Transparent and Personalised AI Support in Air Traffic Control: First Empirical Results

The air traffic management (ATM) community is committed to greater use of automation to accommodate predicted traffic growth. In particular, Artificial Intelligence (AI) is seen as a key enabler of future ATM systems. The ATM community, in turn, is actively engaged in research to help guide the design of AI in future ATM.
Against this backdrop, the SESAR-sponsored MAHALO (Modern ATM via Human-Automation Learning Optimisation) project is exploring a simple but profound question: in the emerging age of Machine Learning (ML), should we be developing AI that matches air traffic controller behavior (i.e., is strategically conformal), or AI that is understandable (i.e., transparent) to the controller? Further, what tradeoffs exist, in terms of controller trust, acceptance, and performance? The latter issue is currently receiving a great deal of research attention The growing Explainable Artificial Intelligence (XAI) movement recognizes that system transparency and understandability are key to the successful development of ‘third wave’ AI. However, relatively little attention has been paid to the issue of strategic conformance.