Vision: The physical world is the first teacher. Long before language, biological intelligence learns through structured perception: tracking objects, predicting trajectories, inferring causality from interaction. Current AI skips this curriculum entirely. The dominant paradigm relies on scaling text-based models, and is hitting diminishing returns against the data wall, language-centrism, and the inability to learn beyond what is already written down. The intellectual trajectories of AI and cognitive science have long overlapped; today, the successes and failures of deep learning demand that we take this overlap seriously. My research asks which properties of physical reality and human cognition transfer to machine intelligence and perception, how they transfer, and why, not only to match human capabilities, but to identify the computational primitives that could allow systems to learn new things beyond current human knowledge through grounded interaction with the world. This question spans computer vision, representation learning, and cognitive science: How should the structure of motion, objects, and causes inform what machines learn and in what order? Can the developmental ordering that biology converged on serve as a training curriculum for artificial systems? And when these systems are deployed in social settings, how do we ensure that they produce more equitable outcomes?