关键词: 机器发现, 机器思维, 科学发现, 认知双重过程理论, 图神经网络

Abstract: Machine learning has made significant advancements in scientific research, especially in data-intensive fields.However, its application in scientific discovery still faces limitations.In recent years, computational scientists have proposed several methods that enable machines to autonomously discover scientific laws from data sets.These attempts have sparked widespread debate over whether they truly capture the essence of scientific discovery.Machine learning systems differ significantly from human discovery processes in aspects such as data selection, model construction, the relationship between theories and phenomena, and the nature of machine thought itself, showing that they lack autonomous thinking capabilities.Their outputs are still influenced by existing human knowledge systems and cognitive frameworks.Despite this, artificial intelligence has profoundly changed the way scientific research is conducted.It is essential to recognize both the irreplaceable role of machines in scientific research and the unique role of human creativity and intuition in scientific discovery.The future should aim for effective human-machine collaboration, not for machines to replace humans as the primary agents of scientific research.

Key words: machine discovery, machine thinking, scientific discovery, the dual process theory of cognition, graph neural networks