Teaching Practice Research Project
在全球化與數位轉型的浪潮下，高等教育中的商管課程愈發強調「問題解決」及「跨文化與跨國管理視角」的培養。其中，國際企業管理（International Business Management）課程，由於其涵蓋國際市場選擇、進入策略擬定、跨國知識與人才管理、全球供應鏈與市場進入策略選擇、地緣政治與反全球化議題、新興市場機會識別以及永續發展行動等之整合與分析等複雜主題，往往成為學生從初階入門課程進階至中級難度課程時的巨大挑戰。特別是以英語授課並採用英文個案（Case Study）的教學模式，不僅增加了語言門檻，更要求學生能以批判思維和深度理解能力來面對真實國際企業情境中的決策難題。
在此類中級難度的課程中，學生面臨多重挑戰：首先，他們必須在語言障礙下理解20頁以上的英文企業個案，這對於仍在訓練英語學術能力的大二學生而言相當吃力。其次，課程突然難度的提高（相較於大一的入門課程）使學生往往不知從何下手，缺乏系統解題的歷程（Hwang et al., 2013; Lu & Xie, 2023）。第三，傳統教學模式倚賴人力助教，但助教訓練成本與人力資源有限，難以為所有學生成為有效的「學習引導者」（Chen et al., 2000）。第四，學生在準備個案討論或課中小組討論時，常缺乏策略化提問與層層推進的能力（Hwang & Kuo, 2011）。隨著生成式人工智慧（Generative AI）的可及性提高，學生雖已自發性使用此類工具，但多數只是「一問一答」地將課前問題丟給AI求解，欠缺資料查核與深入互動的過程，導致生成內容的正確性與深度無從保證（Lu & Xie, 2023）。
在此背景下，將AI做為「支架」（scaffolding）的概念受到重視。支架的本義在於透過臨時性、動態且漸進淡出的教學輔助，幫助學習者克服學習困難並最終獨立解決問題（Kim & Hannafin, 2011; Janson et al., 2020）。現有研究顯示，數位工具與教學策略的整合，有助於發展學生的問題解決能力（Lu & Xie, 2023）。然而，少有研究探討生成式AI（如ChatGPT或Claude）在中級難度商學院課程中，如何結合適當的教學設計，為學生提供層層推進的引導，培養其深度提問、資料查核與策略思考等能力。
因此，本研究擬透過將AI作為支架的概念引入大二的國際企業管理課程中，為學生量身訂製教學策略，並探討此模式對學生問題解決能力與學習成效的影響。此項研究不僅對課程本身有所助益，亦可為其他類似中級難度的管理課程提供一個可參考的教學介入策略。藉由課程設計、AI引導介面及分階段問題提問與查核任務的安排，本研究期望為學界提供一個更成熟的模式，闡明在高速進化的數位學習環境中，如何有效整合生成式AI與支架概念，以提升學生對複雜商管個案的理解與解題能力。
In the wave of globalization and digital transformation, business and management courses in higher education increasingly emphasize the cultivation of "problem-solving" and "cross-cultural and multinational management perspectives." The International Business Management course, in particular, presents a significant challenge for students transitioning from introductory to intermediate-level courses. This is due to its complex topics, including international market selection, entry strategy formulation, cross-border knowledge and talent management, global supply chain and market entry strategy choices, geopolitical and anti-globalization issues, emerging market opportunity identification, and sustainable development actions. The challenge is amplified in English-taught courses using case studies, which not only add a language barrier but also require students to apply critical thinking and deep understanding to solve decision-making problems in real international business scenarios.
In such intermediate-level courses, students face multiple challenges. First, they must comprehend English business cases of over 20 pages despite language barriers, which is demanding for sophomores still developing their academic English skills. Second, the sudden increase in difficulty compared to freshman-level introductory courses often leaves students unsure where to begin, lacking a systematic problem-solving process (Hwang et al., 2013; Lu & Xie, 2023). Third, traditional teaching models rely on human teaching assistants (TAs), but the cost of training and limited human resources make it difficult for TAs to be effective "learning guides" for all students (Chen et al., 2000). Fourth, students often lack the ability to ask strategic questions and progressively deepen their inquiry when preparing for case discussions or in-class group work (Hwang & Kuo, 2011). With the increasing accessibility of Generative AI, students have begun to use these tools spontaneously, but mostly in a simple "Q&A" manner, feeding pre-class questions to the AI for answers. This lacks a process of data verification and in-depth interaction, making the accuracy and depth of the generated content unreliable (Lu & Xie, 2023).
