Open–Closed World Communication: A Boundary-Centered Architecture for Generative AI and Logical Reasoning

Generative AI and logical reasoning systems are often contrasted as probabilistic and symbolic approaches. This paper argues that a more consequential distinction concerns the boundaries of the knowledge worlds in which they operate. Generative AI works effectively in an open-ended semantic space: it combines learned knowledge, current context, retrieved information, and analogy to produce interpretations and candidate solutions. Logical systems such as Prolog, SMT solvers, and constraint engines operate more effectively in bounded spaces whose facts, rules, and admissible inference procedures are explicit. The paper proposes Open–Closed World Communication, a boundary-centered architecture that connects these two modes without collapsing one into the other. A third layer, Boundary Space, governs the transfer of knowledge through explicit states such as Candidate, Hypothesis, Grounded, Verified, and Committed Assertion. It preserves provenance, applies verification and policy, and prevents generated output from becoming authoritative knowledge merely because it was produced by an AI system. The architecture is bidirectional: failure or insufficiency in a Closed Reasoning Space can trigger controlled escalation to an Open Reasoning Space; validated results can then be committed back into a bounded knowledge snapshot and re-evaluated. The paper further places JDR (Job Descriptor) above these spaces as an intent-oriented orchestration layer that specifies reasoners, bridge policies, human checkpoints, and execution conditions. The result is a conceptual framework for combining generative discovery with logical rigor while retaining traceability, accountability, and human control. The proposal is positioned relative to logic programming, closed-world reasoning, retrieval-augmented generation, neural-symbolic AI, SMT solving, provenance models, and human-AI interaction. A minimal prototype and research agenda are presented.