基於生成式人工智慧的智慧輔導系統：自動出題與自動批改的應用與挑戰

A Generative-AI-Based Intelligent Tutoring System: Applications and Challenges of Automatic Question Generation and Automatic Answer Grading

隨著科技迅速發展，教育領域面臨多重挑戰，傳統的出題往往依賴教師的經驗設計，難以滿足個別學生的需求，如何面對學生的多元性及獨特的學習節奏，使傳統的教學方式受到質疑。為了解決這一問題，人工智慧和機器學習技術逐漸被引入教育，尤其是智慧輔導系統的發展，為教學方法帶來了顯著的改變。然而，現有的智慧輔導系統大多依賴預先定義的規則進行互動，使其在回應學生個別需求上仍存在不足。基於此，本研究提出一套整合ChatGPT技術的智慧輔導系統，此系統主要包含兩個功能：自動問題生成與自動答案評分。自動問題生成功能運用生成式人工智慧技術，採用大型語言模型協助生成選擇題及簡答題，幫助學生在不同層次上進行學習；自動問題批改功能則利用ChatGPT即時評估學生的答題結果，根據正確程度將其分類，並提供對應的詳解。根據本實驗結果，生成式人工智慧採用修改現有資料而非重新生成的策略，能夠有效地提升內容的精確性，並結合本研究專屬設計的提示詞，使生成題目的品質大幅提升。這樣的改進不僅提升了傳統出題方式的靈活性，還大幅減少了問題設計所需的人力成本，從而能夠更好地應對當今教育領域所面臨的挑戰。

With the rapid development of technology, the field of education faces increasing challenges. Traditional question design often relies on the teacher’s experience, making it difficult to meet the needs of individual students. The diversity of students and their unique learning paces have led to doubts about traditional teaching methods. To address this issue, Artificial Intelligence (AI) and Machine Learning technologies are gradually being introduced into education. Developing Intelligent Tutoring Systems (ITS) has significantly changed teaching methods. However, most existing ITS rely on predefined rules for interaction, which limits their ability to respond to individual student needs. Based on this, our research proposes an ITS integrated with ChatGPT technology. This system mainly includes two functions: automatic question generation and automatic answer grading. The automatic question generation function utilizes generative AI technology, employing large language models to generate multiple-choice and short answer questions, helping students learn at different levels. The automatic answer grading function uses ChatGPT to evaluate students’ responses in real time, categorizing them based on accuracy and providing detailed explanations accordingly. According to the results of our experiment, generative AI, by adopting a strategy of modifying existing data rather than regenerating it from scratch, can effectively improve content accuracy. Coupled with specially designed prompts, the quality of generated questions has significantly improved. This enhancement not only increases the flexibility of traditional question design but also greatly reduces the manpower required for question creation, better addressing the challenges faced by the modern educational field.