Go to DBLP homepageREALM: A relevance-driven layered protocol for human-AI collaboration in STEM education [Special Issue on Artificial Intelligence for Education: A Signal Processing Perspective]
Abstract: Modern science, technology, engineering, and mathematics (STEM) education faces a growing challenge: how to maintain human relevance in an era of rapidly advancing generative artificial intelligence (GenAI) [1], [2], [3]. The 2024 Nobel Prizes in Physics and Chemistry, awarded for AI-driven discoveries, underscore a shift in how scientific knowledge is produced. GenAI systems[1], [2] not only retrieve but synthesize knowledge across disciplines, often outpacing expert reasoning. This has led to a “cognitive gulf”: a widening gap between the probabilistic knowledge space of GenAI and the bounded cognition of learners and teachers. We propose that STEM education be reconceptualized as a human-in-the-loop (HITL) signal processing system, in which all stakeholders—including teachers, students, GenAI, and policy actors—evolve as adaptive, collaborative learners; exchanging, modulating, and filtering learning signals (such as adaptive prompts, feedback, and project outcomes) in real time. Drawing from systems theory [4], [5], we frame these interactions as composable, ethically governed operations, scaffolding cognitive leverage rather than resisting AI’s stochastic reasoning. By clarifying stakeholder roles and aligning flows with human autonomy, the cognitive gulf can become a conduit for personalized mastery and collaboration. We introduce the Relevance-Enhanced Adaptive Layered Model (REALM), a protocol that not only establishes a novel, self-contained framework for GenAI-enabled STEM education but also unifies and advances foundational educational models—including objectivism [6], constructivism, collaborativism, and socioculturism—by mapping classic learning theories onto a closed-loop, explainable ecosystem centered on relevance, adaptability, and growth.
External IDs:dblp:journals/spm/ZhangRS26
Loading