Artificial intelligence (AI) is becoming an increasingly important part of modern aviation. From predicting traffic congestion and optimizing flight operations to evaluating regulatory compliance and supporting complex airspace decisions, AI offers opportunities to improve safety, efficiency, and operational performance.
However, as aviation organizations adopt increasingly sophisticated AI tools, an important question emerges: Can aviation professionals understand and trust the decisions these systems make?
In an industry where safety depends on accurate information, clear communication, and informed human judgment, producing the right answer is only part of the challenge. Aviation AI systems must also help users understand how recommendations were developed, which factors influenced an outcome, and where uncertainty remains.
This is where explainable artificial intelligence (XAI) becomes essential.
At Mosaic ATM, we believe the future of aviation AI depends on more than increasingly powerful algorithms. It requires intelligent systems that combine advanced analytics with operational expertise, transparent reasoning, and human-centered decision support.
How Mosaic ATM Is Advancing Explainable AI in Aviation
At Mosaic ATM, our approach to aviation innovation combines expertise in air traffic management, artificial intelligence, systems engineering, human factors, and software development.
Through research and development programs supporting the FAA, NASA, and other aviation stakeholders, we are exploring how advanced AI can improve operational decision-making while maintaining transparency and accountability.
Several of our projects demonstrate how these principles can be applied to real aviation challenges, including:
We describe these projects in more detail below, but first, we will explore the importance of explainability in aviation.
What Is Explainable AI in Aviation?
Explainable AI refers to artificial intelligence systems designed to provide meaningful insight into how they produce predictions, classifications, or recommendations.
Traditional machine learning models can operate as “black boxes,” processing large amounts of information and generating outputs without clearly revealing the factors behind their conclusions.
While these systems may demonstrate impressive predictive accuracy, limited transparency can create challenges in safety-critical environments.
For example, an AI-powered air traffic management tool might recommend rerouting aircraft to reduce congestion. But aviation professionals also need to understand:
- Which traffic, weather, and operational conditions influenced the recommendation?
- What constraints or potential safety risks were considered?
- How reliable is the prediction, given the available data?
- What alternative actions are available?
- How might changing conditions affect the recommendation?
An explainable system helps answer these questions, allowing operators to evaluate AI-generated insights using their own expertise and situational awareness.
Importantly, explainability does not necessarily mean exposing every mathematical calculation within a machine learning model. It means providing information that is meaningful, accurate, and relevant to the person making the decision.
Aviation Decisions Are Becoming Increasingly Complex
The National Airspace System (NAS) depends on coordination among air traffic controllers, airlines, airport operators, traffic flow managers, and numerous supporting technologies. These stakeholders must continuously interpret information about aircraft movements, weather, airspace restrictions, airport capacity, and changing operational conditions.
Meanwhile, the expansion of uncrewed aircraft systems (UAS), advanced air mobility (AAM), and increasingly automated flight operations introduces additional complexity.
AI can help manage this growing volume of information by identifying patterns, forecasting disruptions, and evaluating potential responses faster than traditional manual processes. However, increased automation does not eliminate the importance of human judgment.
Consider an AI system that predicts a capacity reduction at a major airport and recommends adjustments to arrival traffic. A useful recommendation should explain the underlying operational conditions and associated trade-offs, allowing traffic managers to determine whether the proposed action is appropriate.
Explainable AI transforms a prediction into actionable decision support. Rather than asking operators to accept an algorithm’s recommendation without context, it provides information they can evaluate and incorporate into established workflows.
Safety-Critical Systems Require Transparency and Accountability
Aviation has always maintained rigorous expectations for system reliability, verification, and safety assurance. AI introduces new challenges because machine learning models often develop their predictive capabilities from data rather than exclusively from predefined engineering rules.
As a result, traditional approaches to verifying software behavior may not fully address the complexities of AI-enabled systems.
The FAA’s Roadmap for Artificial Intelligence Safety Assurance outlines the challenges and priorities associated with establishing acceptable safety assurance practices for AI in aviation.
