The Standpoint Threshold: Architectural Conditions for Processing from Somewhere
Philosophies July 20, 2026 D.H. Lange
Before asking whether an AI system could be conscious, we must first ask what internal organization would make that question meaningful. This paper proposes three jointly necessary conditions: the system must maintain persistent internal context, have differentiated internal possibilities, and operate in a closed loop where outputs feed back into processing. These conditions mark when a system's processing unfolds from an internal standpoint rather than as a chain of local transitions. Simple feedback controllers clearly lack this organization; standard transformer inference is at most a bounded near-case; world-model architectures are more serious candidates. The paper makes no claim that such systems are conscious, but clarifies when consciousness-oriented inquiry has a determinate architectural target.