学生与AI协同的科研选题与构思:方向探索、意图对齐、人机分工、问题化、可行性与认知负担
无明确想法时的方向探索与选题判断依据:新手与经验者的认知差异
合并三份分组中所有讨论“方向从何而来、依据什么判断是否值得继续”的文献,回答第1问。共同点是把选题视为可被刻画、可被支持的认知过程而非灵感事件:以读者/课程/文本/阅读行动等因素框架与选题准则(criteria_to_select_a_research_topic、conceptualization_and_categorization_of_topic_selection)界定“该看哪些线索”,以研究生的数据驱动决策与视觉化选题认知模式(data_driven_decision_making_in_graduate_students)刻画实际判断过程;同时以教师/评审者的“注意(noticing)”能力与临场决策质量对比(teachers_noticing_skills_during_geography_instruction)和专家—新手文化共识差异(variation_in_cultural_consensus_between_expert)说明经验水平如何改变判断依据——新手依赖可共享的显性共识与既有模型,专家则发展出个人化、情境适应的模型。四类证据共同构成“探索行为—外部线索—经验差异”的选题判断基础。
- Criteria to select a research “topic” for postgraduate and doctoral studies in ayurveda(S. Dornala, S. Dornala, 2024, Journal of Ayurveda and Integrative Medicine)
- Conceptualization and Categorization of Topic Selection Factors in Thematic Integrated Reading : A Literature-Based Qualitative Analysis and Delphi Study(Y. Seo, Seonho Park, 2026, The Korean Association of General Education)
- Data-driven decision making in graduate students' research topic selection(Qiao Li, Ping-Hsuan Wang, Yifan Sun, Ying-Long Zhang, Chuan-Fu Chen, 2019, Aslib Journal of Information Management)
- Teachers’ Noticing Skills during Geography Instruction: An Expert-Novice Comparison(Nina Scholten, Sandra Sprenger, 2020, Journal of Geography)
- Variation in Cultural Consensus Between Expert and Novice Brazilian Jiu Jitsu Athletes(K. Bennett, W. Dressler, 2020, No journal)
澄清提问、结构化表征与可解释交互:让人和AI逐步形成并纠正共同理解
合并三份分组中的意图对齐线索,回答第2问“学生说不清时如何建立共同理解、如何发现并纠正AI的理解偏差”。组内文献构成一条“提问—表征—验证—用户反应”的链条:以外部知识(事件图式)与检索文档驱动高质量澄清问题,避免AI靠猜测或重复无效追问(leveraging_event_schema_to_ask_clarifying、generating_intent_aware_clarifying_questions_in_conversational);提供澄清问题自动生成与多视角评估框架,使学生可检验AI的理解假设(agent_cq_automatic_generation_and_evaluation_of);用大规模用户研究证明澄清问题的质量直接决定绩效与满意度、低质量提问反而有害(users_meet_clarifying_questions_toward_a);并以类比推理+思维导图结构化表征并追溯想法演化(ideationweb_tracking_the_evolution_of_design),让“我到底想研究什么”变成可查看、可修正的对象。整体回答“提问而非猜测、显式表征而非隐式推断”的对齐机制。
- Leveraging Event Schema to Ask Clarifying Questions for Conversational Legal Case Retrieval(Bulou Liu, Yi-Ran Hu, Qing-Yao Ai, Yi-Qun Liu, Yue-Yue Wu, Chen-Liang Li, Wei-Xing Shen, 2023, International Conference on Information and Knowledge Management)
- Generating Intent-aware Clarifying Questions in Conversational Information Retrieval Systems(Ziliang Zhao, Zhi-Cheng Dou, Yu-Jia Zhou, 2024, International Conference on Information and Knowledge Management)
- AGENT-CQ: Automatic Generation and Evaluation of Clarifying Questions for Conversational Search with Large Language Models(Clemencia Siro, Yifei Yuan, Mohammad Aliannejadi, M. de Rijke, 2026, ACM Transactions on Information Systems)
- Users Meet Clarifying Questions: Toward a Better Understanding of User Interactions for Search Clarification(Jie Zou, Mohammad Aliannejadi, E. Kanoulas, Maria Soledad Pera, Yi-Qun Liu, 2022, ACM Transactions on Information Systems)
- IdeationWeb: Tracking the Evolution of Design Ideas in Human-AI Co-Creation(Hanshu Shen, Lyukesheng Shen, Wenqi Wu, Kejun Zhang, 2025, International Conference on Human Factors in Computing Systems)
人机分工边界与学术主体性保持:学生先行、AI协助、依赖风险防范
合并三份分组中所有关于“谁负责什么、AI何时主动介入、如何不越界”的文献,回答第3问并衔接第6问的主体性判断。组内给出四条互补的划界依据:以“负责任学术主体性”框架明确认知所有权、认识论核查与归属责任必须保留在学生一侧(from_ai_assistance_to_responsible_academic);以“论文先行”的手工逻辑图要求学生在接触AI前完成认知劳动,从根本上确定介入时机(paper_first_knowledge_engineering_as_a_pedagogical);以流程化陪伴与同意机制在计算教育中维护学生能动性(prosespark_preserving_student_agency_in_computing);以共创模式与AI参与程度的实验揭示心理所有权、感知能动性如何影响创造性与作者身份归属(the_impact_of_human_ai_co_creation_on);并从反面证明被动依赖AI会削弱自我效能与工作意义、而协作式使用可保持能动性(relying_on_ai_at_work_reduces)。正反两面共同界定分工原则与主动性介入的设计底线。
- From AI assistance to responsible academic agency: A human-centered framework for generative artificial intelligence in higher education(Benjamin Batte, 2026, Journal of Progress in Artificial Intelligence)
- Paper-first knowledge engineering as a pedagogical framework for cognitive agency in AI-assisted learning(G. W. Collins, 2026, Contemporary Educational Technology)
- ProseSpark: Preserving Student Agency in Computing Writing with an Agentic AI Companion(Víctor Brenes, 2026, Proceedings of the 2nd ACM Virtual Global Computing Education Conference V.2)
- The impact of human-AI co-creation on perceived creativity in generative AI art: a serial mediation mechanism of psychological ownership, perceived agency, and creative flow(Chen-Yang Wang, Miao Chen, Wen Yang, 2026, Frontiers in Psychology)
- Relying on AI at work reduces self-efficacy, ownership, and meaning while active collaboration mitigates the effects(E. Lee, Y. Yin, N. Jia, Cheryl J. Wakslak, 2026, Scientific Reports)
从宽泛兴趣到具体研究问题:问题表述训练、支架式构思与元认知支撑
合并三份分组中关于“兴趣如何变成可研究问题、卡在哪里、多个候选问题如何比较与修改”的文献,回答第4问。组内既包含问题化的理论模型与能力测评——信息素养视角的问题形成理论模型给出从背景问题到前景问题的路径(question_formulation_for_information_literacy_theory),医学研究生的实证测评定位能力短板(research_question_formulation_competence_among_medical);也包含可直接落地的干预:问题表述技巧(PICO/QFT)的具体实施步骤(implementation_of_the_question_formulation_technique)、AI 支架式规划对研究主题清晰性—可行性—契合度的提升效果(effect_of_artificial_intelligence_ai_scaffolded_planning);并延伸至AI把方向展开为多个候选方案的构思实验,用于比较创造力与多样性(design_ideation_through_large_language_model_driven)。整体呈现“理论模型—能力诊断—技能训练—AI支架—候选比较”的问题化路径。
- Question formulation for information literacy: Theory and practice(Davida Scharf, J. Dera, 2021, The journal of academic librarianship)
- RESEARCH QUESTION FORMULATION COMPETENCE AMONG MEDICAL STUDENTS: DIAGNOSTIC ASSESSMENT(O. M. Sherekhova, 2026, Lifelong education the XXI century)
- Implementation of the Question Formulation Technique as a Teaching Strategy in Renewable Energy Engineering Education(Claudia Torres Garibay, Jessica Kerby, Andrew P. Minigan, 2020, No journal)
- Effect of Artificial Intelligence (AI)-Scaffolded Planning Tools on Research Question Formulation Among Students with Mild Intellectual Disabilities: A Randomized Controlled Trial Analysis(U. Ofem, P. Anake, N. N. Anyin, N. Nsan, M. Okon, C. Abuo, Stephen Olonye, Joy Adie, Vera Mpuon Obibessong, J. Ukatu, Agede Ambor Ogar, A. Ebaye, A. Egu, E. George, Godwin Amuchi Odey, 2025, Journal of Intellectual Disability - Diagnosis and Treatment)
- Design ideation through large language model-driven design operation: a case study of architectural design using pattern language(Daichi Tanaka, Yutaka Nomaguchi, K. Fujita, 2026, Design Science)
研究想法的质量、价值与可行性判断:AI与人的筛选依据及其可靠性
合并三份分组中关于“问题值不值得做、能不能做、依据什么证据判断”的文献,回答第5问。组内形成三层判断资源:一是价值与质量层面,比较LLM与新手设计者所提想法在清晰度、多样性、可执行性与质量上的差异并检验AI能否替代人工筛选(comparative_analysis_of_human_and_large),梳理构思三阶段流程与评价指标、伦理限制,为“值得做”提供评估框架(analysis_of_accelerating_ideation_process_with);二是判断本身的可靠性层面,以随机对照试验检验培训与经验水平如何影响筛选决策的信度与效度(training_and_experience_in_study_selection);三是新颖性—可行性—有效性权衡与外部约束匹配层面,以多维奖励学习生成研究想法(ldc_learning_to_generate_research_idea),并把资助机会、合作伙伴、资源等可行性约束纳入工具化评估(leveraging_open_big_data_from_rd)。共同覆盖“价值—质量—可靠性—资源可行性”四类依据。
- Comparative Analysis of Human and Large Language Model Generated Product Ideas under Practical Constraints(N. Pathinayake, Charitha Darshana Makavita, D. M. Randima Dinalankara, 2026, Moratuwa Engineering Research Conference)
- Analysis of Accelerating Ideation Process with Large Language Model(Yuhan Wei, 2024, Applied and Computational Engineering)
- Training and Experience in Study Selection (TESS): study protocol for a pilot randomised trial within a systematic review(E. Ahern, T. Adedeji, A. Whiston, S. Dillon, F. Lynn, 2025, HRB Open Research)
- LDC: Learning to Generate Research Idea with Dynamic Control(Ruochen Li, Li-Qiang Jing, Chi Han, Jia-Wei Zhou, Xin-Ya Du, 2026, Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing)
- Leveraging Open Big Data from R&D Projects with Large Language Models(D. Ruiz, Y. Cardinale, A. Casas, V. Moscardo, 2025, Big Data and Cognitive Computing)
降低认知负荷与工具负担的设计与评估:让工具促进思考而不替代判断
合并三份分组中所有关于“如何减少理解信息与使用工具的负担、如何判断工具是否真正帮助了思考”的文献,回答第6问。组内从设计规范与效果评估两侧互补:认知负荷理论在界面设计中的分解与应用(内在/外在/相关负荷、视觉层级、内容分块)给出降负原则(cognitive_load_optimization_in_user_interface),面向教育界面的认知卸载与冗余噪声削减实验提供机制证据(neuroergonomics_of_educational_interfaces_a_study),以强化学习动态难度调整配合教学对话代理维持投入、避免挫败(adaptive_cognitive_load_model_and_pedagogical);评估侧则以感知控制与心理所有权解释为何“可控共创”优于“自动替代”(designing_for_meaning_in_aigc_systems),并检验AI脚手架的介入是否真正帮助学生产生自己的想法而非替代思考(impact_of_ai_assistance_on_student)。由此同时回答“负担是否降低”和“学生是否仍在主动判断与选择”这两个测量问题。
- Cognitive Load Optimization in User Interface Design for Info Services(Zaed Balasm, S. S. Rajan, C. Karpagham, Muhidinov Ayubbek Nuritdinovich, Sardorbek Yusupov, Shyam Maurya, 2025, Indian Journal of Information Sources and Services)
- Neuroergonomics of Educational Interfaces: A Study of Cognitive Biases in Modern Educational Platforms(A. Yakubov, Anna Ubaydullaeva, Mirzokhid Musaev, M. Mukhiddinova, Jamshid Vosiev, R. Temirov, 2025, 2025 5th International Conference on Technology Enhanced Learning in Higher Education (TELE))
- Adaptive Cognitive Load Model and Pedagogical Conversational Agent for Preschoolers’ Interactive Training(Yu-Qin Mu, 2026, Proceedings of the 2026 3rd International Conference on Innovation Management and Information Systems)
- Designing for Meaning in AIGC Systems: The Control–Ownership Pathway in Human–AI Co-Creation(Yu-Liang Feng, Ya-Qin You, Shing-Sheng Guan, 2026, IEEE Access)
- Impact of AI assistance on student agency(A. Darvishi, Hassan Khosravi, S. Sadiq, D. Gašević, G. Siemens, 2023, Comput. Educ.)
