The Danish Defence has established a new AI centre. I fundamentally see this as both positive and necessary.
AI will be central to how the Danish Defence carries out its missions. Not because AI is inherently interesting – although it certainly is – but because modern military systems generate vast amounts of data that must be turned into something we can actually act on.
Radars, sensors, drones, data from naval vessels, cyber data and modern platforms give us access to more information than ever before. The challenge is to process that information quickly enough and turn it into better decisions.
This is where AI can make a real difference.
At the same time, AI has become a buzzword. That is why I believe the most important question for the new centre is a very practical one:
What specific problems should AI solve?
That question may be a bit of a rhetorical device, because I recently saw a post outlining an interesting approach by Janus Lind, Head of the AI Task Force in the Danish Army. The post described different levels of AI agents, but its key point was to choose the lowest level capable of solving the task. (link)
That way of thinking makes perfect sense to me and is central to applying AI successfully in the context of the Danish Defence’s complex missions. Identify a specific use case. Choose the least complex solution that solves the problem. Put it in the hands of users. Learn from it. Then build from there.
The same applies when we begin discussing some of the more difficult questions surrounding AI.
Accountability is a good example.
Who is accountable if an AI system supports – or at some point makes – an incorrect decision?
This is, of course, an important question. But we should be careful not to treat accountability as though it was invented alongside AI. If a person makes an incorrect military decision today, we already have command structures, rules, responsibilities and processes for dealing with it. AI may change the decision-making process and create new challenges, but the fundamental discussion about accountability already exists.
The next question is whether the Danish Defence can realistically build this capability on its own. If not, how far should it go in sharing the necessary data with industry partners?
In my view, the Danish Defence has always depended on industry. Private companies develop and produce a large proportion of the platforms and technologies it uses. AI changes the nature of some of those dependencies, but the idea that the Danish Defence should be able to develop everything itself does not make sense to me.
On the contrary, I see it as a strength that this development is not taking place in isolation in Denmark. NATO, our European allies, research institutions and the defence industry are all working on many of the same challenges.
There is, of course, a balance to strike.
Classification, information security, law and military approval processes exist for a reason. Many are based on experience that it would be foolish simply to discard in the pursuit of greater speed.
At the same time, we must be willing to ask whether existing structures still manage risk in the right way; whether the challenges they were designed to address remain the same today; and whether, in some cases, they stand in the way of the decision-making tempo required in a modern conflict.
The war in Ukraine has clearly demonstrated the importance of being able to adopt new technology quickly and adapt to an adversary’s advances.
That is why I find the establishment of the Danish Defence AI Centre genuinely exciting and see it as an important milestone in the transformation currently under way across the Danish Defence.
For me, however, the measure of success will not be how many AI specialists are hired, how many models are developed or how much computing power becomes available.
It will be operationalisation.
If, in the relatively near future – perhaps within a few years – we can point to specific operational solutions in which AI enables the Danish Defence to process information more effectively and make faster and/or better decisions, then the centre will have succeeded.
That is where AI becomes interesting.
When it works in the real world.