Against this backdrop, the concept of using AI as "scaffolding" has gained attention. The essence of scaffolding is to provide temporary, dynamic, and gradually fading instructional support to help learners overcome difficulties and eventually solve problems independently (Kim & Hannafin, 2011; Janson et al., 2020). Existing research shows that integrating digital tools with teaching strategies helps develop students' problem-solving abilities (Lu & Xie, 2023). However, few studies have explored how Generative AI (such as ChatGPT or Claude) can be combined with appropriate instructional design in intermediate-level business school courses to provide students with progressive guidance, thereby cultivating their abilities in deep questioning, data verification, and strategic thinking.
Therefore, this study intends to introduce the concept of AI as scaffolding into a sophomore-level International Business Management course. We will tailor teaching strategies for students and explore the impact of this model on their problem-solving abilities and learning outcomes. This research will not only benefit the course itself but also provide a referenceable instructional intervention strategy for other similar intermediate-level management courses. Through course design, an AI-guided interface, and phased questioning and verification tasks, this study aims to provide academia with a more mature model, clarifying how to effectively integrate Generative AI and the concept of scaffolding in a rapidly evolving digital learning environment to enhance students' understanding and problem-solving capabilities for complex business cases.
本研究旨在探討「生成式AI做為支架」(AI as scaffolding) 的教學策略在中級難度之商學院課程（以大二英文授課之國際企業管理課為例）中對學生的學習成效與問題解決能力所產生的影響。本研究期望達成下列目標：
透過本研究，冀望為教學實務與教育科技研究社群提供一個創新且可行的教學介入模式範本，並為未來在中級難度課程中導入AI支架設計之教學創新，提供實證依據與應用參考。
This study aims to investigate the impact of the "AI as scaffolding" teaching strategy on student learning outcomes and problem-solving abilities in an intermediate-level business school course (using a sophomore-level, English-taught International Business Management course as an example). This study aims to achieve the following objectives:
Through this research, we hope to provide an innovative and feasible instructional intervention model for the teaching practice and educational technology research communities, and to offer empirical evidence and practical references for future teaching innovations that incorporate AI scaffold design in intermediate-level courses.
過往研究已指出，在問題解決導向（problem-solving-oriented）的學習環境中，透過適當的教學設計與科技輔助工具，學生得以形成更深度的知識結構與策略性思維（Janson et al., 2020）。基於對複雜學習介入措施之研究顯示，採用以問題為中心之指導策略搭配技術增強（technology-enhanced scaffolding）介入，能顯著增進學習者對複雜技能（complex skills）的理解與轉化能力（Kim & Hannafin, 2011a）。在此類學習中，複雜技能包括高階思考（如批判思考、創造力）與操作層面的精熟（如程式設計、醫療手術技巧），而管理情境中之策略分析、決策擬定、跨文化溝通與整合國際市場資訊的能力，也可視為一種複雜技能的延伸應用。另一方面，研究探討教育科技（educational technology）在促進問題解決能力（problem-solving skills）方面之運用，指出多元科技工具的綜合設計、教學策略（如支架scaffolding、同儕協作、課程導向與訓練）、以及學習者持續反思與資訊查核流程，均對問題解決績效有正向影響（Clarke & Dede, 2010; Lu & Xie, 2023）。
