窥探机器之窍:机器心理学视角下的大模型教育应用

Peering into the Machine: Large Language Model Applications in Education from the Perspective of Machine Psychology

大模型所呈现的“类人”行为以及潜在风险引发了研究者广泛的兴趣,越来越多的心理学家开始深入探索其背后的原因,以更清楚地划定大模型的能力边界,而机器心理学则成为了理解大模型行为背后机制的重要工具。从探索大模型的智能特征和推动教育研究的发展两方面分析机器心理学对于大模型教育应用的重要性,从智力测试、人格特征、心理理论的适用性和心理学实验的复现等角度分析机器心理学的研究视角,在此基础上,以思维链为例,通过对比人类的思考方式与不同思维链的形成,从机器心理学视角分析思维链的产生与发展过程,探索大模型性能优化的途径。最后,基于此探讨了机器心理学实验的争议以及未来新的发展。现有研究表明,通过深入挖掘模型的认知机制,机器心理学既有助于更准确地判断大模型在教育领域的适用性及潜在风险,也有助于更好地理解和模拟人类心理过程,为大模型教育领域的应用提供新的可能性。

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.