Multi-Strand (12/7) Stand-alone Paper Set

description

416 Informal Generative AI Practices in Science Education: Patterns; Usage Levels; and Impact on Higher Education Mehedi Hasan Anik ORCID iD1; S M Hafizur Rahman1; Nushat Khan2; Shahriar Nafees Chowdhury Raaz3; Fayek Al Hasnain4 1University of Dhaka; Dhaka; Bangladesh. 2Eastern Illinois University; Illinois; USA. 3Green University of Bangladesh; Dhaka; Bangladesh. 4Ohio University; Ohio; USA Abstract The rapid rise of generative AI (GenAI) tools has transformed academic practices; yet their informal and unregulated use raises critical concerns for higher education. In Bangladesh's Science Education context; students are increasingly adopting GenAI without institutional guidance; creating challenges for teaching; learning; and assessment. This study aimed to explore the patterns; levels; and potential impacts of such informal use. A mixed-methods; explanatory sequential design was employed; involving a survey of 83 Science Education students and semi-structured interviews with five faculty members. Quantitative results showed that 96% of students used GenAI for academic purposes; with nearly 69% engaging at advanced creative levels. Qualitative findings revealed inconsistent teaching responses; over-reliance on AI by students; and threats to assessment integrity. The study underscores the need for AI literacy training; clear policies; and AI-inclusive assessment strategies to ensure responsible integration while preserving educational quality. Strands Strand 12: Technology for Teaching; Learning; and Research 542 Enhancing STEM Education with Generative AI: Insights from Students and a Teacher in Thailand Parin Chawalit ORCID iD1;2; Xinying Yin3 1California State University; San Bernardino – MA in STEM Education (Class of 2025); San Bernardino; CA; USA. 2Royal Thai Government Scholarship Recipient; Thailand. 3California State University-San Bernardino; San Bernardino; CA; USA Abstract This qualitative case study explored how generative AI (GenAI) tools supported STEM education in a Thai secondary school. During a one-day STEM camp; 50 students (grades 9–11) and one science teacher used ChatGPT and Gemini in a bridge design challenge. The study examined GenAI's influence on student learning within the engineering design process (EDP); student motivation; and teacher perceptions. Thematic analysis of observations; surveys; interviews; and AI interaction logs found GenAI mainly supported early design phases—brainstorming; comparing materials; and visualizing structures. Challenges included students' weak prompt-writing skills; difficulty interpreting vague or irrelevant AI responses; and limited application of AI feedback during testing and revision. Students' learning and motivation varied based on their digital literacy; confidence; and prior experience; with some students preferring familiar platforms (e.g.; Google and YouTube). The teacher valued GenAI as a scaffold for creativity and discussion but noted risks of over-reliance; uncritical copying; and equity gaps for under-resourced students. She emphasized the need for prompt literacy; critical evaluation; and structured integration across all EDP phases. This study adds to research on GenAI in non-Western K–12 STEM contexts; highlighting the importance of culturally responsive pedagogy; ethics instruction; and teacher facilitation for effective AI use. Strands Strand 12: Technology for Teaching; Learning; and Research 590 Generative AI as Science Tutors: Can LLMs Combine High-Level Problem Solving and Advanced Tutoring? Paul Tschisgale ORCID iD1; Peter Wulff ORCID iD2 1Leibniz Institute for Science and Mathematics Education; Kiel; Germany. 2Ludwigsburg University of Education; Ludwigsburg; Germany Abstract Generative AI offers novel opportunities for individualized and adaptive science learning; for example through large language model (LLM)-based feedback systems. To date; most research on such systems has focused on supporting conceptual understanding. Our study addresses a more complex and multifaceted practice: science problem solving in the domain of physics. To advance our understanding of how AI can support science problem solving; we first examined the capabilities of recent LLMs to solve advanced science problems. To examine the tutoring; we developed an LLM-based feedback system grounded in evidence-centered design to assess the complexity of students' problem-solving processes. Using data from student interactions with the system; we investigated how learners perceive the LLM-generated feedback in terms of usefulness; as well as perceived and actual correctness. Our findings indicate that LLMs can solve advanced science problems on par with; or even better than; high-performing students. However; while problem-solving capabilities were high; producing feedback that was both useful and accurate proved more challenging. Although; students generally perceived the feedback as useful and physically correct; it occasionally contained errors. We problematize the unreflected use of LLMs and LLM-enhanced feedback systems; and outline requirements for implementing LLMs in productive ways for improving science learning. Strands Strand 12: Technology for Teaching; Learning; and Research 346 Mindset in Physics and AI Use Behaviors in the Digital Transformation Era İpek Derman ORCID iD; Sevim Bezen ORCID iD Hacettepe University; Ankara; Turkey Abstract This study examines the relationship between pre-service physics teachers' mindset in physics and their use of artificial intelligence (AI) during the learning process. Designed as a qualitative case study; the research involved semi-structured interviews with 14 pre-service teachers who used ChatGPT in core physics courses such as mechanics and electricity. Data were coded according to growth and fixed mindset indicators; while AI use was categorized as either supporting or not supporting the mindset. Findings revealed that most participants associated physics achievement with innate intelligence and used AI mainly to obtain quick answers or find solutions to given problems. Only one participant demonstrated growth mindset characteristics; using AI for exploration; deepening the learning process; and generating new questions. Four profiles were identified based on mindset and AI use; with the majority falling into the "fixed mindset–non-growth-supporting AI use" category. The study shows that mindset plays a decisive role in shaping the pedagogical potential of AI tools and that approaches fostering a growth mindset are needed to reveal the transformative role of AI in physics learning.