然而，現今生成式人工智慧（Generative AI）之興起帶來新契機和挑戰。學生在面對英語個案（case study）時，往往嘗試藉由將問題直接拋給AI產生答案，但缺乏深入提問與多回合挖掘的過程，容易產生資訊品質不穩、錯誤未被查覺、以及未能內化為學生自身思考策略等問題。本文之研究即旨在回應此落差，嘗試將AI做為一種「支架工具」（AI as scaffolding），透過精心設計之教學模式，引導學生發展深度提問、資訊查核、以及逐步形成決策策略的能力，並同時改善學生在中階難度國際企業管理課程中的學習體驗與成效。
21世紀高等教育強調學生需具備「學習與創新技能」、「資訊、媒體與科技素養」、以及「生活與職涯能力」(van Laar et al., 2020)。在商學教育領域，特別是國際企業管理課程中，問題解決（problem-solving）被視為核心能力之一，因為學生將來不僅面對國際市場的不確定性，也必須因應地緣政治、可持續發展、新興市場崛起與企業數位轉型等多重挑戰。問題解決不僅是單一認知技能，而是一組複雜的過程，包括理解問題、分析情境、發展解決方案、資料查驗與方案執行、以及成效評估與反思（Sternberg, 2003）。其本質涉及多層次認知加工與情境化知識運用。
對於大學二年級階段的學生而言，從較簡易且框架明確的初階課程突然轉移至要求更高、主題更複雜的IBM課程，極具挑戰性（Janson et al., 2020）。課程中使用個案教學法（Case-Based Learning）在商學與管理教育中行之有年，被視為發展高階思考、批判分析與問題解決能力的重要途徑（Smith & Miller, 2005），每個個案約20頁長的英文文本，描述一家國際企業在特定決策節點面臨的難題與環境變遷。學生必須在語言與專業知識雙重負擔下，從冗長資訊中萃取關鍵要點、建立問題框架、比較策略選項、並考慮跨文化溝通與國際供應鏈運作模式。個案教學高度仰賴學生的閱讀理解與課前準備，若學生在前期無法有效吸收背景資訊，課堂討論往往淪為少數語言能力與專業基礎較佳者的獨角戲。有研究指出，複雜問題解決中，若學生缺乏引導或支援，往往淪為淺嘗輒止的資訊蒐集而未形成深度理解（Loibl, Roll, & Rummel, 2017；Kim & Hannafin, 2011）。在個案教學中，若能提供資訊搜尋、概念釐清、及策略提示等，有助於更多學生參與深度討論（Merseth, 1991；Huang & Chuang, 2008）。
除此之外，教師或助教的人力資源有限，傳統實體助教難以針對每位學生給予即時與有效協助。在數位時代中，如何善用教育科技工具，以降低助教訓練成本並擴大支援範圍，已成為關鍵課題（Lu & Xie, 2023）。技術輔助學習（technology-mediated learning）必須與適當的教學策略整合，才有可能真正提升學生問題解決之學習成效。
教育領域中針對「最少指導」(minimal guidance)與「直接教學」(direct instruction)有長期爭論（Kirschner et al.,2006）。前者認為學生應在探索中學習，由試誤中摸索出解決之道；後者則強調在複雜課程中若無適當明確指導，學生可能迷失方向、浪費心智資源。研究顯示，當任務非常複雜且學習者先備知識不足時，過度自由的學習或僅輕微指導可能導致學習效益不彰（Sweller, 1988; Kirschner et al., 2006）。
前述提及的「支架」（scaffolding）理論由Wood、Bruner與Ross（1976）提出，強調在學習者尚未具備足夠能力獨立完成任務前，由教師或有能力的同儕提供適切且即時的協助，待學習者逐漸精熟後再逐步撤除支援。此概念在數位學習環境中屢有應用（Kim & Hannafin, 2011a; Chen & Bradshaw, 2007）。然而，傳統支架或是由人力（教師、助教、同儕）提供，或是由預設教學系統靜態呈現，較難針對學生的多元需求與動態問題理解而快速調整（Janson et al., 2020）。支架技術往往透過提示、範例、線索引導與問題分解等形式出現，並透過數位環境或工具實現。例如，教學平台可在學生解題過程中適時出現提示，協助其將複雜問題拆解為易於處理的子任務；或是利用智能導師系統（intelligent tutoring system）根據學生表現動態調整支持程度（del Olmo-Muñoz et al., 2022；Harskamp & Suhre, 2007）。
Janson等人（2020）研究指出，在管理教育翻轉課堂中設計「技術增強之支架」（technology-enhanced scaffolding）能顯著影響學習者的問題解決表現與學習過程滿意度。程序支架可協助學習者瞭解如何使用學習管理系統（Learning Management System, LMS）與資源；後設認知支架促進學習者自我監控與反思；概念支架為學習者提供理解任務所需的知識線索；策略支架引導學習者考量多元解決方案並評估最適策略（Kim & Hannafin, 2011a, 2011b）。支架方法將探究過程拆解為可管理的階段，並於不同階段提供相應的資訊或指引，使學習者不致在一開始就被問題的複雜性壓垮。在面臨複雜任務時，學生若能獲得多元教學干預及技術支援，有助於降低認知負荷，使他們將心智資源投入於策略思考與知識結構建構。
教育科技在促進學生問題解決能力上已有多方嘗試，包括遊戲化學習（game-based learning）、線上協作平台（online collaborative tools）、智慧教學系統（intelligent tutoring systems），以及虛實整合的學習系統（如AR/VR情境）（Lu & Xie, 2023）。課程設計中能提供系統化的指引、工具使用訓練、同儕互動機會、以及支架式的回饋，學生不僅能增進問題分析與策略擬定的能力，亦能培養更深層的學習動機與自我調控（self-regulation）。Hwang等人(2013)的研究顯示，在高互動性平台中，提供步驟化的指導與提示，能讓學生逐步建構問題解決流程。過往研究多集中於特定數位工具（如遊戲、模擬系統、LMS）促進問題解決能力的成效，但較少提到生成式AI的智慧互動特性。儘管已有研究肯定AI的潛力，但目前仍缺乏針對「中級難度商學院課程」整合生成式AI為支架的實證研究。過去的研究多聚焦於STEM或醫學領域（Frerejean et al., 2023; Costa & Miranda, 2019）或純粹的計算思維、邏輯推理訓練上，對國際企業管理等需多元觀點整合的情境較少著墨。商學課程特別之處在於情境變化多端、資訊不完全、並需考量文化與政治經濟等柔性因素，AI支架該如何引導學生思考，尚待探索。
