Innovative AI research about military decision support

Peter Bovet Emanuel is researching artificial intelligence (AI) and its impact on future warfare. In his research, he created two AI models for decision support. One of the models is very innovative, leading to Saab applying for a patent on the technology.

Today's modern high-tech warfare increases the speed on the modern battlefield, partly due to more automated weapon and sensor systems, and partly because of larger information flows. This development puts pressure on making quick decisions, challenging human cognitive abilities.

"Analyzing and synthesizing large amounts of information in a short time requires computational power, which computers excel at. A calculation that takes a human several minutes, hours, or days to perform can be done in seconds by computers. The same applies to decision-making – what takes about seven seconds for a human can be done in milliseconds by a machine. Depending on the context, seconds spent or gained can be crucial for the outcome. By introducing AI as support for human decision-making, we can compensate for our shortcomings", explains Peter Bovet Emanuel.

He is in the final year of his doctoral studies in war studies at the Swedish Defence University and is a Lieutenant Colonel in the Swedish Armed Forces.

"I have worked in the special forces system for over 20 years and led a unit where we developed capabilities for joint operations. That is where I became interested in advanced and disruptive technologies like AI", he says.

Decision support in modern warfare

Peter Bovet Emanuel is primarily interested in how various forms of AI can contribute to addressing the challenges of high-tech warfare, especially regarding making quick decisions based on large amounts of information.

"Much has been written about the strategic importance of AI, and a lot of work is underway at the sensor and weapons levels. But additionally, it is interesting to study usage and management methods", he argues.

Testing current issues based on real scenarios

In his research, Peter Bovet Emanuel focuses on two types of AI, creating conceptual models for developing and testing the tools. He focuses on current challenges in collaboration with the Swedish Armed Forces. One of the most difficult aspects of his work was to define the exact utility of AI.

"Since we are not looking at AI in general, but rather very narrow application areas, it is important to be specific and know exactly what data to input and what results to expect. I developed several models that I had to discard along the way, either because I aimed too high or because the work proved to be too extensive for a doctoral project, ultimately coming up with two models."

AI based on optimization algorithms

The first model uses a traditional AI tool based on optimization algorithms, aimed at streamlining processes and finding optimal solutions to problems that encompass many different conditions and variables.

"This involves decision support within NATO’s joint targeting process. It is a fundamental process at the operational and tactical command level that involves identifying and prioritizing targets and matching them with suitable ways to affect the targets to achieve desired effects. It is a meticulous process and analysis method where the main goal is to optimize the use of resources," he explains.

Collaboration with IBM and Saab

In this work, Peter Bovet Emanuel collaborated with IBM for technical support in translating and adapting the model to IBM's program for solving optimization problems. The complex problems he developed can be found in dynamic contexts where multiple targets must be assessed simultaneously. Some targets have a higher priority than others, and decisions must be made about what type of weapon to use against which targets.

"Staffs skilled at this can allocate tasks to the appropriate resources even in a dynamic process, but if there are too many variables, such as multiple high-value targets, in a short time, it can become too difficult. Especially considering the trend toward increased autonomy and automation of weapon platforms. Here is where AI can be used as support in the decision-making process", he says.

Additionally, the model can be useful for investment decisions, and new weapon systems can be added to the model to analyze the results generated. By simulating different scenarios and adjusting variables in the model, adaptations can be made to the testing of new tactical and operational concepts.

Support for targeting sensors for intelligence gathering

The second model was developed in collaboration with Saab and uses AI to optimize the use of sensors to locate and monitor targets.

"In this model, we use neural networks and the latest AI technology known as ‘deep learning’, which learns and adapts as the model gains access to more data. The model was trained on large amounts of satellite images and has 'learned to understand' how topographic information affects the spread of radio signals.”

The model includes multi-step problem-solving where the neural network contributes to recommending the best locations for radar systems.

"I chose to focus on radar range, that is, defining the most suitable places to position radar systems to protect a specific object. If we switch perspectives, these places are good locations to scan with your sensors. There are several additional possibilities offered by the model."

Patent in progress

The model can be used for all types of ground-based radio signals, and the technology has a wide range of applications for further development.

"My colleague at Saab and I have not found anyone else who has done this, and Saab is currently applying for a patent on the technology."

Currently, mathematical models are used to calculate radar coverage, but these take considerably longer, and the models are and remain static and therefore cannot be used for other geographic areas.

"Our model is constantly learning, and after training on 500-1,000 images, the response time is down to seven minutes, compared to nine hours for more traditional models. Moreover, when trained on more topographic variations, the model can be used for analyses of new geographic areas. Like a lens you move over the map to generate responses for suitable grouping locations in this case."

AI can change military decision-making

Analyzing and synthesizing large amounts of data using AI may change our view of leadership and shape future concepts for command and control, says Peter Bovet Emanuel. In the long run, he hopes that research will contribute to an increased understanding of AI's growing role in military decision-making.

"Today, the predominant view is that humans are in control of and make decisions. My research challenges that view and may affect how command and control is designed in the future."

He emphasizes that his studies do not imply that we should transfer decision-making entirely to AI but rather regard AI as support in the process.

"Military decision-making sometimes needs to be almost intuitive, and if we have an AI model that can generate recommendations quickly, the commander can get a quick validation of his/her decision from an AI agent that can provide two recommendations even faster, for example."

AI performing part of the work also leads to more time for other tasks.

"It is interesting when we can work together – humans and machines. We can become smarter by compensating for our limitations with the help of AI. At the same time, we can assist in areas where AI still has limitations."

Need for more competence

At the same time, it is important to be aware of the problems AI can bring, and create expert knowledge in the field.

"We can't just buy an app and go for it. For example, it's important that we have control over the data we put into the systems so that it is correct or is aimed for a different type of operation. Just as we have legal advisors, political advisors, and other types of special staff advisors in the Swedish Armed Forces, we may also need AI advisors, i.e., personnel who really understand the sources of errors and can interpret AI-generated recommendations."

Peter Bovet Emanuel points out the need for more research in the field.

"We are not even at the threshold, and therefore we have not seen the extent of what this development may entail. There is a great need for more knowledge, both in society as a whole and within the Swedish Armed Forces", he says.