2021-09-30-兰德-军事准备的人工智能工具（英）_58页_917kb

报告摘要

Summary

NATIONAL DEFENSE RESEARCH INSTITUTE

This report explores the use of artificial intelligence (AI) to enhance military readiness by analyzing unstructured textual data from commanders' reports. Key points include:

Purpose: Military readiness, measured annually, involves subjective narratives alongside categorical assessments. AI can automate readiness analysis, improving accuracy and providing situational awareness.
Methodology:
- Used defese-specific word embeddings trained on defense-related data for semantic analysis.
- Employed deep neural networks to predict readiness levels (C-level 1–4) based on text descriptions.
- Achieved 75% accuracy on the test set, significantly better than baseline logistic regression models.
Findings:
- Defense-specific embeddings improved performance in domain-specific tasks but not in predictive modeling.
- Architectures incorporating LSTMs/GRUs and small batch sizes enhanced model performance.
- The model successfully interpreted text to predict readiness levels, with high recall and precision across all C-levels.
Applications:
- Potential integration into readiness reporting systems to improve commander clarity.
- Future use in chatbots for real-time feedback and refinement of unit assessments.
- Broader implications for transfer learning in defense-related NLP tasks.
Conclusion: The research demonstrates AI's potential to make readiness measurements more objective and timely, with applications extending beyond classification to decision support.