傳統的支架多由教師或預先設計的教學系統提供，缺乏彈性與即時互動特性。生成式AI的興起為此帶來新契機：AI可成為一個「可對談、可自適應」的智慧支架，學生可透過多回合對談，不斷提出問題、要求更多範例或反饋、並檢視AI回應品質，進一步促進深度思考與資料查證意識。
程序支架的核心在於提供學習者對學習環境與工具的導覽、任務步驟的明確指引（Hannafin et al., 2004）。GenAI可在此層面擔任「即時輔助導覽員」的角色。例如，在使用LMS進行學習時，學習者可向ChatGPT諮詢「我該如何使用某個資源？」「下一步該做什麼？」GenAI依據學習平台的功能與教學設計者預設的資訊，給予階段性指令或策略提示。相較於預先錄製的教學影片或靜態文字說明，GenAI可根據學習者當前狀態給出更為精準的程序性指引。例如，如果學習者在進行人力資源管理流程分析（Janson et al., 2020研究中提及的真實商業流程案例）時感到迷失，GenAI可立即回應並提供「請先閱讀該流程的BPMN圖（Business Process Model and Notation），再嘗試指出流程弱點」的步驟建議，這種互動即時性將提升程序支架的有效性。
後設認知支架旨在協助學習者監控、反思與調整自身的學習策略（Azevedo & Hadwin, 2005; Jumaat & Tasir, 2016）。問題解決不僅是純粹的認知歷程，也牽涉元認知（metacognition），即學生對自身理解、策略選擇與學習進度的監控（Azevedo & Hadwin, 2005; Hannafin et al., 2004）。GenAI在此可扮演「元認知教練」的角色。例如，學習者可詢問GenAI：「我剛回答了作業中的問題，你能幫我看我是否已觸及核心概念？」「我花了很多時間在搜尋資料，我該如何提升學習效率？」GenAI可透過對話分析，發現學習者的理解盲點並提出引導性問題，促使學習者反思自己的學習進程。此外，GenAI可建議學習者訂定學習目標並於後續對話中詢問達成狀況，藉此促進學習者自我調節（Wan, Compeau, & Haggerty, 2012）。此類後設認知支架的動態化，有助減少學習者的茫然感，提升學習過程的滿意度（Gupta & Bostrom, 2013）。
概念支架透過提供關鍵概念、例證、圖示，協助學習者理解任務中所需的知識結構（Cagiltay, 2006）。在此方面，GenAI可作為「即時知識引擎」，當學習者遇到概念困惑時，詢問GenAI「什麼是本研究案例中的業務流程再造(BPR)？」「BPMN中『事件』和『任務』有什麼差別？」GenAI可提供簡明扼要的解釋，或在需要時延伸提供更多實例及補充材料（Sun et al., 2018）。這種支援模式不僅降低學習者搜尋概念資訊的認知負擔（外在負荷），也使學習者更能專注於將新知識融入先前認知結構（本質負荷的提升），從而促進深度學習。
策略支架鼓勵學習者探索多種解決問題的途徑並評估選擇（Kim & Hannafin, 2011b）。GenAI在此可作為「策略規劃助理」，對於複雜問題提出多樣解決方法，並促使學習者思考不同方案的優劣。例如，學習者可詢問：「我目前為改善人力資源招聘流程，有哪些可能的流程優化策略？」「若採用策略A會有什麼風險？」GenAI可提出多種策略並協助比較，不僅引導學習者評估選項，也有助於塑造問題解決的高階思考技能（Zohar & Dori, 2003）。在此過程中，GenAI亦可隨著學習者回饋動態調整建議，形成真實的互動式策略支架環境。
近年大型語言模型（Large Language Models, LLM）如ChatGPT、Claude之類的生成式AI工具快速普及，學生能輕易獲取「看似專業」的回答。然而，目前學生多半僅將此視為快速給答案的管道，將課前問題丟進AI模型，拿到簡答後即草草了事，缺乏進一步深究的驅動力。此種使用模式並未產生有效的支架效果，反而可能讓學生放棄自我思考和評估資料正確性的責任，降低學習深度。理論上，若能設計一組教學策略，要求學生必須與AI多回合互動、每一回合提出更具體與尖銳的提問、要求AI提供出處、促進資料比對與查核行為，那麼AI即可成為真正的支架工具。
問題解決的促進需要多種教學策略的整合（Lu & Xie, 2023）。AI不僅是一種資訊提供工具，也是刺激學生問對問題、進行資訊素養訓練（information literacy）的契機。Hwang & Kuo (2011)的研究顯示，當學生在數位學習環境中受到適當支架策略引導，他們能發展出更成熟的問題解決心智模式，包括系統搜尋、策略評估、以及自行構建解決方案。若此心智模式可以透過AI互動層層鞏固，則學生在面對個案時，將不再只是被動接收，而是能主動提出假設、查證、修正並最終形成獨特見解。
AI支架的特色是可動態調整指導程度。初期可提供較明確步驟式建議（類似直接教學的部分），待學生理解基礎後，教師可要求學生減少詢問AI直接答案的次數，而須透過AI來索取資訊做整合，縮小AI引導的份量，逐步讓學生走向獨立探索（Loibl et al., 2017）。此模式提供了一個「適性輔助—逐步淡出」的路徑，使得AI支架在「最少指導」與「直接教學」中取得中間點，既避免學生完全無所適從，又防止其對AI回答過度依賴而喪失自主思考能力。此種策略與Renkl (2002)所提倡的循序淡出支架原則相呼應，亦符合Janson et al. (2020)強調技術增強之教學應在初期扮演強化結構的角色，隨著學習者能力提升，慢慢退居為背景協助。
在國際企業管理課程中，語言為一大阻礙。學習者非英語母語，加上個案多以英語書寫，難度倍增。然而，AI具備多語言能力，可幫助學生將複雜專業文獻解構為易理解的段落，或透過雙語並列讓學生檢視原文與AI產生之中文解釋。此舉可降低語言造成的認知負荷，讓學生更專注於問題的邏輯與策略層面。在強調的教育科技應用中，並未大幅討論語言障礙，但本研究所處脈絡下，語言問題顯然是應用AI輔助的重要場域。
