LLM-Based Human-machine Teaming for Air Traffic Controllers

Abstract

Nowadays, the air traffic control (ATC) system is much more complex, and the number of situations in which controllers have to make decisions that can be assisted by intelligent decision-support systems (IDSSs) has grown. Current ATC systems are mainly based on human skills and rule-driven automation, which may have challenges in coping with dynamic operations, unstructured communication, and the growing airspace density. The advent of Large Language Models (LLMs) has brought new opportunities for human–machine teaming, such as contextual reasoning, natural language understanding, and decision-support capabilities that adjust to the context. This paper explores the feasibility of human–machine teaming in air traffic control employing LLM. The study used a qualitative conceptual research approach using literature review and thematic analysis. Existing studies on human–machine teaming, aviation systems utilizing AI, explainable AI, and the field of air traffic management have been reviewed to explore the opportunities, challenges, and safety implications of implementing LLM in an ATC scenario. The results indicate that LLMs can assist air traffic controllers in various aspects, including communication management, procedural guidance, workload reduction, and situational awareness enhancement. The research also identified important problems of explainability, automation bias, reliability, risk of hallucination, cyber security and regulatory compliance. The study recommended a conceptual human-in-the-loop framework for LLM-assisted ATC operations that focuses on enhancing controller authority, introducing explainable AI mechanisms, ensuring system transparency, and designing systems for safety. The study contributes to the ongoing conversation regarding the impact of AI systems on aviation systems and provides a conceptual framework for integrating LLM with collaborative ATC decision-support systems. The results also indicate that human oversight, safety and reliability in system operation are critical considerations for future LLM-driven ATC systems, necessitating careful and transparent integration of AI capabilities to ensure reliable and safe ATC.