本报告将三份初始化分组中的文献合并去重后,按用户提出的六个问题组织为六个相互并列、互不包含的一级分组,共 30 篇近五年文献,形成一条“选题判断—意图对齐—人机分工—问题化—可行性评估—负荷与主体性设计”的完整链路。第一组(5篇)以选题准则与影响因素框架、研究生选题认知模式、教师/评审者的注意能力对比及专家—新手文化共识差异,回答“方向从何而来、依据什么线索判断值得继续”,并给出经验差异的依据。第二组(5篇)以事件图式与检索驱动的澄清提问生成、澄清问题自动生成与评估、大规模用户实证以及想法演化的结构化表征与追溯,回答“学生说不清时如何与AI形成并修正共同理解”。第三组(5篇)以负责任学术主体性框架、“论文先行”的认知劳动前置、流程化陪伴保主体性、共创参与度对心理所有权与作者身份的实验证据,以及被动依赖削弱自我效能的负面证据,界定“学生负责什么、AI协助什么、AI何时可以主动介入”。第四组(5篇)以问题形成的信息素养理论模型、研究问题形成能力测评、问题表述技巧实施、AI支架式规划与LLM驱动候选方案展开,回答“宽泛兴趣如何转化为可比较、可修改的研究问题”。第五组(5篇)以人机想法质量比较、构思流程评价指标、筛选判断的信度效度实验、多维新颖性—可行性权衡学习及资源约束匹配工具,构建“值得做/做得了”的判断依据。第六组(5篇)以认知负荷界面设计原则、教育界面认知卸载机制、动态难度调节、感知控制与心理所有权、AI脚手架效果评估,回答“如何降低负担并判断学生是否仍在主动思考”。整体上呈现两条主线:一条是“人”的认知与主体性(兴趣判断、问题生成、价值与可行性判断、作者身份),另一条是“AI”的介入机制(澄清提问、显式表征、分阶段分工、降负交互与可控共创);两条主线在多数文献中通过“控制感—心理所有权—主体性”链条相互衔接,可支撑一个以学生主导、AI辅助的科研选题与构思流程设计。
合并三份分组中所有讨论“方向从何而来、依据什么判断是否值得继续”的文献,回答第1问。共同点是把选题视为可被刻画、可被支持的认知过程而非灵感事件:以读者/课程/文本/阅读行动等因素框架与选题准则(criteria_to_select_a_research_topic、conceptualization_and_categorization_of_topic_selection)界定“该看哪些线索”,以研究生的数据驱动决策与视觉化选题认知模式(data_driven_decision_making_in_graduate_students)刻画实际判断过程;同时以教师/评审者的“注意(noticing)”能力与临场决策质量对比(teachers_noticing_skills_during_geography_instruction)和专家—新手文化共识差异(variation_in_cultural_consensus_between_expert)说明经验水平如何改变判断依据——新手依赖可共享的显性共识与既有模型,专家则发展出个人化、情境适应的模型。四类证据共同构成“探索行为—外部线索—经验差异”的选题判断基础。相关文献:S. Dornala et. al, 2024 等 5 篇文献
合并三份分组中的意图对齐线索,回答第2问“学生说不清时如何建立共同理解、如何发现并纠正AI的理解偏差”。组内文献构成一条“提问—表征—验证—用户反应”的链条:以外部知识(事件图式)与检索文档驱动高质量澄清问题,避免AI靠猜测或重复无效追问(leveraging_event_schema_to_ask_clarifying、generating_intent_aware_clarifying_questions_in_conversational);提供澄清问题自动生成与多视角评估框架,使学生可检验AI的理解假设(agent_cq_automatic_generation_and_evaluation_of);用大规模用户研究证明澄清问题的质量直接决定绩效与满意度、低质量提问反而有害(users_meet_clarifying_questions_toward_a);并以类比推理+思维导图结构化表征并追溯想法演化(ideationweb_tracking_the_evolution_of_design),让“我到底想研究什么”变成可查看、可修正的对象。整体回答“提问而非猜测、显式表征而非隐式推断”的对齐机制。相关文献:Bulou Liu et. al, 2023 等 5 篇文献
合并三份分组中所有关于“谁负责什么、AI何时主动介入、如何不越界”的文献,回答第3问并衔接第6问的主体性判断。组内给出四条互补的划界依据:以“负责任学术主体性”框架明确认知所有权、认识论核查与归属责任必须保留在学生一侧(from_ai_assistance_to_responsible_academic);以“论文先行”的手工逻辑图要求学生在接触AI前完成认知劳动,从根本上确定介入时机(paper_first_knowledge_engineering_as_a_pedagogical);以流程化陪伴与同意机制在计算教育中维护学生能动性(prosespark_preserving_student_agency_in_computing);以共创模式与AI参与程度的实验揭示心理所有权、感知能动性如何影响创造性与作者身份归属(the_impact_of_human_ai_co_creation_on);并从反面证明被动依赖AI会削弱自我效能与工作意义、而协作式使用可保持能动性(relying_on_ai_at_work_reduces)。正反两面共同界定分工原则与主动性介入的设计底线。相关文献:Benjamin Batte et. al, 2026 等 5 篇文献
合并三份分组中关于“兴趣如何变成可研究问题、卡在哪里、多个候选问题如何比较与修改”的文献,回答第4问。组内既包含问题化的理论模型与能力测评——信息素养视角的问题形成理论模型给出从背景问题到前景问题的路径(question_formulation_for_information_literacy_theory),医学研究生的实证测评定位能力短板(research_question_formulation_competence_among_medical);也包含可直接落地的干预:问题表述技巧(PICO/QFT)的具体实施步骤(implementation_of_the_question_formulation_technique)、AI 支架式规划对研究主题清晰性—可行性—契合度的提升效果(effect_of_artificial_intelligence_ai_scaffolded_planning);并延伸至AI把方向展开为多个候选方案的构思实验,用于比较创造力与多样性(design_ideation_through_large_language_model_driven)。整体呈现“理论模型—能力诊断—技能训练—AI支架—候选比较”的问题化路径。相关文献:Davida Scharf et. al, 2021 等 5 篇文献
合并三份分组中关于“问题值不值得做、能不能做、依据什么证据判断”的文献,回答第5问。组内形成三层判断资源:一是价值与质量层面,比较LLM与新手设计者所提想法在清晰度、多样性、可执行性与质量上的差异并检验AI能否替代人工筛选(comparative_analysis_of_human_and_large),梳理构思三阶段流程与评价指标、伦理限制,为“值得做”提供评估框架(analysis_of_accelerating_ideation_process_with);二是判断本身的可靠性层面,以随机对照试验检验培训与经验水平如何影响筛选决策的信度与效度(training_and_experience_in_study_selection);三是新颖性—可行性—有效性权衡与外部约束匹配层面,以多维奖励学习生成研究想法(ldc_learning_to_generate_research_idea),并把资助机会、合作伙伴、资源等可行性约束纳入工具化评估(leveraging_open_big_data_from_rd)。共同覆盖“价值—质量—可靠性—资源可行性”四类依据。相关文献:N. Pathinayake et. al, 2026 等 5 篇文献
合并三份分组中所有关于“如何减少理解信息与使用工具的负担、如何判断工具是否真正帮助了思考”的文献,回答第6问。组内从设计规范与效果评估两侧互补:认知负荷理论在界面设计中的分解与应用(内在/外在/相关负荷、视觉层级、内容分块)给出降负原则(cognitive_load_optimization_in_user_interface),面向教育界面的认知卸载与冗余噪声削减实验提供机制证据(neuroergonomics_of_educational_interfaces_a_study),以强化学习动态难度调整配合教学对话代理维持投入、避免挫败(adaptive_cognitive_load_model_and_pedagogical);评估侧则以感知控制与心理所有权解释为何“可控共创”优于“自动替代”(designing_for_meaning_in_aigc_systems),并检验AI脚手架的介入是否真正帮助学生产生自己的想法而非替代思考(impact_of_ai_assistance_on_student)。由此同时回答“负担是否降低”和“学生是否仍在主动判断与选择”这两个测量问题。相关文献:Zaed Balasm et. al, 2025 等 5 篇文献
总计 95 篇相关文献
Research topic selection is a complex process for novice researchers, often characterized by directional ambiguity and structural fragmentation. While Generative AI (GAI) offers potential support, general-purpose tools frequently fail to align students’ tacit intuitions with academic requirements, tending to yield generic feedback. To address this gap, we conducted semi-structured interviews with 33 novice researchers, deriving five distinct user expectations for GAI-supported guidance. Responding to these expectations, we propose a GAI-supported Topic Selection system grounded in Nonaka’s SECI (Socialization-Externalization-Combination-Internalization) knowledge creation model. The system operationalizes the concept of "Ba" within an interactive digital space, orchestrating four modules — Originating, Dialoguing, Systematizing, and Exercising — to facilitate knowledge conversion. Rather than providing direct answers, the mediating agent offers adaptive cognitive scaffolding. This work suggests the value of transitioning AI-supported education from content generation to process guidance, fostering independent research reasoning.
Purpose With the advent of the intelligent environment, as novice researchers, graduate students face digital challenges in their research topic selection (RTS). The purpose of this paper is to explore their cognitive processes during data-driven decision making (DDDM) in RTS, thus developing technical and instructional strategies to facilitate their research tasks. Design/methodology/approach This study developes a theoretical model that considers data-driven RTS as a second-order factor comprising both rational and experiential modes. Additionally, data literacy and visual data presentation were proposed as an antecedent and a consequence of data-driven RTS, respectively. The proposed model was examined by employing structural equation modeling based on a sample of 931 graduate students. Findings The results indicate that data-driven RTS is a second-order factor that positively affects the level of support of visual data presentation and that data literacy has a positive impact on DDDM in RTS. Furthermore, data literacy indirectly affects the level of support of visual data presentation. Practical implications These findings provide support for developers of knowledge discovery systems, data scientists, universities and libraries on the optimization of data visualization and data literacy instruction that conform to students’ cognitive styles to inform RTS. Originality/value This paper reveals the cognitive mechanisms underlying the effects of data literacy and data-driven RTS under rational and experiential modes on the level of support of the tabular or graphical presentations. It provides insights into the match between the visualization formats and cognitive modes.
A research topic is a subject or problem that an investigator is interested in research. The starting point of any research project that is successful is a well-defined subject of research. The selection of topics is an ongoing approach in which investigators analyse, describe, classify, and refine their ideas. For the ease of selection of topics particularly in the field of Ayurveda, Authors have created an acronym ‘TOPIC’ as a guiding tool for the selection of the area of research. The components of the TOPIC criterion are Textual & Time-bound; Objectivity & Outcome measures; Planning & Pragmatic; Innovative & Interesting and Contemporarily relevant & Consent, highlight useful points for the easy selection of the topic for the research by the novice.
This study aimed to conceptualize and structure the factors influencing topic selection in thematic integrative reading and to examine how these factors function throughout the reading process and shape learners’ understanding. To achieve this purpose, a literature-based qualitative content analysis was conducted on 53 domestic and international studies published between 2000 and 2024. The analysis identified four overarching factors influencing topic selection in thematic integrative reading: reader-related factors, curriculum-related factors, text-related factors, and reading-action factors. These factors function by shaping learners’ entry conditions and engagement, aligning reading activities with curricular goals, organizing principles for text sets, and structuring the cognitive and metacognitive actions required during the reading process. To establish the conceptual validity of the proposed factor structure, a two-round Delphi survey was conducted with seven experts in the field. All four factors met the predefined consensus criteria and were confirmed as the final factor structure. This study reconceptualizes topic selection not as a preliminary step, but as a core theoretical and instructional design process that organizes thematic integrative reading. The proposed framework provides a theoretical foundation for future empirical research and instructional design related to thematic integrative reading. This study is significant in that it reconceptualizes topic selection not as a preliminary procedural step in thematic integrated reading, but as a core theoretical construct that organizes reading design. Furthermore, the factor framework derived from this study can serve as both a theoretical reference and a practical analytical tool for examining the validity of topic selection in instructional design, textbook development, and research on reading education.
ABSTRACT The first step in many text-as-data studies is to find documents that address a specific topic within a larger document set. Researchers often rely on simple keyword searches to do this, even though this may introduce considerable selection bias. Such bias may be even greater when researchers lack the domain knowledge required to make informed search decisions, for example, in cross-national research or research on unfamiliar social contexts. We propose expert-informed topic modeling (EITM) as a hybrid approach to tackle this problem. EITM combines the validity of external domain knowledge captured through expert surveys with probabilistic topic models to help researchers identify subsets of documents that cover initially unknown domain-specific topics, such as specific events and debates, that belong to a researcher-defined master topic. EITM is a flexible and efficient approach to the thematic selection of documents from large text corpora for further study. We benchmark and validate the method by discovering blog posts that address the public role of religion within large corpora of Australian, Swiss, and Turkish blog posts and provide researchers with a complete workflow to guide the application of EITM in their own work.
No abstract available
Abstract In supporting their students’ learning in the classroom, noticing is an important professional skill for teachers, encompassing perceiving and interpreting relevant incidents, as well as ad hoc decision-making. While noticing is an established research topic in other domains, it remains largely neglected in relation to the geography teacher. The paper compares the noticing skills of geography teachers at three different stages of their professional development. Qualitative content analysis of teachers’ think-aloud responses to video vignettes of geography teaching revealed that the three groups differed in terms of number of incidents perceived, interpretive stance, and decision-making.
