Abstract
This disclosure describes an open architecture for the physical infrastructure on which governed AI systems run, applied identically at three scales: a personal device in the home, a community site serving a locality, and regional backbone centres. At every scale the hardware separates a control plane, a data plane and a safety plane, and the safety plane has its own power and its own communication path. The safety planes of all levels are joined into a safety mesh, separate from the data network, so that a stop, hold or reset can pass between levels, and be confirmed, even when the data network is down or compromised. Because the architecture, interfaces and certification are the same at every scale, a certified software bundle can move between a personal device, a community site and a backbone centre without re-certification of the bundle. Data stays at the level and in the region where its owner placed it; models and permitted outputs travel to the data. Community sites federate with one another, holding encrypted restore copies for each other at physically separate locations, so that no single site is essential. Every component carries a lifelong identifier and passport, is checked on arrival and measured at every boot. A complete site is certified as installed, including its power, network paths, physical security and the separation of its safety plane, with the separation re-tested in place at intervals. Reference designs specify functions and interfaces rather than parts, so that many manufacturers can build to them.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Wang, Edward D. H., "SI Infrastructure™: One Three-Plane Architecture for AI Computing at Personal, Community and Regional Scale, with a Safety Mesh on Its Own Power and Paths, Data that Stays in Its Region while Models Travel, Federated Restore, and Certification of Sites as Installed", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12277