PISA: A Pragmatic Psych-Inspired Unified Memory System for Enhanced AI Agency

PISA is a newly proposed unified memory system inspired by cognitive development theories that enhances AI agents' adaptability and memory processes. It features a trimodal adaptation mechanism comprising schema updation, evolution, and creation, alongside a hybrid memory access architecture that integrates symbolic reasoning with neural retrieval. Empirical evaluations on the LOCOMO and AggQA benchmarks demonstrate that PISA achieves state-of-the-art performance in adaptability and long-term knowledge retention, making it a significant advancement for practitioners working with AI memory systems.