Designing An LLM-Based Multi-Agent System for Customized Science Assessments Through Multidisciplinary Human-AI Collaboration

Tingting Li

Despite growing interest in generative AI, science education continues to confront two enduring challenges: the persistent difficulty teachers face in designing assessments that reflect the three-dimensional vision of the Next Generation Science Standards (NGSS), and the lack of tools that meaningfully respond to the linguistic and instructional realities of today's classrooms. This project examines how human‚ AI collaboration can be reimagined not as automation, but as a co-constructive process that honors teacher judgment, centers classroom knowledge, and supports students' scientific reasoning. Grounded in hybrid intelligence and collective sensemaking theory, the project will develop a large language models (LLMs)-based multi-agent system to support elementary teachers in designing NGSS-aligned formative assessments. The system will integrate dimensions often treated separately‚ disciplinary rigor, classroom context, student engagement, and language accessibility‚into an interactive, teacher-guided design process. Using a design-based research approach, the study will investigate how teachers and students engage with the system, how they shape its outputs, and how this co-design process informs both instruction and learning. Rather than positioning AI as a static tool, the project conceptualizes it as a responsive partner in pedagogical dialogue‚ adaptable, imperfect, and shaped through iterative human input. This research will explore how such human‚ AI partnerships can generate usable knowledge for teachers while attending to the practical and epistemic tensions of real-world implementation. Findings will contribute to theories of "assessment as learning" and offer a novel framework for designing instructional tools that support teacher agency, student participation, and the situated integration of AI in science education.

About Tingting Li

Tingting Li is an Assistant Professor of Science Education at Washington State University. She holds two distinct Ph.D. degrees: one in Educational Psychology and Educational Technology from Michigan State University, and another in Science Education from Northeast Normal University, completed through a joint study abroad program with MSU. With a disciplinary background in Chemistry and K-12 teaching experience across China and the U.S., her research bridges psychological, technological, and pedagogical perspectives to support accessible science teaching and learning opportunities. She also holds affiliated faculty appointments in Educational Psychology, reflecting her interdisciplinary commitment to understanding learning from both cognitive and motivational perspectives. Her scholarship centers on exploring the mechanisms and applications of human-AI collaboration in augmenting teaching and learning. She investigates how teachers, students and AI can co-design culturally responsive, linguistically accessible, and contextually adaptive learning environments. Rather than treating AI as a neutral or autonomous actor, she conceptualizes AI as a co-participant whose value emerges through thoughtful integration with classroom expertise and local epistemologies. She does not approach AI with techno-optimism, nor does she advocate for refusal. Instead, she carves out a third stance: as a structural resister and rebuilder, committed to designing alternatives that challenge dominant technological logics while centering teacher and student voices, contextual relevance, and accessible learning opportunities. Supported by grants such like Microsoft and Spencer, she advances new models of AI-augmented instruction that position educators and learners as co-designers, not passive recipients.