부산대학교 도서관

The 'human-like' behavior exhibited by large language model and the potential risks it poses have sparked widespread interest among researchers. An increasing number of cognitive scientists are delving into the underlying reasons to clearly define the boundaries of large language models' capabilities. Machine psychology has become an essential tool for understanding the mechanisms behind large language model behavior. We analyzed the importance of machine psychology in the context of large model applications in education, from two perspectives: assisting in exploring the intelligent features of large language models and advancing educational research. We examined machine psychology from the angles of intelligence testing, personality traits, theory of mind, and the recurrence of psychological experiments. Using the example of the chain of thought mechanism, we explored the production and development processes of different chains of thought from a machine psychology perspective, contrasting them with human thought processes. It also explores avenues for optimizing large language models performance. Finally, we discussed the controversies surrounding machine psychology experiments and potential future developments. Existing research suggests that by delving into the cognitive mechanisms of models, machine psychology not only aids in accurately assessing the applicability and potential risks of large language models in the field of education but also contributes to a better understanding and simulation of human psychological processes, opening up new possibilities for applications in the education domain.