The Janus-Faced Nature of AI in STEM Education

Artificial intelligence (AI)—particularly machine learning (ML) and natural language processing (NLP)—is reshaping research, instruction, and learning in STEM education. These technologies enable automated assessment, individualized feedback, and data-driven curriculum design, but they also raise challenges such as technical limitations, interpretability issues, and epistemological concerns. This chapter examines AI’s double-edged role in STEM education: while supervised and unsupervised ML, NLP, and large language models (LLMs) can classify and analyze complex data, support learning, and generate feedback, they also bring risks of algorithmic bias, opacity, and over-reliance that may undermine authentic learning. LLMs in particular struggle with reasoning and domain knowledge in STEM contexts. Although AI can help scale assessment and support teaching in resource-limited settings, successful integration requires explainable AI, rigorous validation, and pedagogically grounded implementation strategies. We call for interdisciplinary frameworks and reporting standards to ensure AI in STEM education remains both effective and responsible.