In agreement with conceptual understanding, identifying problem-solving skills is an important topic in physics. The purpose of this research was to explore students' problem-solving skills based on expert and novice categories. The research used a mixedmethod approach with an explanatory model with 34 prospective physics teachers at Jambi University, Indonesia. The data was obtained by using open-ended questions. Data were collected using tests and interviews. Data analysis was done by using the rubric for physics problem-solving skills. Problem-solving skills were classified into expert and novice. The research showed that most students have problem-solving skills by novice categories. The cause is the weak application of concepts in the problem-solving process. The research concludes application of the concept in the process of problem-solving not well understood by most students be the reason most students have categories novice ability in problemsolving.
Background Systematic reviews can be resource-intensive and require timely completion, yet limited availability of experienced reviewers often necessitates incorporating novice members into review teams. The purpose of this Study Within A Review (SWAR) will be to determine whether training and level of experience within the screening pair affects the reliability of decisions made by novice screeners during study selection for a systematic review Methods A 2(training: task-specific, minimal guidance) x 2(experience level of screening partner, ‘Reviewer 1’: moderate experience, minimal experience) pilot randomised trial will be conducted within a host systematic review in the topic area of depression and psychosocial functioning. Participants ( N = 12), consisting of higher education students with no prior experience in evidence synthesis, will be randomised to one of the four conditions to complete a standardised study selection task at title/abstract level ( k = 219 records) on Covidence systematic review screening software, blindly and independently. Total participation time is estimated at 5 hours. Screening decisions made by participants will be assessed for reliability against the consensus-based decisions by two reviewers with content and methodological expertise (expert standard), through calculation of chance-corrected Cohen’s kappa and percentage of agreement, then compared across the conditions. Secondary outcomes will include reliability within the screening pair (participant and allocated screening partner), validity of screening decisions (false positives, false negatives, sensitivity, specificity), feasibility measures, including time taken to complete the study selection task and success of blinding, as well as acceptability. Conclusions Findings will be used to inform the design of subsequent trial work to determine the efficacy of training and screener pairing for study selection. Ultimately, these insights will help to build capacity among novice screeners to engage with evidence synthesis and work alongside experienced review teams. Registration Northern Ireland Hub for Trials Methodology Research SWAR Registry: SWAR 38.
The practice of martial arts for leisure and exercise is growing rapidly in the United States. In this study, we utilize mixed methods and cultural consensus analysis to examine a cultural model of strategy among Brazilian jiu jitsu athletes in a gym in Atlanta, Georgia, and to explore how that cultural knowledge varies within consensus. This study builds upon previous research in martial arts studies by employing cultural domain analysis and ethnographic research to quantify and link culture to embodied knowledge through a cultural model of strategy. Results show that strategy in Brazilian jiu jitsu revolves around the hierarchy of positional dominance. Low level belt-ranked novice athletes utilize shared understandings of positional dominance to guide their personal strategic selection of techniques, which they habitually practice to the point of embodied reflexivity. High belt-ranked expert athletes eventually reach a level of expertise at which relying exclusively on the cultural model of strategy impedes their performance in competition. These expert athletes develop a personal model of strategic fluidity within the context of the cultural model of strategy which focuses on adaptation to specific opponents and circumstances.
Abstract. Large Language Models (LLMs) have rapidly advanced, demonstrating near-human intelligence in language comprehension, problem-solving, and autonomous decision-making, which can be leveraged in scientific discovery. Previous research of LLM for science often focuses on experiment execution part and the first stage of research, i.e., idea generation with large language models is lack of research. This paper fills this gap by surveying the use of LLM-based agent systems in research idea generation. This research first proposes three stages in research idea generation pipeline, pre-ideation (knowledge preparation), ideation (generation and iteration), post-ideation (evaluation), and provide detailed summary in different methods in each stage, then it summarizes different metrics and angles of evaluating generated research ideas, finally it discusses limitations and ethical concerns of existing works and suggest potential solutions and future directions. These results aim to provide a solid foundation for future research in improving LLM-based ideation systems and fostering responsible AI usage in scientific discovery.
Abstract This research proposes a systematic method for design ideation with large language models (LLMs), grounded in the design operation model inspired by Christopher Alexander’s pattern language (PL). Design operation refers here to a tree of thought for the step-by-step definition of the design context behind a design problem, the functional requirements, the physical attributes of alternative solutions and the generation of design alternatives that synthesize those attributes. To examine how the design operation improves LLM performance for design ideation, we implemented an architectural design ideation case study and compared four prompt methods for generating design alternatives. The four prompt methods are designed with and without the design operation and the PL. Respective prompts generated 100 design alternatives which were evaluated both by human experts and through computational diversity metrics. The results showed that prompts using the PL tend to generate design alternatives whose creativity is highly rated by humans, but are strongly influenced by the given knowledge and lack diversity, whereas prompts incorporating the design operation have the potential to enhance validity and feasibility and attribute diversity.
No abstract available
Artificial intelligence (AI) chatbots powered by large language models (LLMs) such as ChatGPT are rapidly gaining popularity as labour-augmenting tools. This paper is for sociologists seeking to make the best use of this technology in their work. It presents a practice-oriented framework for using AI chatbots in sociology, building on considerations of the technical conditions of LLMs to introduce both a task categorization and the concept of a ‘knowledge funnel’. This model illustrates the relationship between the scope of knowledge and accuracy in outputs to guide sociologists in evaluating the reliability and applicability of AI-generated content in their research. The main argument driving this article is to establish a paradigm of ‘augmented sociology’ that focuses on human–AI interaction and understands LLMs as a resource rather than as a replacement. This augmentation manifests itself clearly in dialogic ideation, enhancing research by bridging domains, and broad methodological assistance. The paper's primary contribution lies in introducing specific terminologies and actionable strategies for sociologists to integrate LLM chatbots creatively and effectively in their work, filling a significant gap in the current academic understanding of generative AI's role in sociology.
Product ideation and screening are resource-intensive, high-cognitive-demand activities, while traditional ideation and screening methods often experience various challenges. LLMs offer opportunities to overcome these challenges by improving the processes. This study compared product ideas generated by five free LLM versions using zero-shot and tree-of-thought prompting and novice designers, all given similar guidelines. Ideas were assessed using grammar, word count, top words, Universal Sentence Encoder, Self-BLEU, and Type-Token Ratio. Screening considered workability, relevance, specificity, and idea quality, with results evaluated by Krippendorff's alpha, Pearson’s correlation, and the first quartile of average idea quality. LLMs produced ideas with greater clarity, elaboration, and vocabulary, while novice designers generated more diverse ideas, though none of the novice designers’ ideas ranked in the top quartile of average idea quality. Gemini emerged as the best LLM for idea quality, and zero-shot prompting was the most effective strategy. No LLM fully simulates human performance in screening. The study results highlight the potential of LLMs for product ideation and screening, as well as the importance of prompt engineering, fine-tuning, and human collaboration to improve idea quality.
Recent studies have highlighted the potential of Large Language Models (LLMs) to become experts in specific areas of knowledge through the utilization of techniques that enhance their context. Nevertheless, an interesting and underexplored application in the literature is the creation of an LLM that specializes in research projects, as it could streamline the process of project ideation and accelerate the advancement of research initiatives. In this regard, the aim of this work is to develop a tool based on LLM technology capable of assisting the employees of technology centers in answering their queries related to research projects funded under the Horizon 2020 program. By facilitating the identification of suitable funding calls and the formation of consortia with partners meeting specific requirements, tasks that are traditionally time-intensive, the proposed tool has the potential to improve operational efficiency and enable technology centers to allocate their resources more effectively. To improve the model’s baseline performance, context extension techniques such as Retrieved Augmented Generation (RAG) and prompt engineering were explored. Specifically, different RAG approaches and configurations, along with a specialized prompt, were tested on the LLaMA 3 70B model, and their results were compared to those obtained without context extension. The proposed evaluation metrics, which aligned with human judgment while maintaining objectivity, revealed that RAG systems outperformed the standalone LLaMA 3 70B, achieving a rate of optimal responses of up to 46% compared to 0% for the baseline model. These findings emphasize that integrating RAG and prompt engineering pipelines into LLMs can address key limitations, such as generating accurate and informative answers. Moreover, this study demonstrates the practical feasibility of leveraging advanced LLM configurations to support research-driven organizations, highlighting a pathway for the further development of intelligent tools that enhance productivity and foster innovation in the research domain.
No abstract available
Large language models (LLMs) are increasingly used as interactive systems rather than single-turn text generators. However, most prompt engineering paradigms remain oriented toward linear or reactive interactions and offer limited support for persistent state, structured workflows, or explicit user control. This paper introduces program simulation prompting, a prompting paradigm in which an LLM is instructed to simulate the behavior of a structured software program using natural language interaction alone. In this approach, the model adopts a persistent program identity, presents menu-driven interaction modes, manages explicit task state, and constrains behavior to a bounded domain. We analyze three domain-diverse program simulation prompts—spanning startup ideation, culinary recipe development, and creative writing—using a qualitative comparative methodology. Across these case studies, we identify recurring design patterns, including main menu abstractions, explicit separation between creating new and continuing existing artifacts, progressive elaboration, summarization-based state compression, and transparent signaling of memory constraints. Based on these findings, we propose a general, reusable framework for program simulation prompting that decomposes prompt design into five layers: initialization, interaction, state management, output constraints, and user control mechanisms. Our results demonstrate that program simulation prompting enables interactive, stateful, and user-controlled LLM behavior without fine-tuning, external tools, or autonomous agents. This paradigm bridges conversational AI and traditional software affordances, offering a lightweight and transparent alternative to agent-based systems. We conclude by discussing implications for LLM usability, system design, and future hybrid approaches.
The development of Multimodal Large Language Model (MLLMs) offers new technological support for cultivating design thinking and innovation capability in medical education. However, the current training of medical professionals remains predominantly centered on knowledge memorization and one-way didactic instruction. The systematic integration of artificial intelligence and innovation methodologies is still insufficient, while challenges such as limited interdisciplinary integration and inefficient teaching iteration have constrained the cultivation of innovative literacy. To address these challenges, this study constructs the “MLLM+EDIPT” integration framework, which deeply couples the design thinking model from Stanford University's D.school with MLLM technology. It systematically elucidates the cognitive support mechanisms of MLLM across the stages of empathy, definition, ideation, prototyping, and testing. Targeting diverse stakeholders, including hospitals, universities, educators, and students, this study proposes a phased cultivation strategy and competency framework based on school-clinician collaboration. This framework emphasizes the full integration of the “human-centered” philosophy, leveraging AI to enhance situational awareness, feedback timeliness, and methodological rigor, thereby driving the transformation of teaching models from experience-driven to intelligent collaboration. Ultimately, this research aims to provide a theoretically grounded and practically viable pathway reference for the cultivation of top-tier innovative medical talents in the AI era.
The rapid growth of published research literature has made manual, unaided reading a bottleneck for students and early-stage researchers, who must extract structured understanding from unstructured PDF documents while operating at varying levels of comprehension. This paper presents the AI Research Partner, a full-stack MERN (MongoDB, Express.js, React, Node.js) web platform that unifies the research-reading workflow — comprehension, synthesis, and ideation — into a single authenticated system. The platform ingests a PDF, extracts its text, and uses the Google Gemini large language model to generate multi-level (basic, medium, technical) section summaries, an interactive D3.js concept knowledge graph, novelty-rated research ideas, citation recommendations, auto-generated quizzes, and abstract/slide drafts, while a Socket.io-based real-time layer enables collaborative annotation among multiple users. The system was implemented end-to-end, evaluated through functional testing across eight modules, and benchmarked for AI feature response latency and concurrent-user scalability. Results indicate pass rates above 87% across all modules, typical AI response times of 3-17 seconds depending on feature complexity, and stable real-time note-broadcast latency under load, demonstrating that a single, prompt-engineered platform can reasonably reproduce the core stages of expert research reading within one coherent, collaborative interface.
In design, good ideas are the source of a designer’s creativity. With the development of artificial intelligence, large language models (LLMs) represented by ChatGPT shows excellent conversational and creative reasoning capabilities. However, we still know very little about the design ideation process and its performance in terms of productivity. In this paper, we propose a representation of the design ideation process based on NLP. On this basis, referring to LASSO and linkography method, we propose a DCLASSO model to calculate the interconnections between design moves in protocol analysis. In the subsequent experiment, we take the design protocols generated by ChatGPT as the experimental sample to verify the proposed model. Finally, with the generated linkographs, the characteristics of ChatGPT and human designers are analyzed and compared from aspects of design ideation process and design productivity.
