SophyBARA
Abstract
We present SophyBARA, a Retrieval-Augmented Generation (RAG) based multi-agent debate system where anthropomorphic capybara-themed philosophers engage in real-time debates over users' everyday concerns. Implemented as a web-based interface, the system facilitates a progressive interaction cycle through a 1:1 Chat for initial exploration and a multi-agent Arena for observed discourse. SophyBARA employs a unique four-axis framework---Ontological, Axiological, Methodological, and Process/History---to transform natural language inquiries into structured philosophical agendas. By adopting the Cross Examination Debate Association (CEDA) format, the system visualizes the logical trajectory of conflicting arguments, allowing novices to track complex reasoning. Users participate through a turn-based intervention mechanism, enabling them to directly challenge agents or invoke a third philosopher’s perspective to redirect the discourse. Grounded in Vicarious Learning and Cognitive Apprenticeship, SophyBARA demonstrates a novel interaction model where AI serves as a cognitive scaffold, lowering the threshold for philosophical literacy and fostering active critical inquiry in the digital age.
System Architecture
