Scientists rethink how AI systems store and retrieve information
Researchers have abandoned the traditional library-like model of digital memory in favor of a landscape-based system that mirrors how brains actually work. The approach could make AI systems more efficient and responsive, with implications for everything from edge computing to real-time decision-making in autonomous systems.
Originaltitel: The Architecture of Resonance
What if memory were not a library, but a landscape? This essay traces the development of SPIRALbase and Hybrid-J - a research program that replaces the idea of memory as indexed storage with memory as a physical potential: a high-dimensional terrain of attractor basins that a system falls into rather than looks up. Beginning with the question of whether biological memory could be realised digitally, we walk through three generations of architecture: from early context-gated attractor networks trained by pseudo-likelihood, through metric-derived projectors that open precise subspaces on demand, to Hybrid-J -- a block-structured substrate with a shared core, local context blocks, and controlled sparse bridges between them.