The U.S. Army’s Synthetic Training Environment (STE) and supporting training and learning concepts define Artificial Intelligence (AI) as a functional requirement to optimize the use of simulation to support individual and team readiness requirements. A current limitation to technologic tools examining AI is access and proper management of meaningful data. Many AI methods are developed under controlled and isolated settings with limited use cases and data-points. These investments prove a methodology from a technology readiness standpoint, but often fail to meet the intent of having ready-to-transition AI services that create valid measures and drive calculated decisions. In this paper, we will present a strategy for defining data requirements and management to support an evolutionary approach to AI development and validation.
How do we directly address this issue? Establishing a data strategy on standards, best practices, acquisition requirements, and mission threds can produce data repositories (i.e., buckets) specifically implemented to drive AI maturation. This emphasizes collecting data with a purpose, and establishing explicit implementation guidelines that align to desired end-state AI capability.
This position is explored at a high-level in the context of STE and future Programs of Record. We will present a framework based on AI services associated with adaptive training management, the type of functions each service provides, and the type of data bucket required to drive its utility. Services explored include building more objective assessments across multi-modal data and across training iterations; building personalized feedback and scenario adaptations that target strengths and weaknesses; creating recommender engines for guided training progression to maintain proficiency; and building realistic synthetic entities that enhance training fidelity. Each of these services demand careful consideration for data instrumentation and management. Beyond persistent storage, we will present recommendations for the capture, contextualization (i.e., metadata/labeling) and retention of data to drive evolutionary maturation of each AI function.
Keywords
AI, BIG DATA, MACHINE LEARNING
Additional Keywords
Training Management, Data Strategy