Cognitive Load Theory（CLT）指出，學習成效取決於學習材料與任務要求是否與學習者有限的工作記憶資源相適配（Sweller, 1988；Paas et al., 2003）。在中級難度國際企業管理課程中，學生面對英文案例所帶來的語言負荷、專業概念的本質負荷（intrinsic load）、以及搜索資訊與建立解題策略的外在負荷（extraneous load）可能過高，影響深度學習（Kirschner et al., 2006）。生成式AI之應用若能在教學設計中平衡認知負荷，將大有助益。例如，在初階時，學生可要求AI簡化案例描述、歸納關鍵要點、或以雙語對照形式呈現資訊，藉此降低不必要的外在負荷（Janson et al., 2020）。隨著課程進行，教師可逐漸要求學生面對更完整、未精簡的案例，並要求學生透過AI得到的資訊進行再次查核與反思，以逐漸提升本質負荷的處理能力（Renkl, 2002）。同時，適度的提示與規範（如要求AI在回答時同時列出思考步驟或資訊來源）能避免過高的外在負荷，使學生能將更多認知資源用於真正問題解決的核心策略思考。
綜合而言，全球高等教育的發展強調學生問題解決與高階思維能力的培養，特別是在國際商學領域中，面對多語言、多文化與複雜案例的挑戰下，如何善用教育科技為學生提供適時且動態支援成為迫切議題。先前文獻（Janson et al., 2020; Lu & Xie, 2023）已清楚顯示技術增強學習對問題解決有正面影響，但對生成式AI之應用則仍有待深入研究。AI的優勢在於可透過自然語言多回合對談提供彈性支架，並使學生在資訊查核、策略形成、方案評估中受益。然而此模式必須有精心設計之教學策略加持，包括要求學生深化提問、查證資訊、進行元認知反思與團隊討論。
Previous research has indicated that in a problem-solving-oriented learning environment, students can develop deeper knowledge structures and strategic thinking through appropriate instructional design and technology-assisted tools (Janson et al., 2020). Studies on complex learning interventions show that adopting a problem-centered instructional strategy combined with technology-enhanced scaffolding can significantly improve learners' understanding and transfer of complex skills (Kim & Hannafin, 2011a). In this type of learning, complex skills include higher-order thinking (such as critical thinking and creativity) and mastery of operational aspects (such as programming or medical surgical techniques). The abilities to analyze strategies, make decisions, communicate cross-culturally, and integrate international market information in management contexts can also be seen as an extended application of complex skills. On the other hand, research on the use of educational technology to promote problem-solving skills points out that the integrated design of multiple technology tools, instructional strategies (such as scaffolding, peer collaboration, curriculum orientation, and training), and learners' continuous reflection and information verification processes all have a positive impact on problem-solving performance (Clarke & Dede, 2010; Lu & Xie, 2023).
However, the rise of Generative AI brings new opportunities and challenges. When faced with English case studies, students often try to get answers by directly throwing questions at AI. This approach lacks in-depth questioning and multi-round exploration, which can easily lead to unstable information quality, unnoticed errors, and the failure to internalize thinking strategies. This study aims to address this gap by attempting to use AI as a scaffolding tool. Through a carefully designed teaching model, we will guide students to develop the abilities of deep questioning, information verification, and gradual formation of decision-making strategies, while also improving their learning experience and effectiveness in an intermediate-level international business management course.
