Nuclear Weapons + AI: The Mathematics of Fatal Inevitability

Source abstract

The article analyzes the systemic risk caused by the integration of neural network algorithms into nuclear triad control loops and ballistic missile early warning systems. The author examines the fundamental gap between the statistical nature of AI (the ‘black box,’ hallucinations) and nuclear deterrence’s need for absolute determinacy. Special attention is paid to time pressure, which removes humans from decision-making as the window is shrunk by the proximate deployment of submarines and the development of hypersonic weapons. Based on the concept of ‘normal accidents’ and a comparative analysis of control systems (from rigid Perimeter algorithms to adaptive neural networks), the article proves that the probability of AI-controlled systems’ catastrophic failure ultimately approaches 1. The article proposes measures to preserve the ‘physical gap’, keep humans in the decision-making loop, and institutionalize control over military algorithms.