ExplainabilitySafetyTransparencycs.AI
Can Large Language Models Explain Flight Safety Events? A Prior-Guided Semantic LLM-based Approach
Lu Xu, Xu Li, Linjiang Zheng, Fan Li, Riquan Zhang, Jiaxing Shang
August 18, 2026
Key Takeaways
-
LLMs can help explain aviation safety incidents in human-readable terms: The research demonstrates that Large Language Models can translate complex flight data into clear explanations of pilot control behavior and underlying causes of safety events—moving beyond technical feature importance maps that require specialized domain knowledge to understand.
-
LLMs need significant guidance to work reliably in specialized domains: The study found that LLMs alone are not strong enough classifiers for flight safety analysis. The researchers had to combine LLMs with statistical machine learning (CatBoost) and inject prediction results as "prior guidance" into prompts, showing that responsible AI deployment in high-stakes domains requires hybrid approaches rather than relying on LLMs independently.
-
Data representation matters for AI reasoning about specialized domains: The research shows that converting raw numerical flight data into qualitative descriptions (through "Semantic Discretization") makes it easier for LLMs to reason about safety events. This suggests that how we prepare domain-specific data significantly impacts whether AI systems can provide trustworthy explanations.
-
Limited training data requires creative AI solutions: Because flight safety incident data is scarce, the researchers used "contrastive few-shot learning" to help the model learn from minimal examples. This highlights a practical challenge in applying AI to safety-critical domains where large labeled datasets are often unavailable.
Abstract
Improving flight safety with flight data requires not only accurate detection of risk events, but more importantly, clear interpretation of their underlying causes at the level of pilot control behavior. Existing explainable AI techniques, such as feature importance maps, often require considerable domain knowledge to translate them into operationally meaningful explanations. Large Language Models (LLMs), which excel at language reasoning, bring a promising solution to this issue. However, applying LLMs in this domain presents key challenges such as modal inconsistency, limited classification ability, scarcity of task-specific data for fine-tuning, and lack of domain knowledge. To overcome these challenges, we propose FlightLLM, a prior-guided semantic LLM-based approach for interpretable flight safety analysis. Specifically, we first perform feature engineering to address modal inconsistency, combining statistical descriptors with physically meaningful flight indicators. This representation is further processed by a Semantic Discretization module, which converts abstract numerical patterns into qualitative descriptions that are more compatible with language reasoning. In addition, since LLMs are not inherently strong classifiers, CatBoost is incorporated as a statistical expert, and its prediction results are injected into the prompt as prior guidance. A contrastive few-shot learning strategy is further adopted to compensate for limited data. Finally, we design structured prompts to embed aviation-specific knowledge into the inference process. Using hard landing, a representative risk event with complex causal mechanisms, as an anchor point, we evaluate FlightLLM on a dataset of 704 real-world A320 flight samples. Experimental results show that the proposed approach achieves competitive classification performance while generating direct and reasonable explanations for event causes.
14 pages, 6 figures, submitted to IEEE Transactions on Intelligent Transportation Systems
Paper Info
Responsible AI Topics
ExplainabilitySafetyTransparency
Links
Citation data from Semantic Scholar