What changes when materials are no longer merely the substrate of computation, but active participants in sensing and learning? This talk explores the reciprocal relationship between artificial intelligence and materials science: AI as a tool for discovering and designing new materials, and materials as the foundation for machines that perceive, learn, and act. Connecting these two directions, I will discuss Physical AI through the interplay of algorithms, functional materials, and embodied interaction with the environment. A central theme will be perceptronics, the integration of electronic sensing and information processing to transform physical signals into meaningful machine perception. Moving beyond cameras, I will examine the role of complementary sensory modalities, particularly artificial olfaction, in giving machines access to dimensions of their surroundings that vision alone cannot capture. Examples from nanoscale devices, molecular sensing, and neuromorphic architectures will illustrate opportunities to bring sensing, memory, and computation closer together. The discussion will also address what separates a compelling laboratory demonstration from a useful technology: energy efficiency, robustness, reproducibility, and scalable integration. Rather than treating hardware as a passive platform for increasingly powerful software, I will explore how materials and device physics can contribute directly to intelligent behavior.