Navigating the Landscape of Robust and Secure Artificial Intelligence: A Comprehensive Literature Review

Abstract

Addressing the multidimensional nature of Artificial Intelligence assurance, this thorough survey is dedicated to elaborating on various aspects of ensuring the reliability and safety of computerized systems. It steers through the turbulent seas of model enervates, unmodelled phenomena, and security menaces to give an elaborate lit review. The review touches upon the boisterous ways of addressing these intricate mitigation strategies for model errors used in the past, the challenges of under-specification with modern ML models, and how understanding uncertainty is crucial. In addition, it evaluates the AI system’s security basis, the emerging Adversary Machine Learning field, and its processes necessary for testing and evaluation of weaker adversarial case studies. The review of literature also looks upon the situation of DoD context, how the terrain surrounding developmental and operational testing is altering with all these shifts in culture that must be implemented if not to implement robust but secure AI implementation.