Artificial intelligence (AI), as an emerging technology, has been widely used in STEM education to promote the educational assessment. Although AI-driven educational assessment has the potential to assess students' learning automatically and reduce the workload of instructors, there is still a lack of review works to holistically examine the field of AI-driven educational assessment, especially in the STEM education context. To gain an overview of the application of AI-driven educational assessment in STEM education, this research conducted a systematic review based on 17 empirical research published from 2011 January to 2023 April. Specifically, this review examined the functions, algorithms, and effects of AI applications in STEM educational assessment. The results clarified three main functions of AI-driven educational assessment, namely academic performance assessment, learning status assessment, and instructional quality assessment. Moreover, the systematic review found that both traditional algorithms (e.g., natural language processing, machine learning) and advanced algorithms (e.g., deep learning, neural fuzzy systems) were applied in STEM educational assessment. Furthermore, the educational and technological effects of applying AI-driven educational assessment in STEM education were revealed. Based on the results, this research proposed educational and technological implications to guide the future practice and research of AI-driven educational assessment in STEM education.