The rapid advancements in large language models (LLMs) have demonstrated their potential to accelerate scientific discovery, particularly in automating the process of research ideation. LLM-based systems have shown promise in generating research ideas. However, current approaches predominantly rely on prompting-based pre-trained models, limiting their ability to optimize generated content effectively. More-over, they also lack the capability to deal with the complex interdependence and inherent restrictions among novelty, feasibility, and effectiveness, which remains challenging due to the inherent trade-offs among these dimensions, such as the innovation-feasibility conflict. To address these limitations, we propose a novel framework that employs a two-stage approach combining Supervised Fine-Tuning (SFT) and controllable Reinforcement Learning (RL). In the SFT stage, the model learns foundational patterns from pairs of research papers and follow-up ideas. In the RL stage, multi-dimensional reward modeling, guided by fine-grained feedback, evaluates and optimizes the generated ideas across key metrics. Dimensional controllers enable dynamic adjustment of generation, while a sentence-level decoder ensures context-aware emphasis during inference. Our framework provides a balanced approach to research ideation, achieving high-quality outcomes by dynamically navigating the trade-offs among novelty, feasibility, and effectiveness.
The integration of Generative Artificial Intelligence (GenAI) into higher education has outpaced the development of pedagogical models capable of directing it towards higher-order learning. Unstructured, zero-shot interaction with Large Language Models tends to yield fluent but uncritical output, while conventional assessment instruments are not designed to capture the quality of a learner's interaction with an AI system. This article addresses that gap by presenting the design of a Personalized Multi-Agent System (MAS) built upon the CIDI (Clarify, Ideate, Develop, Implement) pedagogical framework, together with the assessment architecture required to evaluate it. Following a design-based research methodology, four components are specified. First, the CIDI framework is articulated as a four-stage model of structured human-AI interaction, grounded in Distributed Cognition Theory and Cognitive Load Theory. Second, a dual-agent architecture is specified in full: an Ideation Agent configured for divergent generation and a Critical Evaluator Agent configured for convergent methodological critique, each defined by its system prompt and temperature configuration, and both integrated into an institutional Learning Management System. Third, a complete instructional design is presented, implementing the framework as a four-stage research-planning task within a research methodology curriculum, and introducing the Iterative Refinement Cycle (IRC) as an observable indicator of dialogic engagement recoverable from LMS logs. Fourth, two purpose-built assessment instruments are introduced: a 100-point Cognitive Competence Score rubric addressing problem formulation, methodological rigour, logical structuring and academic literacy; and an analytic rubric operationalizing Torrance's four dimensions of divergent thinking. The article concludes with a validation protocol specifying the design, reliability indices and analytic procedures required for subsequent empirical testing. The contribution is methodological: it supplies a reproducible architecture and a transparent instrumentation set for research on structured human-AI interaction in higher education.
Introduction: Students with mild intellectual disabilities often encounter significant difficulties when faced with complex academic tasks, particularly in developing research topics. These challenges are not only linked to their cognitive limitations but are also compounded by limited instructional support within learning environments. As a result, there is an increasing need for innovative strategies that can simplify learning without diluting academic standards. One promising approach is the use of artificial intelligence–based scaffolding tools, which provide learners with structured guidance to help them navigate demanding academic tasks. Purpose: This study aimed to explore how AI-supported planning tools influence students' ability to formulate research topics. Beyond this central objective, the study also examined whether differences in school type and gender affected students' performance when supported with AI scaffolding. Methods: The study adopted a pretest–posttest randomized controlled trial involving 94 students. Participants were randomly assigned to either the experimental group, which received AI-scaffolded support, or the control group, which did not receive any support. The Research Topic Quality Scale, adapted from Creswell and Clark (2018), was employed to evaluate the clarity, feasibility, and alignment of research topics generated by the students. Data analysis was conducted using ANCOVA to examine both main effects and interaction effects of the independent variables. Results: The findings revealed that students who had access to the AI-scaffolded tool performed significantly better in developing their research topics compared to those in the control group. In addition, results showed that private school students achieved higher scores than their counterparts in public schools, suggesting that resource availability played a role. Gender, however, did not emerge as a significant factor in students’ performance. No significant interaction effects were found between school type, gender, and intervention. Conclusion: The study demonstrates that AI-based scaffolding has the potential to enhance the ability of students with mild intellectual disabilities to engage with higher-order academic tasks, such as formulating research topics. While the intervention proved effective for both male and female learners, the differences observed between private and public schools highlight persistent inequalities in educational resources that still need to be addressed. Originality/Value: This research contributes to the growing body of evidence supporting the integration of artificial intelligence into inclusive education. By demonstrating that AI-driven scaffolding can enhance complex academic skills, the study underscores its value as a practical and innovative approach for promoting educational success among students with mild intellectual disabilities.
Postgraduate students are required to undertake research projects as part of the requirements of the courses they take. In that regard, they need to understand how theories are applied in research work. The aim of this study was to document challenges postgraduate students encounter when applying social research theories in constructing research questions. The key objective was to determine the deficits postgraduate students have in applying theories in the formulation of research questions. The second objective was to identify appropriate strategies as proposed from the respondents in overcoming the deficits identified under the first objective. Thestudy was conducted in Kenya and Armenia in selected public universities. A Quasi-qualitative survey research design was used and target population consisted of lecturers and postgraduate students conveniently selected from the social sciences departments. Focused group discussions (FGD) were used to collect data through the online platform Zoom where one FGD consisted of 9 conveniently selected postgraduate supervisors and two consisted of 11 and 14 postgraduate students respectively. Internet-based research was used to get information from online sources. Thematic and contextual analyses were used and the results showed that students have inadequate skills, knowledge, and competencies suitable for helping them formulate research questions using theoretical constructs. The study recommends improving the content and approaches of teaching research methods, focusing on the application of theory in research studies. The study further proposes strategies for ensuring that the students can design, develop, and use theory in all aspects of conducting research. The study has value by proposing ways of overcoming inadequacies by improving postgraduate students’ research competencies, by applying theoretical constructs in the development of research questions for their research studies.
No abstract available
Developing students' metacognitive awareness and self‐efficacy is crucial for fostering independent learning and higher‐order thinking skills, especially in research‐related tasks. Despite the widespread recognition of the importance of cultivating higher‐order thinking skills in higher education, there remains a significant gap in our understanding of effective instructional methods for fostering these skills among students. To address this gap, this study examines the impact of instructional practices on undergraduate students' metacognitive awareness and self‐efficacy in research question formulation—a fundamental skill in academic inquiry. Using a pretest/posttest design, 291 students from two colleges participated in courses where active and passive instructional methods served as the intervention. Our analysis reveals that active instructional methods were associated with marked improvements in self‐efficacy. The study provides empirical evidence of the effectiveness of active, student‐centred pedagogies in enhancing cognitive and self‐regulatory capacities. These findings offer educators practical guidance for designing instructional interventions that cultivate critical thinking and research skills, ultimately advancing student learning and success.
Following the launch of the generative AI Web application, Ask.SMILE, designed to evaluate the cognitive levels of questions asked, 2559 educators generated 25,973 question-feedback sets over a three-month period, with an average of over 10 questions per participant. Analyses revealed a significant improvement in question quality from initial submissions to later stages. Specifically, participants progressed from Level 1 questions (simple recall) to Level 5 questions (creative and evaluative), based on Bloom’s Taxonomy. However, participants who began with lower-quality questions showed varied progression paths to higher levels. To analyze the data, a Linear Mixed Model (LMM) was used to account for both fixed and random effects, revealing a statistically significant improvement in question quality over time (p < .01). ANOVA results further confirmed the significance of question stage on quality, and a Least Squares Means (LSMEANS) analysis indicated a precise and significant improvement in later-stage questions. Additionally, survey responses highlighted participants’ appreciation for the ongoing development of AI-driven educational tools like Ask.SMILE. Several implications and future research directions are explored, including the potential to expand the platform’s features beyond question evaluation.
Introduction Medical students may arrive at medical school with some research background but not necessarily evidence-based medicine (EBM) skills. First-year preclinical medical students require foundational skills for EBM (formulating background and foreground questions, navigating information sources, and conducting database searches) before critically appraising evidence and applying it to clinical scenarios. Methods We developed a flipped classroom EBM workshop for preclinical students combining prework modules and a 60-minute in-person session. After completing the online modules on foundational EBM skills, students participated in an in-person activity based on patient cases. In small groups, students formulated background and foreground questions based on a case and looked for evidence in resources assigned to each group. Small groups reported back to the whole group how they searched for information for their patient cases. A total of 105 first-year medical students were required to complete this workshop after concluding their basic sciences courses. Results Because current EBM assessment tools do not assess the early steps of EBM, we developed an assessment tool for foundational EBM tools. Before the modules, students completed a pretest on formulating questions and searching for information. After the workshop, students completed a posttest. Students showed improvement in differentiating background and foreground questions (p < .001), formulating answerable clinical questions (p < .001), and developing appropriate database searches (p < .001 and p = .002). Discussion This flipped classroom approach to teaching foundational EBM skills may be adapted for different contexts, but educators should consider time limitations, group size, and tools for interactivity.
The 21st-century learning has undergone a paradigm shift in education from teacher-centered to student-centered. It recently emphasises students' critical and creative thinking abilities, effective communication, innovation, problem-solving, and collaboration. This study examined the Ricosre model's effectiveness with the Question Formulation Technique (QFT) in enhancing students' higher-order thinking skills (HOTS) and science literacy. It involved 296 fifth-grade students from three selected state elementary schools, chosen through cluster random sampling. Each group consisted of 148 students divided into control and experimental groups. Data were collected through multiple-choice and essay instruments and subsequently analysed using multivariate analysis of variance (MANOVA). The research variables were students' higher-order thinking skills (HOTS) and science literacy. The results indicated that the Ricosre model with QFT effectively improves students' higher-order thinking skills (HOTS) and science literacy, as it encouraged active participation during the learning process, trained students to develop questioning skills, find answers, analyse, share ideas, and draw conclusions. Thus, it can foster curiosity and metacognitive abilities among students.
PURPOSE: Research Question: How much training do Dental Hygiene students require in order to use the question formulation rubric effectively: a brief 5-minute overview or a 25-minute training that includes a student peer assessment application exercise? METHODS: Randomized controlled trial. All pre-randomized 24 students took the question formulation pre-test on the first day of the course on January 22nd. As expected, there were no statistical differences between the Intervention or Control group pre-test scores. The instructors administered the post-test on February 19th after the Intervention group had received the training and rubric with the Control group only having received the rubric with a brief explanation. RESULTS: The investigators employed a paired t-test to analyze the pre- and post-test score differences for each student in the Intervention and Control groups. Surprisingly, the students' average post-test scores were 41.75 for the Control group and 43.67 for the Intervention group on a 70-point scale, which were not markedly different. Apparently, some contamination occurred. CONCLUSIONS: Additional analysis suggests that dental hygiene students benefit by experiencing extra instruction, with the cautionary caveats that this study was limited by some contamination and that it took place during the Covid-19 era.
No abstract available
Assessing student engagement and performance in online learning environments is essential for advancing adaptive tutoring systems. This study examines how question formulation influences student response behavior by analyzing response time, help feature usage, and accuracy. An initial online test with a formulation error led to longer response times and greater reliance on help features, indicating comprehension difficulties. A second test with a revised question format, involving 60 students, resulted in shorter response times, reduced help usage, and higher accuracy. These findings provide empirical evidence that unclear assessment items can increase cognitive load, hinder performance, and distort evaluation outcomes. Systematically validating question design can improve student engagement and learning effectiveness. Future research should explore AI-based tools to automatically detect and correct formulation issues in digital education environments.
Abstract Experienced researchers know that research begins with questions. The ACRL Frame for “Research as Inquiry” includes question formulation as one of the knowledge practices. Librarians and instructors in higher education recognize that many students have trouble isolating the main research question in scholarly studies, in generating analytical and critical questions as they read, and in formulating their own research questions as well. Furthermore, undergraduates frequently fail to make a strong connection between developing questions and seeking information for their research projects. What is the role of questioning in information literacy? How can these critical thinking skills best be taught? Is it the task of librarians to do so? The art and study of questioning are not new, but research within LIS has been limited. Thus, this work has several aims. We introduce the idea that question formulation is critical to information literacy, and should be explicitly taught and studied to improve instruction. We investigate existing theoretical frameworks about question formulation and highlight the few practical studies and resources. We articulate a QF-IL Research Model, identify research gaps, and propose a research agenda for academic librarians.
No abstract available
No abstract available
Ethnomusicology is widely recognized as an interdisciplinary field that examines music as a cultural practice embedded within social, historical, and political contexts. Despite increasing interest among postgraduate students, many emerging researchers encounter difficulties in understanding how ethnomusicological research is conducted in practice, particularly with regard to integrating musical analysis with ethnographic inquiry and navigating the ethical and methodological complexities of fieldwork. This article presents a concise methodological guide designed to support postgraduate students undertaking ethnomusicological research. Drawing upon scholarship in qualitative research, ethnography, and music studies, the paper outlines key stages in the research process, including topic formulation, research question development, methodological design, field engagement, documentation, data analysis, and scholarly dissemination. Particular attention is given to reflexivity, researcher positionality, and ethical engagement with communities, which are essential considerations in culturally embedded research. In addition, the paper highlights common methodological and practical challenges encountered by novice researchers and proposes strategies for addressing these challenges within field-based inquiry. By synthesizing theoretical insights with practical guidance, the article aims to provide postgraduate researchers with a structured framework for conducting ethnomusicological studies while contributing to broader discussions on interdisciplinary research methodologies in music scholarship.
