Leverage Large Language Models For Enhanced Aviation Safety

The modernization of the National Airspace System presents numerous opportunities for increasing the capacity for safe aircraft navigation through the exploitation of AI/ML capabilities. In 2021, the authors outlined the promise and challenges of Artificial Intelligence (AI) technology applications in The Journal of Air Traffic Control cover article, "Novel Framework to Advance Verification, Validation, and Certification of Non-Deterministic, AI-Based Algorithms for Safety Critical Applications." [1] Since 2021, the field of AI has been transformed by new breakthroughs in AI/ML technology, such as the surge in Large Language Model (LLM) technology within the field of natural language processing. This intersection of LLM technology and natural language processing affords an automated ability to interpret, analyze, translate, and even generate language. This Paper looks at this technology-application intersection in the context of air traffic control. An AI/ML-enabled aircraft/ATC alerting system, combined with enhanced positional information, can verbalize timely notifications about impending unsafe situations allowing aircraft flight crews to take proactive measures. ATC systems can also benefit from AI/ML-enabled safety monitoring to ensure all parties are informed about real-time events. [GRAPHICS] This paper presents one promising approach to explore Step 1 (see Table 1) that employs deep learning and LLMs to create a system that can process air-ground verbal transactions, detect anomalous situations, and alert air traffic controllers, thus enhancing ATC and flight crew situational awareness based on identified anomalies raised in air traffic communications. Our results indicate that cost-effective data can be generated with LLMs and then used to train a text-based Variational Auto-Encoder that can successfully discriminate between nominal and off-nominal (safety-critical) scenarios observed in air traffic communication.