Scholarlink

<p>This study systematically examines the capabilities and biases of large language models (LLMs) in early childhood education (ECE), focusing on child observation assessment and pedagogical decision-making. Using two validated situational judgment test (SJT) instruments, twelve leading Western and Chinese LLMs were evaluated across multiple developmental and instructional scenarios. Results indicate that LLMs demonstrate high stability and relatively strong performance in child observation assessment, particularly in identifying children’s current developmental levels. However, performance declines in pedagogical decision-making tests that require value-laden judgments and complex trade-offs. While no significant differences were found between Chinese and Western models in overall scores, systematic cross-cultural differences emerged in value prioritization, such as emotional support, safety, and rule enforcement. Moreover, all models exhibited consistent decision-making biases, including limited developmental sensitivity, weak methodological awareness, and a preference for order maintenance. These findings suggest that although LLMs hold potential as supportive tools in ECE, their use requires critical mediation to prevent misalignment with child-centered educational principles.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

From observation assessment to pedagogical decision-making: Capabilities and biases of LLMs in ECE Situational judgment tests

  • 期刊论文

  • 发表日期：

    2026年6月25日

查看全文
  • Fang Wang,
  • Li Ping Guo,
  • Yi Rong Zhang

摘要

This study systematically examines the capabilities and biases of large language models (LLMs) in early childhood education (ECE), focusing on child observation assessment and pedagogical decision-making. Using two validated situational judgment test (SJT) instruments, twelve leading Western and Chinese LLMs were evaluated across multiple developmental and instructional scenarios. Results indicate that LLMs demonstrate high stability and relatively strong performance in child observation assessment, particularly in identifying children’s current developmental levels. However, performance declines in pedagogical decision-making tests that require value-laden judgments and complex trade-offs. While no significant differences were found between Chinese and Western models in overall scores, systematic cross-cultural differences emerged in value prioritization, such as emotional support, safety, and rule enforcement. Moreover, all models exhibited consistent decision-making biases, including limited developmental sensitivity, weak methodological awareness, and a preference for order maintenance. These findings suggest that although LLMs hold potential as supportive tools in ECE, their use requires critical mediation to prevent misalignment with child-centered educational principles.