Human–GenAI Interaction in Higher Education: A Multi-Agent Fuzzy Delphi Framework for Co-Creative Environmental Learning
This study introduces a novel multi-agent human–GenAI fuzzy Delphi framework to empirically model and optimize co-creative environmental learning in higher education, revealing how orchestrated interactions between educators and diverse AI systems foster systems thinking, engagement, and adaptive learning strategies through three iterative interaction loops.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The world of environmental education is facing a profound shift. For decades, the approach to teaching about climate change, pollution, and biodiversity loss has often relied on linear lessons: a teacher presents facts, and students absorb them. Yet, the problems we face are not simple lists of facts; they are tangled, living systems where a change in one area ripples through many others. To understand these complex webs, learners need to think in systems, imagining how actions today might reshape the world tomorrow. This is where a new kind of technology enters the classroom. Generative artificial intelligence, the same technology that can write stories or solve equations, is being tested not just as a tool for delivering information, but as a partner in the thinking process itself. The question is no longer whether machines can generate answers, but how humans and machines can work together to create new understanding that neither could reach alone.
In a recent study, researchers from the University of Tehran set out to map exactly how this partnership might work in higher education. They were not interested in simply asking a single computer program for advice. Instead, they treated artificial intelligence as a group of distinct thinkers, each with its own way of processing information. The team engaged eight different text-based artificial intelligence systems, ranging from widely known general assistants to specialized models designed for research or logical reasoning. They placed these systems into a structured conversation, asking them to help design learning experiences for complex environmental topics like water management and the circular economy. The goal was to see how these different digital minds would approach the same problem, what patterns emerged when they worked together, and how human experts could guide this collaboration.
The researchers found that these artificial intelligence systems did not all think alike. Some were better at generating ethical arguments and reflecting on values, while others excelled at building logical structures or simulating the consequences of specific decisions. One system might focus on the safety and ethics of a proposed solution, while another would immediately start modeling the data behind it. This diversity suggested that relying on just one artificial intelligence tool might limit the depth of learning. Instead, the most effective approach appeared to be a coordinated effort where a human teacher acts as a conductor, bringing together the unique strengths of different systems to create a richer learning environment. The study identified that when humans and these varied artificial intelligences interact, they move through three distinct, repeating cycles. First, they explore a problem and generate initial ideas. Next, they simulate the outcomes of those ideas and reflect on what might go wrong or right. Finally, they implement the best solutions and adapt them based on what they learned, creating a continuous loop of improvement.
To make sense of these interactions, the researchers brought in a panel of four human experts in environmental education and technology. They used a rigorous method to test the ideas generated by the artificial intelligence systems, asking the experts to rate the importance and clarity of each concept. This process, which involved multiple rounds of feedback, helped filter out vague suggestions and solidify a clear framework. The experts confirmed that the most valuable interactions were those that kept the human in the driver's seat for judgment and ethics, while using the machines to expand the range of possibilities. The study highlighted that while artificial intelligence can simulate complex scenarios and offer instant feedback, it cannot fully grasp the emotional, historical, or deeply political nuances of environmental issues. Therefore, the human role shifts from simply teaching content to orchestrating a complex dance between student, machine, and real-world context.
The final picture that emerged is one of a dynamic learning ecosystem. In this model, the teacher is no longer just a source of information but a facilitator who helps students navigate the relationship between human intuition and machine calculation. The study suggests that for this to work, educational systems must be redesigned to support these new interactions, ensuring that students learn to question the biases of the machines they use and to value their own critical thinking. While the researchers noted that their findings are based on simulations and expert reviews rather than long-term classroom trials, the framework they developed offers a concrete path forward. It proposes that by treating artificial intelligence as a diverse set of cognitive partners rather than a single tool, educators can help students develop the deep, flexible thinking required to solve the environmental challenges of the future. The work underscores that the future of learning lies not in replacing human teachers, but in empowering them to guide students through a new, shared landscape of discovery.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.