Abstract: The convergence and coupling between artificial intelligence (AI) and various disciplines are becoming increasingly prominent, which have been a significant force to drive the transformation of scientific research. AI for science is initiating a new paradigm shift in scientific activities towards data-driven, automated, and intelligent processes. From an epistemological perspective, AI for science will bring new types of knowledge such as intelligence emergence and dark knowledge, and realize a cognitive model based on high-dimensional computation. From a methodological perspective, AI for science embraces uncertainty, addresses complex problems, and provides valuable knowledge production through probability models, and creates a new scientific research model that integrates human scientists with AI scientists.