Theoretical physics constructs mathematical models based on abstractions of natural objects and systems to explain, predict, and control natural phenomena. Artificial Intelligence is a simulation of natural intelligence, enabling computers to perform tasks such as pattern recognition, understanding natural language, playing games, and more. Over the last decade, AI capabilities have grown exponentially, largely driven by empirical trial-and-error methods. This growth has led to the creation of revolutionary tools for work as well as scientific research, while raising concerns over our limited understanding of how these tools work. In this talk, I will argue that applying the principles and mindset of theoretical physics could provide a promising approach to explain, predict, and control AI behavior.