Exploring a biocentric LLM-based assistant in environmental decision-making with more-than-human representation of the Tagus Estuary

Abstract Integrating nonhuman perspectives into human decision-making is crucial for addressing environmental challenges, yet practical methods are scarce. This paper explores using Large Language Models (LLMs) to represent nonhuman entities and influence human decisions toward ecological sustainability. We address two research questions: (1) How can LLMs be instructed to represent nonhuman entities? (2) How does a biocentric AI assistant affect people’s decision-making? We developed a biocentric AI assistant by training an LLM on ecological data about a specific estuary. In an experimental study, we compared this biocentric assistant with an anthropocentric counterpart in a speculative decision-making task involving the location of a new university campus. Results show that while the biocentric assistant did not significantly change participants’ final decisions, it prompted increased reflection on environmental considerations in their justifications. Our findings demonstrate AI’s potential and challenges in incorporating nonhuman perspectives into decision-making, contributing to more-than-human HCI, and promoting sustainable outcomes.