Why AI systems don't learn and what to do about it: Lessons on autonomous learning from cognitive science

The paper critically analyzes the shortcomings of contemporary AI models in achieving true autonomous learning. To address this, the authors propose a novel learning architecture directly inspired by human and animal cognition. This framework is designed around three integrated systems: System A for learning from observation, System B for learning from active behavior, and System M, which generates internal meta-control signals to flexibly switch between the A and B learning modes. The goal is to create systems capable of adapting effectively to dynamic, real-world environments, drawing lessons from how biological organisms adapt across evolutionary and developmental timescales.

(Source：arXiv.org)