The schema spectrum: Emergent structures and levels of abstraction in AI and the brain

There is a long history of interplay between the brain sciences and AI in the area of schema theory. Schemas are typically defined as abstract mental structures representing prior knowledge, experiences, and concepts that capture how events unfold in different contexts and that alter how we learn new information. Classical models have treated schemas as being distinct from both detail-rich episodic memories and general semantic knowledge. Motivated by learning phenomena in modern generative AI, we propose that earlier connectionist theories that did not articulate any strict division between schemas and other declarative memories should be revived. According to this perspective, schemas are not a distinct set of mnemonic objects in the brain; rather, they are a conceptual tool that scientists use to describe how existing knowledge can exist along a spectrum of abstraction.