Although the roadmap focuses substantially on aircraft-related applications, the broader importance of understanding model behavior, evaluating risk, and developing appropriate assurance methods extends to other aviation AI use cases.
Explainability can support these objectives by helping engineers and operational experts investigate model outputs, identify unexpected behavior, and evaluate decisions against applicable requirements.
However, explainability alone does not establish that an AI system is safe. It must complement rigorous testing, validation, data quality controls, performance monitoring, and appropriate human oversight.
As AI becomes more integrated into aviation operations, organizations will need to demonstrate not just what their models can accomplish, but how those capabilities can be used responsibly.
Human-AI Collaboration Depends on Appropriate Trust
Successful AI adoption in aviation requires systems that support human expertise rather than simply replacing established decision-making processes.
Air traffic controllers and aviation professionals bring operational knowledge, contextual awareness, and experience that cannot always be captured within a model’s training data.
An AI system may identify an apparently optimal solution based on available information while overlooking circumstances known to the operator, such as an emerging operational restriction or unusual local condition.
If the recommendation is presented without explanation, users may either place too much confidence in the output or disregard it entirely. Both outcomes can undermine the value of AI.
Explainable AI supports appropriately calibrated trust by allowing users to assess whether recommendations align with current conditions and operational priorities.
As we explored in our recent article, AI Isn’t Replacing Air Traffic Controllers; It’s Giving Them Better Decisions, the most valuable applications of AI in air traffic management are those that help professionals recognize problems sooner, evaluate alternatives, and understand the potential consequences of their decisions.
Explainability is a critical part of making that vision possible. By presenting the operational factors, constraints, and uncertainty behind AI-generated recommendations, these systems can strengthen human-machine collaboration without diminishing the importance of professional judgment.
The objective is not to create systems that operators trust automatically. It is to provide the evidence and context necessary for operators to determine when an AI recommendation deserves confidence.
AI-Powered Contingency Planning: Understanding the Why Behind Recommendations
One important application of explainable AI is contingency management.
When unexpected disruptions occur, aviation operators must quickly evaluate possible responses while considering safety, operational constraints, and competing priorities.
Mosaic ATM’s traffic management contingency planning research has explored decision-support approaches that do more than recommend a single course of action.
In this work, AI-supported recommendations were accompanied by relevant operational constraints and alternative actions, giving decision-makers additional context for assessing the available options.
This approach is particularly valuable because the most efficient action is not necessarily feasible under every operating condition.
For example, a recommended traffic management strategy may reduce expected delays but introduce operational limitations elsewhere in the system. Providing insight into those constraints allows decision-makers to evaluate the broader consequences before acting.
Our work on the Contingency Planning Toolkit for Advanced Air Mobility further demonstrates Mosaic ATM’s expertise in developing intelligent services that account for real-world operational constraints.
Developed in collaboration with The Ohio State University and CAL Analytics, the toolkit includes capabilities for evaluating and responding to off-nominal situations in emerging airspace operations.
These efforts reflect an important principle: AI is most valuable when it helps aviation professionals understand their options, not merely when it identifies a preferred outcome.
EASEL-AI: Transparent AI for Regulatory and Compliance Decisions
Explainability is equally important outside real-time flight operations.
Regulatory compliance, safety assessments, and certification-related documentation often require aviation professionals to evaluate large volumes of technical information against complex requirements. AI can accelerate these processes, but unsupported or poorly explained findings can introduce significant risks.
Mosaic ATM’s EASEL-AI platform demonstrates how advanced AI can support more transparent, evidence-based evaluation. EASEL-AI combines large language models (LLMs), retrieval-augmented generation (RAG), and systems engineering approaches to analyze regulatory documentation, identify relevant requirements, and provide structured assessments.
Rather than returning only a generalized conclusion, the platform is designed to connect its findings to supporting documentation and identify potential gaps or areas requiring further review.
For example, when analyzing an application for a Beyond Visual Line of Sight (BVLOS) waiver, EASEL-AI can help reviewers identify whether particular safety considerations have been addressed and locate the relevant supporting information.
