Machine learning has transformed our ability to extract patterns from data, yet understanding and interacting with the physical world presents fundamentally different challenges. Physical observations are often indirect and incomplete, physical laws impose complex constraints, and computationally optimal designs may not be physically realizable. How can physical knowledge be incorporated into learning systems to better understand and interact with the physical world? This talk will present the speaker's recent research at the intersection of machine learning, computational physics, and inverse problems, exploring how physical knowledge can guide reconstruction from indirect measurements, enable expressive and editable neural representations of physical fields, and support the inverse design of structures with desired functionality. It will highlight key challenges in physical modeling, physics-guided optimization, and manufacturing-aware design, pointing toward a broader vision of physical intelligence in which AI systems can model the physical world with greater precision and translate that understanding into new physical capabilities.