21st-century higher education emphasizes that students should possess "learning and innovation skills," "information, media, and technology literacy," and "life and career skills" (van Laar et al., 2020). In business education, especially in international business management courses, problem-solving is considered a core competency. This is because future graduates will face not only the uncertainties of the international market but also multiple challenges such as geopolitics, sustainable development, the rise of emerging markets, and corporate digital transformation. Problem-solving is not a single cognitive skill but a complex process that includes understanding the problem, analyzing the situation, developing solutions, verifying data and implementing solutions, and evaluating and reflecting on the outcomes (Sternberg, 2003). It inherently involves multi-level cognitive processing and contextualized knowledge application.
For sophomore-level university students, the sudden transition from simpler, well-structured introductory courses to the more demanding and complex IBM course is highly challenging (Janson et al., 2020). Case-Based Learning has long been used in business and management education and is considered an important way to develop higher-order thinking, critical analysis, and problem-solving skills (Smith & Miller, 2005). Each case is an English text of about 20 pages, describing the difficulties and environmental changes a multinational company faces at a specific decision point. Students must, under the dual burden of language and professional knowledge, extract key points from lengthy information, establish a problem framework, compare strategic options, and consider cross-cultural communication and international supply chain operating models. Case-based teaching heavily relies on students' reading comprehension and pre-class preparation. If students cannot effectively absorb background information in the early stages, class discussions often become a one-person show for the few with better language skills and professional foundations. Research has shown that in complex problem-solving, if students lack guidance or support, they often resort to superficial information gathering without forming a deep understanding (Loibl, Roll, & Rummel, 2017; Kim & Hannafin, 2011). In case-based teaching, providing support such as information searching, concept clarification, and strategic prompts can help more students participate in in-depth discussions (Merseth, 1991; Huang & Chuang, 2008).
Furthermore, the human resources of teachers and TAs are limited, and traditional in-person TAs find it difficult to provide timely and effective assistance to every student. In the digital age, how to make good use of educational technology tools to reduce TA training costs and expand the scope of support has become a key issue (Lu & Xie, 2023). Technology-mediated learning must be integrated with appropriate teaching strategies to truly enhance students' learning effectiveness in problem-solving.
In the field of education, there has been a long-standing debate between "minimal guidance" and "direct instruction" (Kirschner et al., 2006). The former suggests that students should learn through exploration, figuring out solutions through trial and error. The latter emphasizes that in complex courses, without appropriate and clear guidance, students may lose their way and waste mental resources. Research shows that when tasks are very complex and learners have insufficient prior knowledge, overly free learning or only slight guidance can lead to poor learning outcomes (Sweller, 1988; Kirschner et al., 2006).
The aforementioned theory of "scaffolding," proposed by Wood, Bruner, and Ross (1976), emphasizes that before a learner has sufficient ability to complete a task independently, a teacher or a more capable peer provides appropriate and timely assistance. This support is gradually withdrawn as the learner becomes more proficient. This concept has been frequently applied in digital learning environments (Kim & Hannafin, 2011a; Chen & Bradshaw, 2007). However, traditional scaffolding is either provided by humans (teachers, TAs, peers) or presented statically by a pre-set teaching system, making it difficult to quickly adjust to the diverse needs and dynamic problem understanding of students (Janson et al., 2020). Scaffolding techniques often appear in the form of prompts, examples, cues, and problem decomposition, and are realized through digital environments or tools. For example, a teaching platform can provide timely prompts during a student's problem-solving process to help them break down a complex problem into manageable sub-tasks, or use an intelligent tutoring system to dynamically adjust the level of support based on the student's performance (del Olmo-Muñoz et al., 2022; Harskamp & Suhre, 2007).
Janson et al. (2020) found that designing "technology-enhanced scaffolding" in a flipped classroom for management education significantly affects learners' problem-solving performance and satisfaction with the learning process. Procedural scaffolds help learners understand how to use the Learning Management System (LMS) and resources; metacognitive scaffolds promote self-monitoring and reflection; conceptual scaffolds provide learners with the knowledge cues needed to understand the task; and strategic scaffolds guide learners to consider multiple solutions and evaluate the optimal strategy (Kim & Hannafin, 2011a, 2011b). The scaffolding method breaks down the inquiry process into manageable stages and provides corresponding information or guidance at different stages, so that learners are not overwhelmed by the complexity of the problem from the beginning. When facing complex tasks, if students can receive multiple instructional interventions and technical support, it helps to reduce cognitive load, allowing them to devote their mental resources to strategic thinking and knowledge structure construction.
