The long-term value of artificial intelligence (AI) techniques for water sciences remains controversial. Here we argue that the “genes” of AI, meaning its most transformative core ideas, can be absorbed into water sciences to give us unprecedented ways of making inquiries. Supervised AI allows us to capture universal hydrologic laws; differentiable programming can inquire about partially known relationships (e.g., human impacts); transfer learning permits cumulative buildup and abstraction of knowledge; operator learning can introduce more complicated physics and partial differential equation solvers with high efficiency; and generative AI can not only generate hypotheses but also capture and predict future joint changes of Earth systems. Tools incorporating these core ideas of AI will serve as the basic infrastructure that democratizes access to reliable water predictions and gives wing to solving some of the most elusive and important problems in water: human roles, the coevolution of landscapes with water, land-atmosphere interactions, parameter nonuniqueness, spatiotemporal scaling, and groundwater sustainability.