The integration of Artificial Intelligence (AI) into creative and design processes has shifted from automation towards co-creation, positioning AI as a collaborative partner rather than a replacement. As AI-driven tools become more embedded in human-centred design, understanding their impact on interaction dynamics, ethics, and usability is critical. This review examines key advancements in human-AI co-design and co-creation fields, focusing on interaction frameworks, ethical considerations, non-linear collaboration models, domain-specific applications, and user experience (UX) design. Recent research emphasizes the need for structured frameworks that facilitate effective communication and partnership between humans and AI in creative tasks. Mixed-initiative and explainable AI (XAI) approaches play a crucial role in enhancing transparency and interpretability, allowing designers to co-create with greater trust and autonomy. Ethical concerns, such as AI’s influence on user perception and decision-making, are also gaining prominence, calling for responsible AI deployment in co-creative settings. Additionally, non-linear collaboration models redefine AI’s role as an adaptive assistant throughout iterative design stages, aligning with the dynamic nature of creative processes. Domain-specific applications, ranging from game and product design to choreography and smart manufacturing, illustrate the versatility of AI in augmenting human creativity. AI-assisted UX design further extends this impact by personalizing user experiences and streamlining workflows, ultimately improving efficiency and engagement. Despite these advancements, challenges remain in balancing AI autonomy with human control, evaluating its impact on creative workflows, and developing inclusive methodologies that cater to diverse design disciplines. This review synthesizes current research trends and identifies future directions for designing AI systems that empower, rather than replace, human expertise in creative industries.
Growing interest in eXplainable Artificial Intelligence (XAI) aims to make AI and machine learning more understandable to human users. However, most existing work focuses on new algorithms, and not on usability, practical interpretability and efficacy on real users. In this vision paper, we propose a new research area of eXplainable AI for Designers (XAID), specifically for game designers. By focusing on a specific user group, their needs and tasks, we propose a human-centered approach for facilitating game designers to co-create with AI/ML techniques through XAID. We illustrate our initial XAID framework through three use cases, which require an understanding both of the innate properties of the AI techniques and users’ needs, and we identify key open challenges.
No abstract available
Generative AI agents have been widely used in creative activities. Complex creations in the real world require multiple interactions with agents to approach the creative goal gradually. However, there are currently no well-defined frameworks to guide human-agent co-creation in complex creative tasks. In this study, we propose GenFlow, a human-agent co-creation framework, to enhance the generation quality and user experience by providing explainable and controllable mixed-initiative with different granularities at different phases of creation. We demonstrated a GenFlow-based picture book agent and conducted an early user test with 11 participants. The results proved the framework’s effectiveness and usefulness in guiding users’ collaborative and interactive behaviors. Our research emphasizes the importance of explainable and controllable interfaces in human-agent co-creation and proposes design strategies. These strategies contribute to designing AI agents that can effectively support the creation process.
No abstract available
The practice of sound design involves creating and manipulating environmental sounds for music, films, or games. Recently, an increasing number of studies have adopted generative AI to assist in sound design co-creation. Most of these studies focus on the needs of novices, and less on the pragmatic needs of sound design practitioners. In this paper, we aim to understand how generative AI models might support sound designers in their practice. We designed two interactive generative AI models as Creative Support Tools (CSTs) and invited nine professional sound design practitioners to apply the CSTs in their practice. We conducted semi-structured interviews and reflected on the challenges and opportunities of using generative AI in mixed-initiative interfaces for sound design. We provide insights into sound designers’ expectations of generative AI and highlight opportunities to situate generative AI-based tools within the design process. Finally, we discuss design considerations for human-AI interaction researchers working with audio.
No abstract available
No abstract available
Due to the remarkable content generation capabilities, large language models (LLMs) have demonstrated potential in supporting early-stage conceptual design. However, current interaction paradigms often struggle to effectively facilitate multi-round idea exploration and selection, leading to random outputs, unclear iterations, and cognitive overload. To address these challenges, we propose a human-AI co-ideation framework aimed at tracking the evolution of design ideas. This framework leverages a structured idea representation, an analogy-based reasoning mechanism and interactive visualization techniques. It guides both designers and AI to systematically explore design spaces. We also develop a prototype system, IdeationWeb, which integrates an intuitive, mind map-like visual interface and interactive methods to support co-ideation. Our user study validates the framework’s feasibility, demonstrating enhanced collaboration and creativity between humans and AI. Furthermore, we identified collaborative design patterns from user behaviors, providing valuable insights for future human-AI interaction design.
As generative artificial intelligence (GAI) becomes increasingly integrated into the design domain, research has begun to explore how it can be meaningfully incorporated into traditional design practices, fostering the development of more collaborative design processes. This study proposes a Human–AI Co-Creative Design Process (HAI-CDP) model and evaluates its impact on designers’ creativity through a comparative experimental design. The results indicate that the HAI-CDP substantially improves creative performance over the traditional design process. For novice designers, its primary value lies in facilitating idea generation, whereas for experienced designers, it contributes more to elevating the quality and refinement of creative outcomes. Although the Human–AI Co-Creative Design Process lowers the entry barrier to creative engagement, the findings also reaffirm that design experience remains a critical factor shaping creative output.
With the rapid advancement of generative AI technologies, the collaboration between designers and generative AI during the conceptual design stage has fostered a novel paradigm of Human-AI Co-creation in design. However, relevant systematic reviews remain relatively scarce. This study analyses 80 research papers and publications from prominent databases such as Web of Science (WOS), Scopus, ScienceDirect, and Google Scholar, based on rigorous selection criteria and utilizing the CiteSpace bibliometric analysis tool. We developed a comprehensive systemic model for Human-AI Co-creation Design: cognition, process, methodology, and outcome. Key findings encompass the transformation of design cognition from designer-centric to Human-AI collaborative cognition, the shift of design process from experience-driven to constraint-driven, the evolution of design method from unimodal to multimodal interaction, and the indirect influence of generative AI on design outcome. Furthermore, this paper discusses the boundaries of Human-AI Co-creation Design with generative AI and envisions future research directions for multidisciplinary collaboration and the future models of AI and designer creative fusion. Through a systematic analysis and summary of the current paradigm shift, this paper identifies the limitations of existing research and provides insights into future research directions for building a more efficient Human-AI Co-creation design paradigm.
Abstract This study employs a mixed-methods design - two 2 × 2 between- subjects experiments and a PLS-SEM analysis (N = 639) to examine the dual pathways driving user self-disclosure in Human-AI co-creation. Results indicate that Self-congruence and AI attachment not only directly promote disclosure intention but also exert an indirect mediating effect by enhancing perceived value and reducing perceived privacy risks. The study further found that self-congruence has an inverted U-shaped effect on perceived value, and that privacy self-efficacy positively moderates the relationship between self-congruence/AI attachment and perceived value, while negatively moderating the relationship between self-congruence/AI attachment and perceived privacy risk. This study integrates Privacy Calculus, Self-congruence, and Atachment Theory in Human-AI co-creation scenarios, revealing the “over-personalization paradox” and deepening our understanding of privacy tradeoffs in Human-AI collaboration. It also provides empirical evidence and practical implications for personalized adaptation, the establishment of user trust, and privacy feature development in AI collaborative design.
Generative AI (GenAI) is rapidly transforming creative fields, raising critical questions about its impact on artistic identity and authorship, which also influence art education. While studies have explored GenAI adoption, the diversity of art students’ subjective perspectives on collaborating with these tools remains under-examined. This study uses Q methodology, a mixed-methods approach to investigating subjectivity, to explore the distinct perspectives art students hold regarding their creative identity when collaborating with GenAI. We conducted a Q study with 40 university students from various art-related disciplines (e.g., visual arts, design, music, animation) across different academic levels. Participants sorted 42 statements reflecting diverse viewpoints on GenAI collaboration, authorship, ethics, and creative control. By-person factor analysis revealed five distinct perspectives: (1) Human-centered directors, who treat GenAI as a tool under strong human intention; (2) Authorship guardians, who resist AI co-authorship and prioritise authenticity and ethical boundaries; (3) Conflicted co-creators, who use GenAI but struggle with ownership, disclosure, and earned pride; (4) Pragmatic integrators, who adopt GenAI for workflow efficiency while keeping craft expertise central; and (5) Enthusiastic explorers, who embrace AI’s creative possibilities, open to co-agency and novel expression. These findings offer a nuanced typology of student viewpoints, moving beyond simple acceptance or rejection. We further propose a two-dimensional framework that maps these perspectives according to students’ openness to GenAI and their sharing of creative agency with it. This research contributes empirical insights into how art students negotiate creative identity with GenAI, provides a valuable framework for educators to develop tailored pedagogical strategies addressing diverse student needs and concerns, and offers implications for designing more human-centered GenAI tools for creative practice.
This paper explores Imagined Realities, an experimental project that investigates the dynamic interplay between human creativity and artificial intelligence in the realm of storytelling. Conducted at Jawaharlal Nehru Architecture and Fine Arts University, Hyderabad, the initiative brought together interior design students and OpenAI's ChatGPT-4 to collaboratively generate narratives based on a curated set of AI-generated images and character briefs. The dual-phase study examined both human-written and AI-generated stories through a detailed rubric focusing on narrative structure, spatial and sensory language, emotional resonance, and character analysis.By integrating pedagogical goals from the students’ communication skills curriculum with cutting-edge generative AI technology, the project aimed to deepen self-expression, storytelling proficiency, and critical engagement with technology. The comparative analysis revealed key differences in emotional nuance, spatial perception, and cultural depth between human and machine-authored texts. The resulting book, Imagined Realities, serves not only as a creative anthology but as a reflective commentary on the evolving landscape of human-AI collaboration in the arts. Ultimately, this project underscores the complementary strengths of human intuition and AI logic while prompting vital discourse on authorship, creativity, and the future of storytelling.
No abstract available
No abstract available
While Generative AI (AIGC) reshapes creative production, there is a growing need to understand user experiences beyond functional utility. Drawing on meaning-making theory and self-determination theory, this study explores the relationships between human–AI interaction factors and perceived meaning—the extent to which users find personal relevance and expressive alignment in generated content. Survey data were collected from 299 AIGC users and analyzed using a mixed-method approach combining partial least squares structural equation modeling (PLS-SEM) and fuzzy-set qualitative comparative analysis (fsQCA). PLS-SEM results indicate that system interactivity is positively associated with perceived meaning not only directly but also indirectly through a sequential pathway involving perceived control and psychological ownership. Similarly, post-editing behavior is linked to meaning primarily through these psychological factors. Furthermore, fsQCA identifies diverse configurational paths; while task openness shows no significant linear relationship, it emerges as a critical condition in specific combinations associated with high perceived meaning. Theoretically, these findings highlight the role of the control-ownership pattern in virtual co-creation. Practically, they suggest that AIGC design should prioritize controllable co-creation, supporting user agency to foster stronger content ownership and meaningful experiences.
Generating clarifying questions can effectively clarify users' complicated search intent in conversational search systems. However, existing methods based on pre-defined templates are inadequate in understanding explicit user intents, making generated questions monotonous or inaccurate in some cases. In this paper, we define the ''intent'' of a query as a verb representing the potential behavior, action, or task the user may take. We study generating clarifying questions from a new perspective by incorporating the intents explicitly to form ''intent-aware'' questions with high informativeness and accuracy. Since obtaining gold intent-aware questions is expensive, we propose a rule-based method and a continual learning model to generate intent-aware questions as weak supervision signals. The former leverages search results to mine contextual intent-aware words or phrases, and the latter relies on parallel corpora to paraphrase template-based questions by incorporating the intents. The generated weak supervision data are then applied to fine-tune a BART-based model for end-to-end intent-aware question generation. We also explore to prompt a large language model to generate intent-aware questions. Experimental results on a public clarification dataset demonstrate that our proposed methods improve users' search experience compared to existing methods.
In legal case retrieval, existing work has shown that human-mediated conversational search can improve users’ search experience. One of the key problems for a practical conversational search system is how to ask high-quality clarifying questions to initiate conversations with users and understand their search intents. Previous works demonstrated that human-annotated external domain knowledge (such as event schemas) can improve the legal utility of clarifying questions generated by large language models. However, these methods are restricted to specific law systems or languages and cannot be generalized to others. To this end, we propose to generate context and domain-specific questions with LLMs without external annotations or knowledge by extracting information from top-retrieved documents given the current conversation context. Specifically, we construct a conversational legal case retrieval system CARQ that iteratively selects neighbor candidate case documents from the retrieved list at each conversation step to ask clarifying questions. We pretrain CARQ to capture the differences between legal cases and employ the reward augmented maximum likelihood to optimize the system directly for retrieval metrics. Extensive automated and human evaluations on three widely adopted legal case retrieval datasets demonstrate the superior effectiveness of our approach as compared with the state-of-the-art baselines.
