Abstract
Adaptive automation in complex training environments continuously adjusts to varying human expertise levels, yet there is limited synthesis of how these systems scale support, extract knowledge, and enhance performance as learners progress from novice to expert. As training environments grow increasingly complex and reliant on human-AI teaming, there is a critical need to understand how automation can not only support performance but actively foster expertise development through intelligent knowledge capture. This meta-analysis examined two decades of empirical research on adaptive automation across high-stakes domains, identifying how AI-driven training systems adapt based on expertise, utilize knowledge elicitation techniques, and formalize knowledge extraction to refine system intelligence. This research provided the first cross-domain synthesis of adaptive automation's role in AI-enabled knowledge elicitation for complext training environments (medical, military, maritime, construction, etc.). Findings revealed three major gaps: 1) an overemphasis on novice-level support with minimal longitudinal tracking of expertise development, 2) inconsistent integration of real-time knowledge elicitation methods, and 3) underdeveloped strategies for converting elicited knowledge into reusable models for adaptive systems. Cross-study themes highlighted importance f dynamically balancing automation control with human input, the need for scalable feedback tailored to cognitive and psychomotor growth, and the critical role of transparent, explainable AI adaptation to build trust across expertise levels. These insights the design of a framework, grounded in learning engineering design, that aligns adaptive automation with continuous knowledge capture, supporting both learner growth and system evolution. By embedding expertise-driven scaffolding and iterative knowledge elicitation, the framework closes the loop between human insight and AI adaptation. It provides the foundation for future research and design of next-generation human-AI systems that advance knowledge elicitation, extraction, and expertise development in complex training environments.