The Transformative Collaboration of Human Intelligence and Artificial Intelligence in Designing Knowledge-in-Use Science Assessment for Learning

This study investigates how human and artificial intelligence (AI), specifically GPT-4, can collaborate to design knowledge-in-use science assessments aligned with three-dimensional (3D) learning goals. Using a design-based research approach grounded in Evidence-Centered Design and the NGSA framework, we guided GPT-4 through structured development processes supported by interdisciplinary expert feedback. Focusing on two NGSS performance expectations (3-PS2-1 and 3-LS4-3), we examined how human scaffolding—prompt design, expert evaluation, and iterative refinement—shaped the quality, equity, and classroom usability of AI-generated tasks. Findings indicate that GPT-4, when supported by principled human guidance, can co-produce standards-aligned and instructionally relevant assessments. We identified three key supports for productive human–AI collaboration: unpacked disciplinary goals, structured prompts aligned with learning performances, and expert-informed exemplars. Expert feedback improved linguistic accessibility, cultural relevance, and 3D integration across iterations. Notably, GPT-4 began anticipating feedback categories over time, suggesting emerging responsiveness to design expectations. This work offers practical insights into using generative AI as a collaborative design partner—rather than an automated tool—to support the development of equity-focused, NGSS-aligned classroom assessments. We propose a transferable refinement framework that can guide teachers, designers, and developers in producing high-quality tasks, particularly in under-resourced settings. By making the co-design process visible, this study contributes to the growing field of human–AI collaboration in education and offers actionable design heuristics for integrating AI into formative assessment practice.