Search queries are often short, and the underlying user intent may be ambiguous. This makes it challenging for search engines to predict possible intents, only one of which may pertain to the current user. To address this issue, search engines often diversify the result list and present documents relevant to multiple intents of the query. An alternative approach is to ask the user a question to clarify her information need. Asking clarifying questions is particularly important for scenarios with “limited bandwidth” interfaces, such as speech-only and small-screen devices. In addition, our user studies and large-scale online experiments show that asking clarifying questions is also useful in web search. Although some recent studies have pointed out the importance of asking clarifying questions, generating them for open-domain search tasks remains unstudied and is the focus of this paper. Lack of training data even within major search engines for this task makes it challenging. To mitigate this issue, we first identify a taxonomy of clarification for open-domain search queries by analyzing large-scale query reformulation data sampled from Bing search logs. This taxonomy leads us to a set of question templates and a simple yet effective slot filling algorithm. We further use this model as a source of weak supervision to automatically generate clarifying questions for training. Furthermore, we propose supervised and reinforcement learning models for generating clarifying questions learned from weak supervision data. We also investigate methods for generating candidate answers for each clarifying question, so users can select from a set of pre-defined answers. Human evaluation of the clarifying questions and candidate answers for hundreds of search queries demonstrates the effectiveness of the proposed solutions.
The use of clarifying questions (CQs) is a fairly new and useful technique to aid systems in recognizing the intent, context, and preferences behind user queries. Yet, understanding the extent of the effect of CQs on user behavior and the ability to identify relevant information remains relatively unexplored. In this work, we conduct a large user study to understand the interaction of users with CQs in various quality categories, and the effect of CQ quality on user search performance in terms of finding relevant information, search behavior, and user satisfaction. Analysis of implicit interaction data and explicit user feedback demonstrates that high-quality CQs improve user performance and satisfaction. By contrast, low- and mid-quality CQs are harmful, and thus allowing the users to complete their tasks without CQ support may be preferred in this case. We also observe that user engagement, and therefore the need for CQ support, is affected by several factors, such as search result quality or perceived task difficulty. The findings of this study can help researchers and system designers realize why, when, and how users interact with CQs, leading to a better understanding and design of search clarification systems.
Asking clarifying questions is an interactive way to effectively clarify user intent. When a user submits a query, the search engine will return a clarifying question with several clickable items of sub-intents for clarification. According to the existing definition, the key to asking high-quality questions is to generate good descriptions for submitted queries and provided items. However, existing methods mainly based on static knowledge bases are difficult to find descriptions for many queries because of the lack of entities within these queries and their corresponding items. For such a query, it is unable to generate an informative question. To alleviate this problem, we propose leveraging top search results of the query to help generate better descriptions because we deem that the top retrieved documents contain rich and relevant contexts of the query. Specifically, we first design a rule-based algorithm to extract description candidates from search results and rank them by various human-designed features. Then, we apply an learning-to-rank model and another generative model for generalization and further improve the quality of clarifying questions. Experimental results show that our proposed methods can generate more readable and informative questions compared with existing methods. The results prove that search results can be utilized to improve users' search experience for search clarification in conversational search systems.
Legal case retrieval is a special IR task aiming to retrieve supporting cases for a given query case. Existing works have shown that conversational search paradigm can improve users' search experience in legal case retrieval. One of the keys to a practical conversational search system is how to ask high-quality clarifying questions to initiate conversations with users and understand their search intents. Recently, Large Language Models, such as ChatGPT and GPT-4, have shown superior ability in both open-domain QA and conversations with human. Thus it is natural to believe that they could be applied to legal conversational search as well. However, our preliminary study has shown that generating clarifying questions in legal conversational search with SOTA LLMs (e.g., GPT-4) often suffers from several problems such as duplication and low-utility contents. To address these problems, we propose LeClari, which leverages legal event schema as external knowledge to instruct LLMs to generate effective clarifying questions for legal conversational search. LeClari is constructed with a prompt module and a novel legal event selection module. The former defines a prompt with legal events for clarifying question generation and the latter selects potential event types by modeling the relationships of legal event types, conversational context, and candidate cases. We also propose ranking-oriented rewards and employ the reward augmented maximum likelihood (RAML) method to optimize LeClari directly based on the final retrieval performance of the conversational legal search system. Empirical results over two widely adopted legal case retrieval datasets demonstrate the effectiveness of our approach as compared with the state-of-the-art baselines.
Clarifying questions enable Conversational Search (CS) systems to resolve underspecified queries by eliciting missing information from users. However, how prompting strategies shape the quality of clarifying questions and how such questions should be evaluated at scale remains understudied. We present Automatic GENeration and evaluaTion of Clarifying Questions (AGENT-CQ), a framework for systematically generating and evaluating clarifying questions and simulated user responses using Large Language Models (LLMs). To support scalable and multi-perspective evaluation, we introduce CrowdLLM, an LLM-based evaluation paradigm that simulates diverse annotator judgments through distinct evaluator personas. Our experiments span both open-domain CS and a regulatory question-answering setting, allowing us to examine the extent to which clarification strategies generalize across domains with different interaction constraints. Across settings, temperature-variation prompting leads to higher quality clarifying questions than baseline prompting and human-authored questions on several dimensions of the task. In addition, LLM-generated clarifying questions lead to improved downstream retrieval performance than human-authored questions in open-domain search. Together, AGENT-CQ and CrowdLLM provide a practical framework for studying and improving clarification strategies in conversational IR systems.
Question answering (QA) with disambiguation questions is essential for practical QA systems because user questions often do not contain information enough to find their answers. We call this task clarifying question answering, a task to find answers to ambiguous user questions by disambiguating their intents through interactions. There are two major problems in building a clarifying question answering system: data preparation of possible ambiguous questions and the generation of clarifying questions. In this paper, we tackle these problems by sentence generation methods using sentence structures. Ambiguous questions are generated by eliminating a part of a sentence considering the sentence structure. Clarifying the question generation method based on case frame dictionary and sentence structure is also proposed. Our experimental results verify that our pseudo ambiguous question generation successfully adds ambiguity to questions. Moreover, the proposed clarifying question generation recovers the performance drop by asking the user for missing information.
Recent advances in Large Language Model (LLM) “agent” systems have moved language models beyond single-turn generation toward goal-directed workflows that can plan, ask clarifying questions, and iteratively refine outputs. In parallel, retrieval-augmented generation (RAG) has become a practical way to ground these agents in enterprise knowledge—enabling models to leverage internal documentation and policies without expensive and recurring retraining. However, RAG is only as strong as the underlying corpus: when key information is missing, outdated, or fragmented, retrieval cannot fill the gaps. In many organizations, assembling high-quality, up-to-date documents remains a persistent bottleneck, limiting both the reliability of downstream applications and the speed at which knowledge can be operationalized. To address this, we introduce the Assistant-Scribe-Knowledge Checker (ASK) framework for generating structured specification documents through guided interviews. ASK decomposes the end-to-end workflow into three specialized LLM agents: an Assistant that asks questions to the user and steers the interview through adaptive follow-ups; a Scribe that continuously summarizes what has been said and records it into a schema-constrained specification document with explicit content requirements; and a Knowledge Checker that evaluates the evolving document against those specifications, detects gaps or weakly supported entries, and advises the Assistant on the next best questions to ask to reach the desired completeness and quality. This separation supports seamless document creation while improving quality control over content, structure, and coverage. By distributing responsibilities across three lightweight, role-specific agents, ASK reduces reliance on a single large model and enables modular scaling across teams and domains. We evaluated the ASK framework in a consulting-firm setting where consultants are required to produce project “return-of-experience” documents (successes, failures, challenges, and solutions) to capture reusable knowledge. Compared to documents authored manually, ASK-guided interviews consistently produced more complete, better-structured, and higher-quality specifications with less variability across authors. Beyond measurable quality gains, consultants also reported a clear preference for the interview-based workflow, citing lower effort and a more natural way to articulate tacit project knowledge.
Introduction: Research is a purpose-directed activity. Although aims, goals, objectives, and targets are interrelated statements of purpose, they differ in hierarchy, scope, measurability, timeframe, and significance. This brief review clarifies these terms and describes the characteristics of a good research objective. Methods: Relevant papers were analyzed to delineate the definitions, interrelations, and distinctions among aims, goals, objectives, and targets in research projects. Results: The apparent purpose of a study (objectives and goals) addresses specific research questions, whereas its deeper purpose (aim) represents the underlying reason for conducting the study. Aims express ideals and long-term direction; goals outline the primary intent and problem to be addressed; objectives link intention to action by specifying how the research question will be answered; and targets indicate desired values of outcome indicators. A strong objective should be SMART (specific, measurable, achievable, relevant, and time-bound). Conclusion: Clear articulation of research purpose facilitates translation of findings into practice and enhances the overall quality of research.
No abstract available
This research develops theoretical and methodological foundations for modular, mixed-initiative conversational recommender systems (CRS). It investigates how such systems can be conceptualized and systematically evaluated. The aim is to improve preference elicitation, dialogue management and intent detection by investigating the role of large language models (LLMs) in these tasks. First, a literature review is provided to conceptualize the use of generative models in CRS, and behavioral archetypes in user sessions that inform system design are identified. A user study compares a traditional faceted search interface with system-driven CRS to assess trust, usability and preference extraction quality in order to assess system acceptance. The outlined work extends this prototype to support mixed-initiative dialogues, with components for intent detection, question generation and state-tracking and focuses on a modular evaluation, often missing in the current literature. The project aims to deliver a holistic CRS designed and evaluated with consultancy of end users and platform stakeholders.
Shopkeeper robots face a critical trade-off when recommending products to first-time customers. Robots must balance recommendation efficiency with unconstrained natural language interaction. Sparse modeling techniques like Lasso regression offer efficient feature selection from limited samples. However, they typically demand numerical evaluations from customers, degrading the social experience. We define this dilemma as the Qualitative-Quantitative Divergence Problem and systematically compare the impact of numeric and non-numeric (natural language) recommendation dialogues on the HRI experience. As a verification platform, we implemented Impression-SmartClerk. This system extracts recommendation parameters from natural utterances via LLMs, rather than forcing customers to adapt to mathematical models, and ensures robustness through intent-adaptive conversation repair. Results from a physical robot experiment with 20 participants demonstrated that the non-numeric approach significantly improves the freedom of self-expression and the robot's likeability, while maintaining recommendation accuracy equivalent to numerical input methods without increasing cognitive load. Ultimately, this article demonstrates the resolution of the Qualitative-Quantitative Divergence Problem, and through an empirical comparison, provides valuable interaction-level insights into the UX trade-offs among expressiveness, robustness, and recommendation quality. GRAPHICAL ABSTRACTTwo comic style illustrations compare a shopper rating shoes with numbers to a shopper giving free form comments to a friendly robot assistant.The figure shows two comic style illustrations placed side by side that contrast two ways a shopper communicates with a robot assistant about shoes. On the left illustration, a human character appears three times in a vertical sequence, looking uncertain with a hand to the face or crossed arms. To the right of each appearance, a rectangular speech balloon from a boxy robot with antennae shows questions about rating shoes. The first robot message reads, How much do you like this? Please give it a score. Next to the text, a high top sneaker is shown. A reply balloon from the human reads, Umm... 3 points. Farther down, the robot asks, Got it. So, whats the score for this one? with a different sneaker beside the robot, and the human replies, Hmm... 8 points. At the bottom left, a cloud shaped thought balloon from the human contains the sentence, Answering with a number is difficult! It makes it harder to interact. On the right illustration, the same human and robot appear in a mirrored layout. The robot first asks, How much do you like this? beside the first sneaker. The human replies, The shape is nice, but the color is not really for me. The robot then asks, Got it, then how about this one? with the second sneaker nearby. The human responds, I like this one! At the bottom right, a cloud shaped thought balloon from the human reads, This is easy to interact with! I can talk to it just like a human shopkeeper. A large arrow between the two illustrations suggests a transition from the numeric rating interaction to the conversational description interaction.
Co-creation engages diverse stakeholders, including marginalized populations, in collaborative problem-solving to enhance engagement and develop contextually appropriate solutions. It is increasingly recognized as a way to democratize research and improve the impact of interventions, services, and policies. However, the lack of synthesized evidence on co-creation methods limits methodological rigor and the establishment of best practices. This review aimed to identify co-creation methods in academic literature and analyze their characteristics, target groups, and associated benefits and challenges. This scoping review follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews. The search was conducted in the Health CASCADE database v1.5 (including CINAHL, PubMed, and 17 additional databases via ProQuest) from January 1970 to March 2022. Data was aggregated and summarized, with qualitative data analyzed using Braun and Clarke’s six-phase thematic analysis approach. The review included 266 articles, identifying 248 distinct co-creation methods published between 1998 and 2022. Most methods were rooted in participatory paradigms (147 methods), with 49 methods derived from co-approaches like co-creation, co-design, and co-production, and 11 from community-based health promotion and action research. Methods were applied across 40 target populations, including children, adults, and marginalized groups. Many methods (62.3%) were delivered face-to-face, with 40 articles incorporating digital tools. Thematic analysis revealed nine benefits, such as enhanced creativity, empowerment, and improved communication, and six challenges, including resource constraints and systemic and structural barriers. This review emphasizes the importance of robust documentation and analysis of co-creation methods to inform their application in public health. Findings support the development of collaborative co-creation processes that are responsive to the needs of diverse populations, thereby enhancing the overall effectiveness and cultural sensitivity of the outcomes. This review highlights the potential of co-creation methods to promote equity and inclusion while emphasizing the importance of evaluating and selecting methods tailored to specific objectives, offering a critical resource for planning, conducting, and evaluating co-creation projects.
