 2023, the Office of the Auditor General of Norway (the audit agency of the Norwegian parliament) found that the leadership and collaboration on sharing data in the Norwegian state does not work well enough, and that there were still considerable barriers to sharing and reusing data. Barriers with respect to the public sector as a whole include GDPR legislation, questions around the actual legality of the data itself, the difficulty of working across sectors, and a lack of overview on what data exists. For the military, there is the added layer of difficulty with respect to classified data on top of all the other challenges. Another challenge to applying the Norwegian military strategy is the acquisitions process. The same audit agency of the Norwegian parliament found that the procurement process is “strongly objectionable”, the worst rating they can give. This is due to the time it takes to acquire and implement new equipment effectively. How will the Norwegian defense acquisitions process, which takes years, fare in the face of rapid AI development, which currently is measured in months?

Despite Norway’s size, there have been earlier niches in defense industry production where the country has been able to punch above its weight, which could also possibly be applied to the current AI strategy. The NASAMS Air Defense System by Kongsberg (developed in collaboration with Raytheon) and its importance in Ukraine is an example of this. This is often considered to have come about due to the so-called “trekantsamarbeidet” or the tripartite system, consisting of the Norwegian state and military, the Norwegian Defense Research Establishment (or FFI), and the defense industry. Hugin, an autonomous underwater vehicle developed firstly at FFI, is touted as another example of the fruits of this type of collaboration. It remains to be seen how the tripartite system will function with the private tech companies that develop the most advanced AI, as they are noticeably different to the traditional defense industry.

There is potential for Norway to be a meaningful player within AI due to its already high level of digitalization, as well as the apparent political will. However, the challenges pertaining to the practical reality of working “tverrsektorielt” (across sectors) and the bureaucratic barriers to defense acquisition is exemplified by the Norwegian audit office’s conclusions that neither data sharing nor defense acquisition is well managed. While Norway’s success hinges on the ability to turn wishful strategy into reality, clear barriers remain.

Exploding or suicide drone with aerial bomb. Loitering munition concept. Photo: colourbox.

Defense studies sheds light on security issues in theory and practice.  

This article is published in response to readers' interest war and digitalization.

Further reading:

Links:

Choose a topic

Revised 20.12.2024