Military training strives to maximize warfighter readiness to mission effectiveness while minimizing the time spent in training exercises. Optimizing training outcomes requires an ability to tailor training to individual learners, considering their previous training, their operational experience, and their role. Compound AI systems (Zaharia et al, 2024), with multiple interacting AI models and tools, offer a promising solution to this problem by combining the advances seen in leveraging large language models (LLMs), human digital twins, and multi-agent simulation.

This paper describes a possible future concept in which an LLM-driven ecosystem of agents could enhance curriculum-based training. This paper will describe what compound AI systems are and describe a framework for future learning systems that integrates three core components: real-world human training, simulated training, and feedback mechanisms for continuously improving training outcomes. Human learners interact with specialized LLM-driven agents acting as instructors, and separate AI agents evaluate trainees’ competencies, identify knowledge gaps, and generate personalized training content to address deficiencies. To guide real-world instruction, the system leverages simulation using human learner digital twins that model personalities, backgrounds, competencies, and cognitive states. Compound agents interact with the digital twins, evaluating their skills and tailoring instructions, generating data to refine instructional tactics, techniques, and procedures for human learners. Ongoing training improvement stems from two additional AI capabilities: competency evaluation that employs LLMs to assess human and digital twin proficiencies, and procedural generation of curriculum-relevant content to assist learners struggling with knowledge components.

Together, these mechanisms continually optimize instruction to maximize training effectiveness. This novel compound ecosystem of LLM-driven agents, simulated human digital twins and their optimized interactions aims to enhance procedural knowledge acquisition, evaluate conditional understanding, and improve the comprehension of unique military concepts. By learning from human-AI exchanges and simulated outcomes, compound AI systems offer a path towards revolutionizing military training.

Keywords
AGENT-BASED SIMULATION;AI;CONTENT GENERATION;DEEP LEARNING;EMERGING TECHNOLOGIES;MODELING;NATURAL LANGUAGE PROCESSING;TRAINING

Additional Keywords
large language models, human digital twin