The FLY AI Forum 2025 organised by EUROCONTROL will take on the 22nd and 23rd of April in Brussels. I attended the previous edition in 2024 and this article summarises my takeaways. To put it in a nutshell, if you think artificial intelligence (AI) will replace air traffic controllers, think again. Maintaining tactical separation between aircraft is only one of many potential use cases for AI in ATC. It is likely that AI based assistants will help air traffic controllers and not replace them.
EUROCONTROL organised a new edition of the FlyAI event about AI at the end of April 2024 and I was lucky enough to secure a seat. The two-day event was the opportunity to get to hear the latest information about the status of AI in air traffic management and let me tell you: things are moving forward on many fronts.
If you think that AI in ATC means replacing controllers with AI, think again. This could become one potential application in the very, very long term but it has so many implications and raises so many questions that it is not even close to happening. One takeaway is that AI has multiple potential applications in the ATM context and not only in the tactical provision of separation between aircraft.
The number of applications presented at the FlyAI forum goes beyond this overview but here is a short list to give you an idea:
- Air traffic controller assistants (more on this later)
- Support EUROCONTROL in their role of Network Manager, using pattern analysis to identify the impact of weather and regulations
- NOTAM interpretation and filtering according to impact, importance and relevance
- Speech recognition
- Analysis of the surveillance infrastructure for monitoring purposes.
There were more topics, many with their own sub-topics. Speech recognition for example, can be used in many ways: read-back confirmation, clearances analysis for safety and statistics, pseudo-pilot for autonomous simulation training and transcription of communication for incident investigation.
AI in ATM needs the human touch
Different AI solutions are at different levels of maturity, some are very close to being operational – or could even be in operation by the time you read this – and others are still at the research and development level.
Replacing the ATCOs with AI, however, is not for today nor for tomorrow. There are multiple reasons for this, ranging from the handling of AI failure to keeping the job interesting, but also liability and teaming with humans. This last factor triggered some interesting discussions. If AI becomes an assistant, looking over a human’s shoulder and nudging them when something is not correct, how to make sure that this interaction does not come in as a disturbance? What if AI gives a hint, that the human applies, which ultimately results in an accident? Who is liable in such a scenario? The AI can’t be liable, but its developer could? Or the human giving the clearance? There is a lot to clarify there.
When humans are trained to become ATCOs, they learn much more than maintaining separation in an efficient way. They learn communication, human interaction and so on. That being said, a human assistant can also be annoying to an ATCO, even if they tend to know instinctively when an intervention is ok or not. And if they fail at this, they usually adjust quite rapidly when getting their colleague’s feedback, just because they are humans and this is what humans do. AI will have to learn how to best interact with each human, in each situation to become a valuable “partner”.
The regulatory framework moves forward in Europe
Another less exciting but nonetheless super-important topic that was discussed is the legal framework and certification. There is progress there too and the following hierarchy from theEuropean Union was presented and discussed:
- The basic document in Europe is the European Union Artificial Intelligence Act. This document defines four levels of AI applications and abstractly defines what has to be done to make them “legal” in the UE. It is independent from ATM applications, it is valid for all uses of AI. Note that it also makes some applications like social scoring fully illegal in the EU, giving the EU organisation somewhat of a disadvantage vs. other regions.
- For aviation applications, EASA issued a concept paper, based on the EU AI Act and defines how it applies to aviation. At the moment of writing [2024], the second edition of this concept paper is in proposed status.
- Getting more specific, EUROCAE workgroup 114 is working on AI topics and should produce standards, based on the concept papers.
This progress is important to make sure that when more advanced and tactical applications of AI will become more concrete, the legal framework will be ready.
Thanks for the open discussion
The FlyAI forum was a great opportunity to get updated on those topics but also to discuss them. Time was given to the participants for raising their questions after each topic and this resulted in lively discussions.
Thanks to EUROCONTROL, not only for organising and running this forum but also for making it open and available to smaller organisations like FoxATM.
A full recording of the event is available here.
