abstract
- Marr’s 1971 simple memory theory of archicortex set the stage for over four decades of subsequent research on the medial temporal lobe memory system. Marr’s ideas contributed fundamentally to current computational models of the hippocampus. In particular, this theory led to the widely held view of the hippocampus as a temporary memory store that accurately memorizes events by creating orthogonalized representations (pattern separation), while using its associative pathways to retrieve the original event from a partial cue. Although Marr’s model has garnered considerable empirical support, subsequent research also calls into question some of its key assumptions. Future extensions to the model must address these data. There have also been several game-changing neuroscientific discoveries, such as neurogenesis in the adult hippocampus and the fate of remote memories in densely amnesic patients; these findings require us to rethink the notion of the hippocampus as a simple memory system. Even the long-held view that the hippocampus performs a fundamentally different computation than the neocortex may be incorrect. Converging evidence suggests instead that the hippocampus is at the apex of a multi-level cortical hierarchy that encodes progressively more abstract information at each level, and is as involved in predictive perceptual coding as it is in memory.