Educational technology has made many attempts to promote students' problem-solving abilities, including game-based learning, online collaborative tools, intelligent tutoring systems, and integrated physical-virtual learning systems (such as AR/VR scenarios) (Lu & Xie, 2023). When course design can provide systematic guidance, tool usage training, peer interaction opportunities, and scaffolded feedback, students can not only improve their problem analysis and strategy formulation abilities but also cultivate deeper learning motivation and self-regulation. The research by Hwang et al. (2013) shows that in a highly interactive platform, providing step-by-step guidance and prompts allows students to gradually construct a problem-solving process. Past research has mostly focused on the effectiveness of specific digital tools (such as games, simulation systems, LMS) in promoting problem-solving abilities, but has rarely mentioned the intelligent interactive features of Generative AI. Although some studies have affirmed the potential of AI, there is still a lack of empirical research on integrating Generative AI as a scaffold for "intermediate-level business school courses." Previous studies have mostly focused on STEM or medical fields (Frerejean et al., 2023; Costa & Miranda, 2019) or on pure computational thinking and logical reasoning training, with less attention paid to contexts like international business management that require the integration of multiple perspectives. The unique aspect of business courses is that the situations are ever-changing, information is incomplete, and soft factors such as culture and political economy need to be considered. How AI scaffolding should guide students' thinking remains to be explored.
Traditional scaffolding is mostly provided by teachers or pre-designed teaching systems, lacking flexibility and real-time interaction. The rise of Generative AI brings a new opportunity: AI can become a "conversational, adaptive" intelligent scaffold. Students can, through multi-round conversations, continuously ask questions, request more examples or feedback, and examine the quality of AI responses, further promoting deep thinking and data verification awareness.
The core of procedural scaffolds is to provide learners with a guide to the learning environment and tools, as well as clear instructions on task steps (Hannafin et al., 2004). GenAI can act as a "real-time assistant guide" in this aspect. For example, when using an LMS for learning, a learner can ask ChatGPT, "How do I use this resource?" or "What should I do next?" Based on the platform's functions and the information pre-set by the instructional designer, GenAI can provide phased instructions or strategic prompts. Compared to pre-recorded instructional videos or static text descriptions, GenAI can offer more precise procedural guidance based on the learner's current state. For example, if a learner feels lost while analyzing a human resource management process (a real business process case mentioned in the study by Janson et al., 2020), GenAI can immediately respond and provide a step-by-step suggestion like, "First, read the BPMN (Business Process Model and Notation) diagram of the process, then try to identify its weaknesses." This real-time interactivity enhances the effectiveness of procedural scaffolds.
Metacognitive scaffolds aim to help learners monitor, reflect on, and adjust their own learning strategies (Azevedo & Hadwin, 2005; Jumaat & Tasir, 2016). Problem-solving is not just a purely cognitive process; it also involves metacognition, which is the student's monitoring of their own understanding, strategy choices, and learning progress (Azevedo & Hadwin, 2005; Hannafin et al., 2004). GenAI can play the role of a "metacognitive coach" here. For example, a learner can ask GenAI, "I just answered a question in my assignment. Can you help me see if I've touched on the core concepts?" or "I've spent a lot of time searching for information. How can I improve my learning efficiency?" Through dialogue analysis, GenAI can identify the learner's blind spots in understanding and ask guiding questions to prompt them to reflect on their learning process. Additionally, GenAI can suggest that the learner set learning goals and inquire about their achievement in subsequent conversations, thereby promoting self-regulation (Wan, Compeau, & Haggerty, 2012). This dynamic nature of metacognitive scaffolds helps reduce the learner's sense of confusion and increases their satisfaction with the learning process (Gupta & Bostrom, 2013).
Conceptual scaffolds help learners understand the knowledge structure required for a task by providing key concepts, examples, and diagrams (Cagiltay, 2006). In this regard, GenAI can serve as a "real-time knowledge engine." When a learner is confused about a concept, they can ask GenAI, "What is Business Process Reengineering (BPR) in this case study?" or "What's the difference between an 'event' and a 'task' in BPMN?" GenAI can provide concise explanations or, if needed, extend to offer more examples and supplementary materials (Sun et al., 2018). This support model not only reduces the cognitive load of searching for conceptual information (extraneous load) but also allows the learner to focus more on integrating new knowledge into their existing cognitive structure (increasing intrinsic load), thereby promoting deep learning.
Strategic scaffolds encourage learners to explore multiple ways of solving a problem and to evaluate their choices (Kim & Hannafin, 2011b). Here, GenAI can act as a "strategic planning assistant," proposing various solutions to complex problems and prompting learners to think about the pros and cons of different options. For example, a learner could ask, "What are some possible process optimization strategies for improving the human resources recruitment process?" or "What are the risks if I adopt Strategy A?" GenAI can present multiple strategies and help compare them, not only guiding the learner in evaluating options but also helping to shape higher-order thinking skills for problem-solving (Zohar & Dori, 2003). In this process, GenAI can also dynamically adjust its suggestions based on the learner's feedback, creating a truly interactive strategic scaffolding environment.
