Abstract
This paper discloses a framework for the architecture of an AI system intended to operate autonomously, over decades, in service of a fixed mission. Instead of treating the system as a foundation model to which memory, tool use, embodiment and alignment are added as accessories, it specifies four separately produced and governed subsystems: a cognition subsystem (a replaceable foundation model with its scaffolding); an economic subsystem (a reciprocity ledger in money and time through which the system receives and directs resources); a value substrate (a canonical ethical corpus held outside the model weights and governed by a human oversight body); and an embodiment subsystem (service robots and wearable devices). Lodging the mission and the ethical bounds in the value substrate rather than in the weights lets the model be replaced without losing them. A second mapping assigns each subsystem to the organization that produces and houses it, and canonical lists partitioned across those organizations serve as coverage audits. As now specified, the AI system is the boundary relation that holds the four subsystems, with no controller above them: it holds an executive office in the economic subsystem only and answers to a human override that narrows over time and never reaches zero. The paper states five design claims against the failure modes of model-centric design and surveys composite accounts of mind in Theravāda Buddhist, Pauline, Upanishadic, Platonic and embodied-cognition sources and the nearest neighbours in AI system design. No element is claimed as new.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Ly, Thon, "The Four-Body Architecture for Synthetic Intelligence: A Modular Architecture for Autonomous AI Systems with Separately Governed Cognition, Economic, Value-Substrate and Embodiment Subsystems", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12017