“Transforming Physics Assessments using Generative AI" - Zhongzhou Chen скачать в хорошем качестве

“Transforming Physics Assessments using Generative AI" - Zhongzhou Chen

This talk (“Transforming Physics Assessments using Generative AI") was recorded on February 12, 2026 as part of the USC Department of Physics and Astronomy's ongoing colloquium series. Dr. Chen is an associate professor in the Department of Physics at University of Central Florida. To learn more about Dr. Chen, feel free to visit his website ( https://sciences.ucf.edu/physics/pers... ). Abstract: In this talk, I will introduce my research into transforming Physics assessments utilizing Generative AI along three strategic directions. First, to address multiple barriers such as accessibility, test security, and test development, we facilitate a transition from single attempt high-stakes exams to multi-attempt, formative assessments using large isomorphic problem banks. Supported by the NSF, we employ generative AI to assist instructors in generating large numbers of structurally consistent isomorphic problems, and transform tests in Physics, Chemistry, and Math. Second, we shift from product- to process-focused evaluation by automating the grading of students' written reasoning. Our results demonstrate that GPT-4o is capable of achieving human-level accuracy in partial-credit grading and feedback writing for written response of introductory physics problems. Third, we move beyond assessing rote proficiency toward evaluating Discipline-based AI Literacy (DAIL) through "Human-AI Collaboration" tasks. These tasks exploit the "jagged frontier" of AI intelligence, requiring students to collaboratively evaluate and correct AI outputs to solve authentic disciplinary problems.