In recent years, the rapid popularization of Large Language Models (LLMs) like ChatGPT and Claude has made it easy for students to obtain "seemingly professional" answers. However, most students currently view these tools merely as a quick way to get answers. They throw pre-class questions into the AI model and, after receiving a brief response, consider the task done, lacking the motivation for further inquiry. This usage pattern does not create an effective scaffolding effect. Instead, it may lead students to abandon their own thinking and the responsibility of evaluating the accuracy of information, thereby reducing the depth of their learning. In theory, if a set of teaching strategies could be designed to require students to engage in multi-round interactions with AI, asking more specific and pointed questions in each round, requesting sources, and promoting data comparison and verification, then AI could become a true scaffolding tool.
Promoting problem-solving requires the integration of multiple teaching strategies (Lu & Xie, 2023). AI is not just an information provider but also an opportunity to stimulate students to ask the right questions and to train their information literacy. The study by Hwang & Kuo (2011) shows that when students are guided by appropriate scaffolding strategies in a digital learning environment, they can develop more mature mental models for problem-solving, including systematic searching, strategic evaluation, and constructing their own solutions. If this mental model can be reinforced layer by layer through AI interaction, then when students face a case study, they will no longer be passive recipients but will be able to actively propose hypotheses, verify them, revise them, and ultimately form their own unique insights.
A key feature of AI scaffolding is its ability to dynamically adjust the level of guidance. Initially, it can provide more explicit, step-by-step suggestions (similar to direct instruction). Once students have understood the basics, the teacher can require them to reduce the number of times they ask AI for direct answers and instead use AI to gather information for integration, thus reducing the amount of AI guidance and gradually moving students toward independent exploration (Loibl et al., 2017). This model provides a path of "adaptive support—gradual fade-out," allowing AI scaffolding to find a middle ground between "minimal guidance" and "direct instruction." It prevents students from being completely lost while also preventing them from becoming overly reliant on AI answers and losing their ability to think independently. This strategy aligns with the principle of fading scaffolds advocated by Renkl (2002) and is also consistent with the emphasis by Janson et al. (2020) that technology-enhanced teaching should initially play a role in strengthening structure, and as the learner's ability improves, it should gradually recede into a background support role.
In the International Business Management course, language is a major obstacle. The learners are non-native English speakers, and the cases are mostly written in English, which doubles the difficulty. However, AI has multilingual capabilities and can help students deconstruct complex professional literature into easily understandable paragraphs or use bilingual parallel text to allow students to compare the original text with the Chinese explanation generated by AI. This can reduce the cognitive load caused by language, allowing students to focus more on the logical and strategic aspects of the problem. While the application of educational technology often does not heavily discuss language barriers, in the context of this study, the language issue is clearly an important area for applying AI assistance.
Cognitive Load Theory (CLT) states that learning effectiveness depends on whether the learning materials and task requirements are compatible with the learner's limited working memory resources (Sweller, 1988; Paas et al., 2003). In an intermediate-level International Business Management course, students face a high cognitive load from the language of English cases, the intrinsic load of professional concepts, and the extraneous load of searching for information and building problem-solving strategies, which can affect deep learning (Kirschner et al., 2006). The application of Generative AI can be highly beneficial if it can balance cognitive load in the instructional design. For example, in the initial stage, students can ask AI to simplify case descriptions, summarize key points, or present information in a bilingual format to reduce unnecessary extraneous load (Janson et al., 2020). As the course progresses, the teacher can gradually require students to face more complete, unsimplified cases and to re-verify and reflect on the information obtained from AI, thereby gradually improving their ability to handle intrinsic load (Renkl, 2002). At the same time, appropriate prompts and regulations (such as requiring AI to list its thinking steps or information sources when answering) can prevent excessive extraneous load, allowing students to devote more cognitive resources to the core strategic thinking of problem-solving.
In summary, the development of global higher education emphasizes the cultivation of students' problem-solving and higher-order thinking skills. Especially in the field of international business, facing challenges of multiple languages, cultures, and complex cases, how to effectively use educational technology to provide students with timely and dynamic support has become a pressing issue. Previous literature (Janson et al., 2020; Lu & Xie, 2023) has clearly shown that technology-enhanced learning has a positive impact on problem-solving, but the application of Generative AI still requires further in-depth research. The advantage of AI lies in its ability to provide flexible scaffolding through multi-round natural language conversations, benefiting students in information verification, strategy formation, and solution evaluation. However, this model must be supported by carefully designed teaching strategies, including requiring students to deepen their questions, verify information, and engage in metacognitive reflection and team discussions.