This provides a foundation for more consistent, traceable regulatory evaluations while preserving the role of qualified personnel in interpreting requirements and determining appropriate outcomes.
For Mosaic ATM, this represents a practical application of explainable AI: combining intelligent automation with the evidence and transparency necessary for high-stakes decisions.
Mosaic Fuser: Building a Reliable Data Foundation for AI
Explainable AI also depends on the quality and consistency of the information it processes. A model may produce an understandable recommendation, but that explanation offers limited value if the underlying data is incomplete, outdated, or inconsistent.
Aviation data frequently originates from multiple sources, including surveillance systems, flight plans, weather services, airline information, and airport operational platforms. Integrating these sources is essential to building a reliable operational picture.
Mosaic ATM’s proprietary Mosaic Fuser technology supports this challenge by integrating and harmonizing information from multiple aviation data streams. This data integration expertise supports capabilities such as our Mosaic Situation Viewer for UAS and AAM.
By providing a more consistent information foundation, data fusion supports downstream analytics, visualization, prediction, and decision-support capabilities.
Data provenance, quality monitoring, and the identification of conflicting or missing inputs are also important parts of building explainable aviation AI systems.
This illustrates why trustworthy AI cannot be treated as an isolated algorithm-development problem. It requires an integrated approach to data engineering, system architecture, and operational validation.
Evaluating Large Language Models for Traffic Flow Management
As generative AI gains attention across industries, aviation organizations are beginning to investigate how large language models might support complex analytical and operational tasks.
However, LLMs introduce challenges related to factual accuracy, reproducibility, reliability, and the potential generation of unsupported information. These concerns make systematic evaluation especially important.
Mosaic ATM’s Performance-Based Flow Management Benchmarking LLM Evaluation (PBLE) project investigates whether large language models can support the implementation of performance-based traffic flow management concepts and improve associated decision-making.
Research of this kind is necessary to better understand where advanced AI tools may provide value, where limitations exist, and how performance should be evaluated before considering operational applications.
For aviation, the objective should never be to introduce generative AI simply because it is available. The objective should be to identify applications where it can deliver measurable improvements while meeting the industry’s expectations for reliability and oversight.
What Does Trustworthy, Explainable Aviation AI Look Like?
As adoption increases, aviation organizations should evaluate AI solutions against several fundamental characteristics:
- Transparent recommendations: Systems should identify the meaningful factors, assumptions, and constraints influencing their outputs.
- Data quality and traceability: Users should have appropriate visibility into supporting information, data limitations, and relevant sources.
- Meaningful uncertainty communication: Predictions should communicate limitations and uncertainty where those factors affect decision-making.
- Human-centered design: Explanations must be understandable and actionable for the aviation professionals using the system.
- Rigorous testing and evaluation: AI performance should be assessed under representative operating conditions, including unusual and off-nominal scenarios.
Together, these principles help establish the foundation for AI systems that can be evaluated, maintained, and improved over time.
Explainability should be designed into aviation AI solutions from the beginning rather than added after an algorithm has already been developed.
The Future of Aviation AI Is Explainable
Artificial intelligence has the potential to transform aviation operations, from traffic management and airspace coordination to regulatory analysis and the integration of emerging aircraft technologies.
However, the industry’s long-term success with AI will depend on more than computational power or predictive performance. It will depend on whether intelligent systems provide aviation professionals with reliable information, understandable recommendations, and the ability to exercise informed judgment.
As airspace operations grow more complex and AI technologies become increasingly capable, explainability will become a fundamental component of responsible aviation innovation.
At Mosaic ATM, we are helping advance this future by combining deep aviation domain expertise with AI research, data science, and systems engineering. Our work demonstrates how intelligent technologies can support complex decisions while keeping human expertise, transparency, and safety at the center of the process.
Explore Mosaic ATM’s Aviation Artificial Intelligence capabilities to learn how we are developing advanced technologies to address some of aviation’s most challenging operational problems.