Strategic management of cognitive load enhances user interface (UI) design in a two-fold manner, particularly in information services that require frequent use of sophisticated data. This paper addresses cognitive overload and elevated user experience by decomposing UI elements and information flow. The primary concern is having users cope with vital information within a reasonable timeframe, which is critical to productivity and overall satisfaction. We explore cognitive load theory, specifically its application in UI design regarding intrinsic, extraneous, and germane cognitive load. Through case studies and usability testing, we focus on visual hierarchies, navigation, and overall design consistency that advance feedback and content chunking to show the extreme reduction in mental effort. Also, cognitive load solutions that alter content and interaction based on user activity are advanced as novel adaptive interfaces. The paper frames real-world scenarios to illustrate assessing cognitive load in information service systems to aid UI decision-makers and developers. Emphasis on enabling real-time interface navigation supports optimal information retention and swift decision making. A systems approach focused on users' cognitive capabilities enables a UI design which ultimately makes information services more effective, efficient, and user-friendly.
With the increasing complexity of human-computer interaction systems, the cognitive load caused by interface information overload has become a key bottleneck affecting user experience and operational efficiency. Therefore, this paper proposes an interactive interface adaptive design model based on dynamic cognitive load evaluation and constructs a closed-loop optimization framework of “perception-evaluation-decision-execution”. Therefore, this paper proposes an interactive interface adaptive design model based on the dynamic evaluation of cognitive load, and constructs a closed-loop optimization framework of “perception evaluation decision execution”. First, a dynamic multimodal cognitive load assessment model is designed, which integrates behavioral, eye-movement, and physiological signals via a cross-modal attention mechanism, combined with time series modeling and uncertainty estimation, to achieve continuous and accurate perception of cognitive load. The root mean square error of prediction is 0.592. Second, a cognitive-driven interface adaptive decision-making framework is constructed, which uses the results of cognitive load assessment and task context as the basis for decision-making. On this basis, a cognitive-constrained reinforcement learning optimization algorithm is proposed. By introducing an upper-limit constraint on cognitive load and a strategy clipping mechanism, interaction efficiency is guaranteed and decision stability is improved. Experimental results show that the proposed method reduces task completion time by 22.7%, increases user satisfaction by 41.9%, decreases cognitive load by 25.0%, and keeps total response delay within 81 milliseconds. This study provides a systematic solution for the construction of intelligent adaptive interface with both evaluation accuracy and decision stability.
: The possibilities offered by digital tools for visualization, distribution, and storage of educational resources, as well as for interaction and communication between professors and students, are broadly acknowledged. However, the role of graphical user interface design in digital educational environments has still not been sufficiently valued. In this paper, we support the idea that the arrangement, distribution, and visual attributes of graphic elements (i.e., what in the field of art is called composition) in the design of the graphic interface of educational environments is not only a matter of aesthetics. It also results in better understanding of the educational message and in a significant reduction of cognitive overload in the learning process. The relationship between visual stimuli and cognitive processes that favour learning is highlighted.
Interactive cognitive games are essential for preschoolers’ executive function, yet existing digital tools often rely on static difficulty progressions. This rigidity fails to accommodate volatile cognitive boundaries, causing either severe overload or boredom. We propose a dual-track intervention integrating a Reinforcement Learning-based Dynamic Difficulty Adjustment (RL-DDA) algorithm with a Pedagogical Conversational Agent (PCA). The RL-DDA optimizes cognitive load via Markov Decision Process (MDP) modeling, while the PCA provides affective scaffolding through automated navigation and a Wizard of Oz (WoZ) parental voice interface. A within-subjects study (N = 24) compared our system against Static-Easy and Static-Hard baselines. The dual-track system effectively mitigated frustration, reducing maximum consecutive errors by over 50%, and extended sustained attention spans by 61%. Qualitative analyses revealed emergent Human-Computer Interaction (HCI) risks, notably "distraction crises"—where children deliberately failed tasks to trigger PCA animations—and potential social dependency on algorithmic encouragement. We propose concrete design guidelines, emphasizing intervention granularity and progressive weaning, for future affective educational technologies.
Industrial plant management environments are becoming more complicated, with managers having to integrate production operations, maintenance activities, safety compliance, manpower allocation, and operational documentation across many digital platforms. The extensive use of fragmented tools including spreadsheets, emails, dashboards, and calendars frequently results in information overload, an increased cognitive effort, and reactive decision making. To solve these issues, this article introduces AI TaskManager, a human-centered, AI-assisted task and workflow management system intended to aid plant managers in manufacturing and industrial settings. AI TaskManager was built with Google AI Studio, which allows for the rapid creation of a high-fidelity prototype without the need for traditional software development. The system supports important management workflows with natural-language interface and multimodal AI capabilities, such as AI-assisted task creation, automated budget estimation from structured data, technical drawing interpretation, and adaptive job prioritizing. These functions are combined into a single interface to improve situational awareness, minimize cognitive load, and maintain human decision authority. A usability and human performance evaluation was conducted with ten participants, including plant supervisors, engineering professionals, and graduate researchers with experience in operational task management. With an average job execution time of 2.4 minutes, a 93% task completion rate, and a 6% mistake rate, the findings show excellent system performance. The findings demonstrate that AI TaskManager effectively facilitates cognitive ergonomics, decision support, and human–AI collaboration in industrial plant management, underscoring the promise of human-centered AI systems to improve managerial performance and operational resilience.
No abstract available
No abstract available
Modern educational platforms show a tendency toward information overload and cognitive biases resulting from suboptimal interface design. Effects of “false knowledge structuring” and information illusions are becoming common problems reducing the effectiveness of material comprehension. Particular concern arises from the unintentional influence of visual noise and information redundancy on the formation of distorted mental models in learners. This article presents a conceptual model of educational design promoting neuroergonomics and cognitive offloading. The proposed approach focuses on highlighting significant and primary information while minimizing distracting elements. The model includes principles of cognitively adaptive design considering neurophysiological features of perception and information processing. The proposed cognitive-centered ecosystem of educational interfaces is based on a multi-level system of design patterns aimed at minimizing cognitive load and increasing learning efficiency. This conceptual model provides a foundation for developing new standards for designing educational platforms with priority given to cognitive ergonomics and natural knowledge acquisition processes.
E-learning platforms play a crucial role in the modern education system with higher flexibility and accessibility for the users. However, they often adopt uniform interface designs and neglect learners’ cognitive differences, leading to cognitive overload and disengagement. While content personalization is common, dynamic interface-level adaptation remains underexplored. To address this gap, this study introduces an AI-driven adaptive user interface model to personalize the interface dynamically based on individual cognitive attributes, namely attention span, memory capacity, and cognitive load, with the consideration of layout modification, navigation structure, and information density. Three validated methods are used for cognitive profiling, namely attention and focus level via WebGazer, cognitive load through the N-Back test, and memory capacity via the Digit Span Test. A within-subjects experiment was conducted by using 30 undergraduates in Sri Lanka. All the participants interacted with both static and AI-driven adaptive interfaces, along with a post-interaction evaluation based on validated instruments combining NASA-TLX, SUS, and UEQ scales. Results indicated a higher adaptation success rate, along with positive post-feedback evaluations across cognitive load, navigation efficiency, personalization, usability, and engagement. Correlation patterns indicated that cognitive attributes influenced perceived outcomes. The impact of the AI-driven adaptive user interface is evaluated using quantitative analysis along with statistical data analysis using SPSS. The proposed system is designed as a web-based platform, ensuring AI-driven personalization to enhance user engagement and learning effectiveness. Further, the research findings contribute to the field of Human-computer interaction and the domain of education by validating AI-driven adaptive UI generation.
In the modern software development lifecycle, effi-cient project management is critical for success. Traditional tools often suffer from information overload, complex navigation, and a lack of intelligent insights, leading to decreased productivity. This paper introduces FlowTrack, a next-generation project management platform designed to address these challenges. FlowTrack leverages Artificial Intelligence (AI) to provide ac-tionable insights and automate routine tasks, while offering an immersive, aesthetically pleasing user interface built with modern web technologies. The platform aims to reduce cognitive load through an intuitive visual workspace and introduce AI-driven automation that significantly minimizes manual overhead. This paper details the system architecture, key features, and technical implementation strategy of FlowTrack, demonstrating how it bridges the gap between powerful functionality and user-centric design while maintaining scalability and performance. The study also highlights the importance of real-time collaboration, pre-dictive analytics, and workflow automation in modern project environments.
No abstract available
AI-powered learning technologies are increasingly being used to automate and scaffold learning activities (e
The increasing integration of artificial intelligence (AI) into English as a Foreign Language (EFL) education has created new opportunities for personalized learning, adaptive instruction, and learner autonomy. However, concerns remain regarding students' ability to maintain independent decision-making, critical engagement, and ownership of their learning processes when interacting with AI technologies. This study aims to systematically review the development of student agency in AI-assisted English language learning, focusing on agentic practices, critical AI literacy, AI-supported scaffolding, and psychological and pedagogical influences. Employing a qualitative systematic literature review (SLR) approach, six publications published between 2025 and 2026 were examined using qualitative content analysis and cross-study thematic synthesis. The findings reveal four interconnected themes. First, student agency develops dynamically through receptive, participatory, and co-creative practices, reflecting increasing independence and control over AI-mediated learning activities. Second, critical AI literacy strengthens students' ability to evaluate AI-generated information, recognize technological limitations, and make informed learning decisions. Third, AI-supported scaffolding facilitates English language development, particularly listening and writing, through personalized materials, adaptive feedback, and interactive learning opportunities. Fourth, psychological and pedagogical factors, including self-efficacy, motivation, positive emotions, metacognitive awareness, and autonomy-supportive teaching, significantly influence students' agentic engagement. Nevertheless, excessive reliance on AI, inaccurate feedback, and diminished authorial ownership remain important challenges. This review concludes that effective AI-assisted EFL instruction requires a balanced integration of technological assistance, critical AI literacy, independent decision-making, and meaningful teacher guidance. The findings highlight the importance of positioning AI as a supportive learning partner rather than a replacement for students' intellectual responsibility and autonomy.
Computing students experience challenges with negative perceptions toward writing and communicating technical ideas. While generative artificial intelligence (GenAI) writing tools provide immediate support, they often generate complete written content, which may reduce students' active participation and sense of ownership. ProseSpark is introduced as a consent-based agentic AI writing companion designed to support computing students without writing on their behalf. ProseSpark is grounded in human-centered AI and self-regulated learning theory to maintain student autonomy and encourage active engagement throughout the writing process. This poster presents ProseSpark's design principles, platform, and responsible-AI safeguards. Future evaluation will examine how workflow-oriented AI support influences computing students' motivation, engagement, and writing self-efficacy.
Generative artificial intelligence has reshaped writing practices by blurring the boundaries between human authorship and machine assistance. This study conducts a systematic literature review of AI authorship in AI-assisted writings, focusing on authorial agency, originality, and ownership. Using the SLR framework proposed by Kitchenham and Charters. After the process elimination of related primary studies, this review analyzed 25 primary studies published between 2024 and 2026 from SpringerLink and ACM Digital. The findings reveal five dominant research trends: epistemology and ethics of AI authorship, AI perception and literacy, linguistic and stylistic analysis, writing process and education, and systematic review. The primary studies highlight that authorial agency is increasingly understood as distributed across human writers, AI systems, developers, and datasets. Originality is also shifting from solitary creation to human-led curation, prompting, synthesis, and critical evaluation. Meanwhile, ownership remains contested, ranging from hybrid authorship to human-led responsibility and uncertainty over intellectual accountability. This study highlights the need for transparent AI use, stronger AI literacy, and ethical frameworks to preserve human responsibility in AI-mediated writing.
