CareGenerative AI in Teacher Education: Capabilities and Biases of Large Language Models in Situational Judgment Tests and Pedagogical Decision-Making

The integration of generative artificial intelligence, particularly Large Language Models (LLMs), into teacher education represents a transformative yet contested development. This article examines the capabilities and biases of LLMs when applied to two critical domains in teacher preparation: situational judgment tests (SJTs) for assessing professional knowledge, and pedagogical decision-making in classroom contexts. Drawing on recent empirical research, we find that LLMs demonstrate significant promise in generating SJT items with psychometric properties comparable to expert-developed assessments, while substantially reducing development time and resource costs. However, LLMs also exhibit systematic biases—including racial, gender, and socioeconomic biases—when evaluating teaching scenarios and generating educational content. These biases manifest across multiple modalities, from text-based evaluations to vision-language models processing classroom imagery. We synthesize findings on LLM capabilities in automated scenario generation, response analysis, and simulation-based training, while critically examining the ethical implications of algorithmic bias in teacher evaluation. The article concludes with recommendations for responsible integration, emphasizing the necessity of human oversight, critical digital pedagogy, and continuous bias auditing in teacher education programs.