This year has seen remarkable progress in artificial intelligence for pathology, with the emergence of general-purpose models that are readily adaptable to a wide range of applications. The first part of this talk will discuss these developments and our experience working with these models that have been trained from large datasets containing diverse pathologies. I will present several applications where we have investigated these models to illustrate their remarkable capabilities, limitations, and tradeoffs with simpler alternatives. The second part of this talk will shift focus from data driven models to hypothesis driven modeling approaches that go beyond prediction to help us better understand disease. I will discuss the development of an elaborate model of breast cancer that incorporates prior knowledge and theories about the disease to elucidate how tissue components contribute to patient prognosis.