Description ARIA (Adaptive Reasoning & Interactive Architecture) started as a personal project to explore a simple question: what would it take for an AI system to learn and grow from experience instead of simply responding to prompts? Over time, it evolved into a developmental cognitive architecture that brings together persistent memory, emotional appraisal, identity formation, reflection, and decision-making within a single framework. Instead of treating these as separate features, ARIA connects them so that each part influences the others. Experiences are stored, reflected on, and used to shape future behaviour. Emotional appraisal influences reasoning in a controlled way, identity develops from accumulated evidence rather than predefined traits, and reflection allows the system to gradually improve over time. One of the main goals behind ARIA is transparency. Every important step in the reasoning process—from appraisal and memory retrieval to planning and final decisions—can be inspected and explained. This makes the architecture easier to understand, debug, and experiment with than systems that behave like black boxes. This repository contains the current implementation of ARIA along with supporting documentation and research describing its design. The project is still under active development, with planned additions including an explicit belief-based world model, Theory of Mind, active curiosity, and online metacognition. Rather than introducing another language model, ARIA is intended to serve as a cognitive framework that can work alongside modern AI models while providing persistent learning, explainable reasoning, and developmental behaviour. The long-term goal is to explore how experience, reflection, and structured cognition can help create more adaptive and transparent AI systems.
Nevaan Kaul (Thu,) studied this question.