We are becoming increasingly dependent on computers and automation, both in and outside work, while systems are becoming increasingly advanced and thereby more difficult to understand.
Even experts in their fields may have difficulty understanding why an AI or automated system acts in one way or another. When an increasing number of functions are automated, this can cause problems in collaboration with humans. It all comes down to trust.
Adaptive AI-system
A recently finalised Horizon 2020 project, MAHALO, which included researchers from Linköping University, investigated how future AI-aided automation in air traffic control can be as safe and user-friendly as possible. The researchers investigated both an explained traditional automation which behaves as the software developers intended and an adaptive system based on training AI on the individual.
“Think of Spotify or YouTube recommending new entertainment based on what you have listened to and watched earlier. You can design other systems in a similar way. The idea is that you as an individual will come to the conclusion that you understand why the system acts as it does,” says Carl Westin, researcher at the Department of Science and Technology.
To test which of the two alternatives for air traffic control automation was the best, 34 air traffic controllers were put through simulations of real situations where the degree of adapted automation varied between low, medium and high. All air traffic controllers performed the tests 18 times randomly.
Human factors
The results show that a system that gets to know the user through AI is the best way forward.
“All humans have a personality and a unique way of behaving. The advantage of AI systems is that they can take this into account. This means that we can learn more about ourselves through AI, and discover new things about ourselves and our behaviour. This is the great benefit of AI right now,” says Carl Westin.Carl Westin, researcher at the Department of Science and Technology. Photo credit Linköpings universitet
His field of research is called human factors, and his focus is on human interaction with automation from a physiological as well as a psychological perspective. He is also a commercial airline pilot, which comes in very handy in this project.
Safety above all
Air traffic control is currently an almost entirely manual decision system, where people with many years of training and experience make many decisions in a short time. Also, communication with all pilots needs to be crystal clear for air traffic to function safely.
Although the potential for automation is very high, letting AI take over this task completely is not on the cards, according to Carl Westin. The limitations do not lay with the technology, however. The issue is how much automation can be trusted.
“The consequences would be huge if anything were to go wrong. There is an enormous safety protocol to comply with in air traffic control, which means that new methods have to be tested thoroughly. There is also very little competition in the field, which slows down developments,” he says.
Digital colleagues
Carl Westin makes a comparison with the automotive industry, where competition is fierce and changes in vehicle automation are implemented much faster. For better or for worse. Progress is made, but at the same time systems that are not yet ready may be found on our roads because car manufacturers need to keep their market shares.
But change will come to air traffic controllers also, albeit slower.
“AI is on the rise everywhere, and the big hype now is digital colleagues. This means that there must be a symbiosis between man and machine. But when technological advances are so fast, it is easy to forget the human who is going to use the system.
MAHALO stands for Modern ATM via Human/Automation Learning Optimisation. The project was financed by the EU framework programme Horizon 2020 and coordinated by the consultancy company Deep Blue. The project partners were Linköping University (the Department of Science and Technology, the Department of Computer and Information Science), Delft University of Technology, the Center for Human Performance Research and the LFV Group – Air Navigation Services of Sweden.
Read more about the project at
Read more about projects within the same research area at
