Sociology is a science concerning itself with interpretive understanding of social action as well as causal explanation. A causal explanation should be the foundation of prediction. For many years,constrained by data and computing power,quantitative research of social science has primarily focused on statistical test to analyze correlation and causality, leaving predictions largely ignored. By sorting out the historical context of"social prediction",this paper redefines this concept by introducing why and how machine learning can help prediction in a scientific way. This paper further summarizes the academic value and governance value of social prediction,and suggests that it be a potential breakthrough in contemporary social research paradigm.We believe,through machine learning,we can witness the advent of an era of paradigm shift from correlation,causality to social prediction,providing a rare opportunity for sociology in China to lead the international frontier of computational social sciences,and accelerating the construction of philosophy and social science with Chinese characteristics.