ע 237 מרסלו דורפסמן ספר הכנס העשרים ואחד לחקר חדשנות וטכנולוגיות למידה ע"ש צ'ייס: האדם הלומד בעידן הדיגיטלי א' בלאו, ד' אולניק - שמש, נ' גרי, א' כספי, י' סידי, י' עשת - אלקלעי, י' קלמן ו נ' ברנדל )עורכים(, רעננה: האוניברסיטה הפתוחה הערכה בסיוע בינה מלאכותית: נקודות מפגש ופערים בין שיקול הדעת של המרצים לבין ChatGPT בעבודות גמר אקדמיות )פוסטר( מרסלו דורפסמן האוניברסיטה העברית בירושלים marcelo.dorfsman@mail.huji.ac.il AI-Assisted Evaluation: Convergences and Divergences between Teacher Judgment and ChatGPT in University Final Works (Poster) Marcelo I. Dorfsman The Hebrew University of Jerusalem marcelo.dorfsman@mail.huji.ac.il Abstract Background: The emergence of large language models and generative artificial intelligence (GAI) tools has rapidly transformed educational practices. Their use remains an uncertain field, full of risks but also opportunities. Objective: this ongoing research provides a comparative analysis between assessments carried out by a university instructor and those generated by ChatGPT on twenty final projects within a graduate-level education program course. Methodology: We employ an interpretive multiple-case study design to analyze the convergence and divergence between human and ChatGPT-based evaluations of identical academic products. The comparison was conducted using the same evaluation rubric, maintaining the instrument unchanged in order to isolate interpretive differences effects. Conclusions: The results demonstrate a substantial overall correlation between human and automated grades, with an approximate margin of ±5%. However, notable discrepancies are evident in their respective evaluations. The model perceives itself as emphasizing formal coherence and argumentative structure, whereas it perceives human evaluators as tending to value originality, contextual relevance, and personal reflexivity more highly. Implications: Extending this line of research can offer university instructors and evaluators valuable insights for optimizing the deployment of generative AI as a research and assessment aid within a framework that safeguards academic integrity and professional ethics. We examine the scope and limitations of AI as an assessment assistant, the epistemological tensions between professional judgment and automation, and propose guidelines for a more ethical and formative utilization of these technological tools. Keywords: AI-assisted assessment, ChatGPT, Higher education, Assessment rubrics, Teacher judgment, AI epistemology.
RkJQdWJsaXNoZXIy Mjk0MjAwOQ==