Habibi Husain Arifin, Industry Process Consultant, APSouth Digital Engineering Strategy, Dassault Systèmes
As artificial intelligence (AI) continues to make inroads across industries, its application within mission-critical systems such as aerospace and defense must be approached with precision and care. At the heart of this integration lies model-based systems engineering (MBSE), a key enabler of digital engineering. When thoughtfully combined, MBSE and AI can transform how complex systems are designed, validated, and deployed. Unlocking this value, however, requires a deep understanding of AI’s limitations and risks, and a strong commitment to maintaining human oversight.
The intersection of AI and systems engineering has led to two complementary approaches: “AI for Systems Engineering” (AI4SE) and “Systems Engineering for AI” (SE4AI). AI4SE involves applying AI techniques such as large language models (LLMs) (a subset of deep learning) and/or natural language processing (NLPs) to enhance the efficiency and automation of systems engineering tasks. SE4AI, in contrast, uses structured systems engineering methodologies to define the context and improve the explainability and traceability of AI deployments. A balanced focus on both is necessary to achieve robust and accountable digital transformation.
MBSE serves as the digital backbone for engineering workflows, especially domains where safety is paramount such as aerospace and defense. It supports the creation of system architecture models that structure and visualize information, capturing the interdependencies between requirements, components, functions, and constraints. When AI is introduced into this framework, it should be treated not as an infrastructural pillar but as an application-layer enhancement. AI must be driven by clearly defined use cases – mission-focused, data-informed, and aligned with regulatory expectations. The success or failure of AI in this space hinges entirely on its grounding in specific, high-value business goals.
Black box AI vs White box AI: Why it matters for MBSE
To fully understand AI’s role in MBSE, it’s important to distinguish between “black box” and “white box” AI. Black box systems, such as LLMs (derived from neural networks), are powerful but opaque. They can provide impressive outcomes such as identifying facial features or interpreting unstructured text but cannot explain how they reach those conclusions. This lack of transparency is a problem in contexts where explainability, traceability, and legal accountability are essential. There are many attempting research to improve this explainability and traceability factors within black box AIs.
White box AI, which includes expert systems or rule-based engines, offers full explainability. These systems have long been used in many mission-critical systems, such as banking, aerospace, and other industries where decisions must be justified and reproduced. However, white box AI tends to be less adaptable, less interpretable by novice human, and could be less accurate than black box systems (depending on the knowledge given to systems).
The key trade-off is between accuracy and explainability. “There is no silver bullet”, means no single AI model is right for every problem. Each application must be evaluated with a clear understanding of risk and desired outcomes.
In mission-critical environments, these concerns are magnified. The consequences of an AI failure can be catastrophic, including the loss of life. This is why the quality of data and the structure of models are so critical. AI systems trained on incomplete, imbalanced, or synthetic data often struggle to make reliable predictions, especially during failure scenarios where historical data may be limited. For instance, while millions of flights may occur each year, only a small fraction involve significant failures. This creates imbalanced training datasets. Without addressing this issue, AI models may underperform in precisely the situations in which performance matters most.
Many organizations are also navigating a knowledge gap. A common misconception is that LLMs represent the full extent of AI’s capability. Business leaders may believe that simply deploying an LLM will solve complex engineering problems, without recognizing the underlying risks or integration challenges. This is further complicated by the need for technical expertise: Successful AI implementation in MBSE requires professionals who can bridge between automation and domain knowledge, build multi-agent systems, write secure services, manage data pipelines, and perform organic-synthetic data creation, cleansing, training, and testing.
Why human-in-the-loop is essential to MBSE
A key solution to many of these challenges is a human-in-the-loop approach. Human oversight is necessary to ensure that AI-generated outputs are relevant, reliable, and ethically sound. This is particularly important in regulated industries, where decisions must be explainable and defensible.
Human-in-the-loop systems allow domain experts to review and validate AI outputs before they are used in decision-making processes. This enhances quality control, strengthens trust, and helps ensure compliance with legal and regulatory standards.
CATIA Magic bridges modelers and non-modelers
Modeling tools like Dassault Systèmes’ CATIA Magic offer a practical bridge between modelers and non-modelers, enabling efficient data integration through features like spreadsheet synchronization and specification modeling. Many engineers still rely heavily on spreadsheets and documents rather than modeling tools. MBSE platforms help unify these workflows by supporting automated data imports and enabling collaborative validation of specifications. Even without AI, MBSE automation tools can reduce manual work and simplify model creation.
When AI is applied – for instance, using LLMs to extract critical information (such as requirements) from unstructured documents – it can further boost productivity by generating initial drafts for human as a starting point to model.
This layered approach supports both the AI4SE and SE4AI paradigms. AI can help transform unstructured content into structured data, while MBSE provides the models and their relationships that shape AI agent behavior. This two-way flow is vital in hybrid systems, where black box AI can be supplemented with white box constraints to enhance explainability. For example, orchestration frameworks inspired by agentic AI or LangChain/LangGraph architectures can define multi-agent pipelines, where agents manage specific specification types — such as requirements, parameters, systems/subsystems/components, or functions — and interact based on a dependency model defined in MBSE.
Human judgement underpins sound decision making
Context matters. The transformation of raw data into actionable insight – and eventually, into domain-specific knowledge – requires contextual understanding. The DIKW (data, information, knowledge, wisdom) hierarchy is a useful reference. AI can help structure data into information, but the leap from information to knowledge requires context, which MBSE provides. The final step, from knowledge to wisdom and sound decision-making, remains firmly within the realm of human judgment. Machines cannot yet grasp the nuance of ethical, strategic, and contextual factors needed for critical decisions.
This leads to a frequently asked question: Does AI accelerate MBSE adoption? The answer is nuanced. AI can help traditional, document-based organizations transition to systems engineering by automating specification extraction and improving data integration. But it does not eliminate the need for MBSE tools or practices.
AI supports the broader systems engineering ecosystem, and MBSE benefits from that support. However, full adoption still demands organizational change, targeting training, and sufficient technical expertise.
Understanding AI’s limitations is key to success
Looking ahead, the path to successful AI-enhanced MBSE depends on domain alignment, contextual modeling, and a thorough understanding of AI’s limitations. Human-in-the-loop systems will continue to play a central role, along with structured frameworks that clearly define each component’s place within the digital engineering ecosystem.
Whether applying AI to MBSE or using MBSE to structure AI, the core principle remains the same: Intelligent systems are only as trustworthy as the human judgment behind them.