Abstract: Whether there is the logic of scientific discovery has always been controversial. In the early development of artificial intelligence (AI), there were limitations in the research of automatic scientific discovery through rule-based and data mining methods, including the need for prior knowledge or only the discovery of empirical laws in specific fields. Through two case studies, it was shown that the recent research based on machine learning can find the concept of science and even a simple theory without prior knowledge. But there are still some problems such as the source of training data, the choice of observation and experiment, the construction of scientific theory and the causal modeling which need interdisciplinary research combination with philosophy of science and history of philosophy.