中立基底：持续性解释分歧下共享记录的设计约束
Neutral Substrates: A Design Constraint for Shared Records Under Persistent Interpretive Disagreement
AI 导读
一篇 arXiv cs.AI 论文提出"中立基底"（neutral substrate）设计约束，将共享问责记录中的共同指称承诺与归因性解释断言分离，使持不同因果与责任解释的各方仍可共用同一记录。论文通过可容许框架、指称承诺与解释性不承诺等概念形式化该分离，证明在指称共同地面条件下，将基底层承诺限于指称承诺或许可归因命题足以实现相对所有可容许框架的语义中立，并明确了该结果作为条件性充分性的边界。
正文 · 原文
Abstract:Shared accountability records must often remain useful to parties who disagree about causation, responsibility, and normative interpretation. Suppressing these claims would erase information needed for accountability; asserting one disputed interpretation as foundational would privilege it. A neutral substrate instead separates shared referential commitments from attributed interpretive assertions.
This paper formalizes that separation through admissible frameworks, referential commitments, permitted attributions, interpretive non-commitment, and extension stability. Under Referential Common Ground, restricting every substrate-layer commitment to a referential commitment or a permitted attribution proposition is sufficient for semantic neutrality relative to every admissible framework. It is not a design-time guarantee: such a guarantee requires a permitted basis establishing the neutrality result, rather than only its truth.
The essential assumption is joint compatibility of the complete referential commitment set with each admissible framework. Separate compatibility of individual commitments is not sufficient, and admissible frameworks need not be mutually consistent. Furthermore, the formal referential interfaces do not independently establish the identity criteria, denotation relations, attributional bases, or interpretive adequacy required by a particular application. When those premises cannot be justified, neutrality cannot be certified by the present construction.
The contribution is a precise conditional layer restriction and an explicit account of its semantic and scientific boundaries. Attribution and provenance are established representational mechanisms; their presence alone cannot establish referential fixing.
| Comments: | 26 pages. Revised main result as conditional sufficiency for semantic neutrality; distinguished design-time guarantees; clarified assumptions and limits; aligned with Lean 4 formalization |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2601.14271 [cs.AI] |
| (or arXiv:2601.14271v4 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2601.14271 arXiv-issued DOI via DataCite |
Submission history
From: Denise M. Case [view email]
[v1]
Thu, 8 Jan 2026 02:36:13 UTC (35 KB)
[v2]
Wed, 8 Jul 2026 02:17:10 UTC (17 KB)
[v3]
Fri, 2 Oct 2026 18:55:10 UTC (17 KB)
[v4]
Sun, 25 Oct 2026 14:57:23 UTC (20 KB)
来源：arXiv cs.AI · arxiv.org