The Geometry of Forgetting

A team of researchers has proposed a novel geometric framework to explain fundamental memory phenomena, including forgetting and false memories, in a paper published on the arXiv preprint server on April 26, 2026. The study, titled "The Geometry of Forgetting," challenges conventional biological explanations by demonstrating that the mathematical properties of high-dimensional spaces can naturally replicate key patterns of human memory without requiring specific biological modeling. The core of the research lies in modeling memories as points within a high-dimensional geometric embedding space. The scientists subjected this space to realistic cognitive pressures: noise, interference from other memories, and gradual temporal degradation. Remarkably, this simple geometric system produced quantitative signatures that closely match empirical human data. A primary finding is the emergence of a power-law forgetting curve, with the model's exponent (b = 0.460 ± 0.183) aligning almost perfectly with the established human average of approximately 0.5. This suggests that the familiar pattern of rapid initial memory loss followed by a slow, long-term decline may be a fundamental property of information storage in complex, noisy systems, rather than a quirk of neural biology. Furthermore, the model provides a compelling explanation for memory distortions and false memories. In a high-dimensional space, the natural overlap and interference between memory representations can cause one memory to be misidentified as another or for a composite memory to be recalled. This geometric interference offers a parsimonious, principle-driven account for why we sometimes vividly remember events that never occurred. The work represents a significant shift in perspective, moving the explanatory focus from the biological 'hardware' of the brain to the abstract, mathematical 'software' governing how representations interact, with potential implications for both cognitive science and artificial intelligence research.