Artificial intelligence is becoming increasingly important in nearly every aspect of society, but is completely dominated by the United States and China. Leaving the field to foreign powers and large companies may entail a number of risks. Recent events have shown Europeans cannot rely on anyone but themselves.

It is also worrying that we do not really understand how artificial intelligence currently works. It is as if we put data into a black box and then an answer magically pops out. Why? We don’t quite know.

We need to understand more about what happens inside the box.

Norwegian professor Harald Martens believes that Europe should take a different path, one that provides safer, more cost-effective and more understandable AI solutions.

“Many of today’s AI systems deliver impressive results, but they do not make us smarter, and we do not always understand how they arrive at their results. This can be a problem when AI is used in critical situations,” said Martens.

For over 50 years, Professor Martens has worked with interpretable, minimalist and reality-based data modelling, based on machine learning methods developed and used in fields other than computer science, without neural networks.

Nowadays, he is a professor emeritus, still affiliated with the Norwegian University of Science and Technology (NTNU), while also working on technical artificial intelligence at the Trondheim-based company Idletechs. He is now raising concerns about developments in modern artificial intelligence.

“We need more reliable problem solving, lower energy consumption and less ‘black box’. At the same time, we need to understand more about what is happening inside the box. We must also try to learn as much as we can about the world we are living in and the system we are working on. And it must all remain under democratic control, based on our Western European values.”

Martens believes this is entirely possible.

The field of AI is currently dominated by large, data-hungry neural networks. The so-called ‘black box’ models seem easy to use, but they must be trained on enormous amounts of data. This type of ‘machine learning’ is energy-intensive, and the solution is difficult to interpret and criticize.

Martens and his collaborators have developed an alternative they call CIM-ML, or ‘Continuous, Interpretable, Minimalistic Machine Learning’. So what is it exactly?

Instead of feeding the system as much data as possible and hoping for the best, machine learning can start with the current understanding of the system to be described and the measurement techniques to be used.

Subsequently, machine learning proceeds more or less automatically, based on the flow of modern measurement data. This is the simplest possible solution, but no simpler – compressed and visually interpretable.

“We have chosen to spend more time on the groundwork to combine domain knowledge with modern measurement data. This gives us models that can explain themselves,” said Martens.

In a time when algorithms are influencing an increasing number of decisions, perhaps the most important issue is not how fast the systems are, but whether we can trust them.

CIM-ML is self-learning, self-expanding and self-correcting, and describes both known and unknown data variations. In addition, it can process large data streams on small computers.

“In a time when algorithms are influencing an increasing number of decisions, perhaps the most important issue is not how fast the systems are, but whether we can trust them,” emphasizes Martens. This involves combining modern measurement data and interpretable mathematics.

The method is currently aimed at moderately complex systems, not, for example, ChatGPT or DeepSeek’s enormous language models, where there is often more room for error.

CIM-ML is specifically aimed at professional use in fields such as the process industry, technical systems, drone- and satellite-based environmental monitoring and earth observation.

Examples include:

• Vibration sensors on turbines to distinguish sound, abnormal noise and measurement interference – before something goes wrong.
• Thermal video monitoring of a smelting furnace or an engine for oversight and optimization.
• Multi-channel drone or satellite imaging of oceans and land to identify as many observable phenomena as possible, including spectral correction of shadow problems.

So, it is not just about getting a reliable answer, but also about understanding how – and preferably why – you arrived at that particular answer.

“In critical applications, it is important to receive early warnings without many false alarms. It is also crucial that the people using the system understand what is happening,” explained Martens.

The debate about AI often revolves around who has the largest models and the most data – but Europe can choose to take a different path.

The artificial intelligence we are developing is rooted in a proud Western European tradition of democratic enlightenment.

The question is whether Europe will invest in more independent and transparent technology, or should it remain dependent on global tech giants that operate according to the arbitrary whims of transient heads of state?

“The artificial intelligence we are developing is rooted in a proud Western European tradition of democratic enlightenment. My dream for Idletechs has been to establish high-tech jobs in Norway and Europe. These must be based on secure, simple, ethical and understandable machine learning that also promotes human learning,” concluded Martens.