As generative AI (GenAI) becomes increasingly embedded in higher education, students face contradictory narratives about its use as it is often presented either as a risk to academic integrity or as a valuable educational resource (Luo, 2025). Navigating these tensions can be particularly complex for students from equity groups, who already contend with challenges in adapting to university learning environments (Stokes, 2024). While institutional policies attempt to regulate GenAI use, these often lack the nuance needed to guide context-sensitive decision-making (Corbin et al., 2025). This presentation outlines an action research project that investigated how dialogic teaching approaches can support students in developing critical GenAI literacies through collaborative exploration and reflection. The research took place within an Australian university’s Enabling Education program and was grounded in the principles of Dialogic Pedagogy which is a critical framework rooted in reciprocal communication, shared meaning-making, and the co-construction of knowledge (Shor & Freire, 1987). Rather than imposing rules, this pedagogical model encourages transparency, inquiry, and inclusive negotiation of acceptable practices. Encouraging students, who are new to university, to become active agents supports them to build their critical AI literacy (Toncelli, et al., 2025; Wang & Wang, 2025). This approach shifts the emphasis from compliance to empowerment, allowing students and educators to engage in meaningful dialogue about the ethical and effective use of GenAI in their learning. Using a critical participatory action research methodology (Kemmis et al., 2014), the study collected data across three foundational courses: language and literacy, digital literacy, and information literacy. In the language and literacy unit, students explored the use of GenAI tools in developing their academic writing. Through classroom activities and reflective discussions, students examined GenAI-generated feedback on their work, considering how such input could enhance their language development while identifying its limitations and ethical implications. In the digital and information literacy courses, activities focused on the broader societal and epistemological impacts of GenAI. Students investigated how AI technologies influence knowledge creation, information access, and authorship. Educators facilitated discussions that encouraged students to evaluate when and how GenAI might be used responsibly in their coursework, particularly in creative and research-based projects, without compromising academic ownership. Findings from this project highlight the value of dialogic approaches in Enabling Education programs to help students take ownership of their learning. Students’ initial uncertainty shifted to confidence in sharing their experiences of GenAI. Further, this approach created a safe space for open discussion enabling co-construction of knowledge about critical AI literacy. By centering student voice and agency, this model offers a compelling alternative to top-down enforcement of AI-related policies. It suggests that fostering critical engagement with GenAI builds students’ confidence and skills and strengthens trust between educators and learners. This work contributes to emerging conversations around AI literacy in higher education by proposing a framework that is both inclusive and responsive to the real-world complexities of student experience. It underscores the importance of pedagogical strategies that are participatory, flexible, and grounded in shared inquiry, particularly for students navigating the margins of academic culture.
Generative artificial intelligence has moved from a peripheral educational technology to an ordinary part of university study, writing, research, and assessment. Existing scholarship has documented substantial opportunities in personalised support, accessibility, feedback, and productivity, while also identifying persistent concerns involving misinformation, academic misconduct, privacy, bias, inequitable access, and cognitive dependence. However, the debate is too often organised around a narrow question: whether artificial intelligence should be allowed or prohibited. This conceptual article argues that the more consequential question is whether students retain responsible academic agency when artificial intelligence participates in their learning. Drawing on research in artificial intelligence in education, self-regulated learning, academic integrity, human-centred artificial intelligence, and institutional governance, the article develops a Responsible Academic Agency Framework. The framework integrates six interdependent domains: purpose and learning alignment; cognitive ownership and human judgment; epistemic verification; transparency and attribution; equity and privacy; and institutional governance and assessment validity. The central proposition is that artificial intelligence contributes to education only when it strengthens rather than substitutes the learner's capacity to set goals, evaluate evidence, make decisions, explain conclusions, and accept responsibility for academic work. The framework provides practical guidance for students, educators, university leaders, and technology providers, and offers a research agenda for evaluating learning quality rather than focusing solely on adoption. The article contributes to the literature by shifting the discussion from tool access and rule compliance toward the preservation, demonstration, and development of human agency within artificial intelligence-mediated higher education.
Generative artificial intelligence (GenAI) has rapidly entered educational practice as a tool for explanation, drafting, feedback, assessment support, and personalized learning assistance. Much of the early debate has focused on whether GenAI improves efficiency or threatens academic integrity. This article argues that the deeper educational question is whether GenAI use preserves or erodes student agency. From an educational perspective, agency is not reducible to student choice, motivation, or technical competence; it refers to learners' capacity to pose meaningful questions, construct knowledge through disciplined inquiry, regulate their own learning, exercise judgment, and assume responsibility for their intellectual work. Drawing on educational theory, self-regulated learning, knowledge-building pedagogy, cognitive offloading, and recent research on AI-assisted learning, this article develops a conceptual account of how GenAI may both support and weaken student agency. It identifies three core risks: the displacement of question-formation by prompt optimization, the substitution of fluent AI output for knowledge construction, and the diffusion of responsibility in authorship and judgment. The article then proposes an agency-preserving pedagogy of GenAI use built around six principles: problem-posing before prompting, process transparency, critical verification, scaffolded rather than substitutive assistance, assessment redesign, and teacher-guided ethical reflection. The central claim is that GenAI should not be treated as either a threat to be banned or an efficiency tool to be adopted uncritically. Its educational value depends on whether it is embedded in learning environments that strengthen students' epistemic, metacognitive, dialogic, and responsible agency.
Writing tools are ubiquitous and have become indispensable to many. However, the integration of AI assistance into writing process lacks a structured approach that addresses the complex, personal, and unique processes among writers. We present Writing Process Graphs, a framework designed to assist in understanding the underlying mechanisms that occur during creation. By offering a method to organize the writing process, this new work aims to mitigate potential negative impacts of assistive writing technologies by preventing irrelevant feedback, ensuring that these tools augment rather than interfere with the writer’s agency and way of writing. We envision this proposal for structuring process as a catalyst for discussions on AI writing assistance and writing process, psychological ownership, as well as human-AI creativity.
This study examines differences in confidence, independence, and authorship among university students based on the level of Artificial Intelligence (AI) tool usage in academic writing. A quantitative cross-sectional survey design was employed with 200 participants, who were categorised as frequent, occasional, or non-users of AI writing tools. Descriptive statistics and one-way analysis of variance (ANOVA) were used to compare the responses between groups. The findings showed that frequent AI users reported significantly higher confidence in completing academic writing tasks than did non-users. Findings regarding independence and authorship were interpreted with caution because of lower internal consistency within the scales; therefore, they were analysed at the item level. Independence showed mixed item-level findings, with significant differences observed for some independence-related items but not others. There were no statistically significant differences in authorship across the AI usage groups, indicating limited evidence of variability in students’ perceptions of writing voice and ownership. This study contributes to the ongoing discussion on AI-assisted academic writing in the EFL (English as Foreign Language) context and highlights the significance of guiding students towards the responsible and reflective use of AI in academic writing settings. These findings have implications for post-secondary educators and higher education policymakers.
Introduction Generative AI is shifting digital art from one-step generation toward interactive human-AI co-creation. Across two between-subject experiments (N = 600), we examined how creation mode and prespecified AI-involvement workflows related to perceived creativity. Methods In Study 1, participants completed digital art tasks under human-only creation, AI-only generation, or human-AI co-creation conditions. In Study 2, three workflows differing in AI involvement (low, moderate, high) were evaluated. Psychological ownership, perceived agency, creative flow, and perceived creativity were measured. Results In Study 1, human-AI co-creation produced higher perceived-creativity ratings than AI-only generation, with indirect effects consistent with a serial pathway through psychological ownership, perceived agency, and creative flow. In Study 2, the moderate-involvement workflow yielded higher perceived-creativity ratings and psychological states than the low- and high-involvement workflows. Discussion Perceived creativity in generative-AI art depends on how users experience contribution, control, and engagement during creation. Within the workflows tested, balancing generative assistance with consequential user choice fosters the most favorable subjective creativity evaluations.
The use of generative AI (GenAI) technologies has become increasingly prevalent across Australian higher education. As these tools become an integral part of learning, they also raise concerns about over-reliance and learner agency (Resnick, 2024), authorship, ownership, and problematic forms of co-creation (Bozkurt, 2024). In Australia, 31% of the student population identify as having English as an additional language or dialect (EAL/D) (Australian Government Department of Education, 2024). Apart from cultural and linguistic challenges, EAL/D students also need to adapt to new academic expectations and practices that account for the impacts of GenAI technologies (Kebble, 2023). Given the unique challenges facing EAL/D student cohorts with their diverse pre-existing technological backgrounds, a gap exists in understanding how EAL/D students perceive their own work and academic contributions in relation to their use of GenAI tools. This study explores EAL/D students’ AI-assisted learning experiences and perspectives on agency through qualitative interviews conducted with 34 Master of Digital Communication students over 18 months. Findings suggest that EAL/D students are instinctively adjusting and adapting their learning practices as they engage with these technologies and making agentic decisions that support their learning. To maintain what they considered to be original and creative in their work, students developed several strategies to distinguish their own work from AI-generated output. We also found that as EAL/D students’ engagement with AI increases, they develop critical self-scrutiny. They maintain agency and ownership over their assessed work by ensuring that their (human) creativity remains central throughout the learning process. While there are some commonalities, the students in our study took diverse paths, with some students developing more nuanced and balanced approaches to using AI than others. For these students, the shift enabled them to integrate GenAI strategically while maintaining their own voice and sense of agency.
No abstract available
The increasing use of Artificial Intelligence (AI) in education has provided students with convenient access to information and assistance in completing academic tasks, but excessive dependence may reduce independent learning, critical engagement, and information evaluation, particularly in English reading comprehension. This study aims to describe junior high school students’ dependence on Dola AI, their responses to AI errors and inaccessibility, strategies for verifying AI-generated answers, and perceptions of teachers’ roles in responsible AI use. A qualitative descriptive design was employed with 10 students from Grades 7 and 8 at State Junior High School 9 Satap Tondano, selected through purposive sampling based on mobile-phone ownership and active use of Dola AI for English reading tasks; data were collected through non-participant classroom observations and semi-structured interviews and analyzed thematically. The findings revealed three patterns of AI dependence: high dependence among 20% of participants, moderate dependence among 50%, and low dependence or independent learning among 30%. Highly dependent students experienced panic when AI was unavailable, trusted AI-generated answers without verification, and sometimes submitted system error messages as answers, whereas moderately dependent students used AI for convenience with limited verification and independent students verified information and continued working without AI. The findings indicate that AI dependence is shaped by students’ self-efficacy, digital literacy, and teacher guidance, highlighting the need to strengthen independent reading skills, critical AI literacy, verification strategies, and explicit teacher guidance for responsible AI use.
Large language models and generative artificial intelligence, as representative language intelligence technologies, are driving foreign language teaching to evolve from the traditional "teacher–student" dyadic structure toward a "teacher–student–AI" triadic collaborative paradigm. Based on the rapid development of artificial intelligence at home and abroad in recent years, this paper systematically reviews the current applications of AI in foreign language teaching, the transformation of teacher roles and identity crises, as well as ethical and equity-related challenges. The study finds that AI applications in foreign language teaching are growing rapidly, yet exhibit marked imbalances across regions, educational stages, and domains—namely, an overrepresentation of higher education research relative to basic education, and a predominance of writing assistance studies at the expense of oral communication and critical reading research. Teachers' roles are shifting from knowledge transmitters to human–AI collaborative instructional designers, while simultaneously facing a dual crisis of technological anxiety and identity recognition. Issues of ethical governance and educational equity urgently demand systematic responses. This paper argues that the future development of foreign language teaching must move beyond mere technology adoption; instead, it should be teacher-centered, constructing systematic coping strategies through the enhancement of AI literacy, human–AI collaborative instructional design, curriculum and assessment reconstruction, ethical governance, and institutional support.
This study investigates how a structured “paper-first” approach can support cognitive engagement during artificial intelligence (AI)-assisted academic work in higher education. Growing reliance on generative AI has increased concerns about surface-level learning, trial-and-error prompting, and reduced student ownership of academic writing. Using a two-year design-based research approach involving nine iterative cycles, the study developed and refined a framework that required students to construct manual logic diagrams before engaging with AI tools and expert system environments. Participants included a six-member interdisciplinary design team and 150 communication for academic purposes students from the faculty of information and communications technology at a South African university. Data were collected through iterative observations and focus group interviews and analyzed using grounded theory and constant comparative analysis. Findings indicate that the framework supported more structured reasoning, reduced inefficient prompting behaviors, and improved students’ ability to identify weaknesses in their own arguments before digital interaction. Students also reported increased confidence and greater ownership of their academic work. The study argues that structured pre-AI planning activities can strengthen critical thinking and academic literacy by positioning AI as a support tool rather than a substitute for cognitive effort. The findings contribute practical guidance for educators seeking to integrate AI into higher education while preserving student agency and deep learning.
Artificial intelligence (AI) promises major productivity gains, but it also raises fundamental questions about how technology can reshape people’s relationship to their work. Historical debates over industrialization warned that technological change could undermine people’s connection to work and sense of meaning. Similar concerns now surround AI, where the key issue may not be whether AI is used, but how it is used. Across a pre-registered experiment (N = 269) and a follow-up survey (N = 270), we examine how different modes of AI use affect the confidence individuals have in completing work without AI assistance (self-efficacy), their sense of ownership over task output, and the meaning they perceive in their work. Participants completed occupation-specific writing tasks under one of three conditions: no AI use, passive AI use (copying AI-generated content), or active collaboration (drafting first and then using AI to refine). We find that passive use undermined self-efficacy, psychological ownership, and work meaningfulness, with declines in efficacy and meaningfulness persisting even when participants returned to manual work. In contrast, collaborative AI use preserved psychological connection to the task, producing outcomes comparable to independent work. Although passive use initially boosted enjoyment and satisfaction, these benefits reversed once participants resumed manual work. A complementary real-world survey mirrored these patterns across tasks beyond writing. Together, these findings show that the psychological consequences of AI use hinge on how it is integrated into human workflows, underscoring that strategies promoting active, collaborative use may help capture AI’s productivity benefits while preserving human workers’ agency, competence, and connection to their work.