Endogenous Reference Fields and Experiential Attribution: A Slow Shared-Modulation Hypothesis and Its Discriminating Tests
Preferred published edition, checked 12 October 2026: v1.2 — Endogenous Reference Fields and Experiential Attribution: A Multidimensional Process Theory of Experience, Subject Boundaries, and Artificial Realization. DOI 10.5281/zenodo.22855837 · PDF. The earlier edition described below is preserved as version history.
《内生参照场与体验归属:慢变共享调制假说及其判别性检验》
Hongju Liu(刘烘炬) · TA-TR-2026-10 · Version 1.1 · 20 September 2026
Published preprint; not externally peer reviewed. Complete Chinese text with an English abstract. Substantial AI assistance in literature research, theoretical development, code, critical revision and manuscript production; the human author assumes publication responsibility.
- DOI: 10.5281/zenodo.22854705 · Zenodo record and 18 files
- 中文全文 PDF · 可编辑 Markdown
- Method and sources · Code and reproduction instructions
- Publication receipt · Exact file manifest
- BibTeX · RIS · CSL JSON
Abstract
We propose endogenous reference fields (ERFs) as a hypothesis linking reference-dependent experience to slow, content-updated states that jointly modulate multiple content channels. Shared modulation, content maintenance, and timescale separation identify a candidate organization; experiential attribution remains a separate bridging hypothesis. We distinguish modulation, mediated return influence, decomposition distance, response differentiation, and intrinsic persistence. A symmetry argument establishes equal persistence profiles within a transitive orbit under an invariant intervention protocol, excluding strict timescale separation at that specified grain rather than at every possible macroscopic grain. Conditional contrasts with IIT, global workspace theories, functionalism, dendritic information theory, and artificial architectures clarify the proposal’s commitments. A small computational proof of concept trains the same two-dimensional state-space architecture on three tasks with three fixed seeds each. The resulting probe-dependent profiles exhibit sustained shared modulation, near-zero modulation, or rapidly decaying modulation. Finite-sample mean modulation is distinguished from the theoretical supremum, and numerical residuals are not treated as exact zeros. The experiment supports training-dependent organizational variation, while neither establishing the complete ERF criteria nor demonstrating machine consciousness. Finite construction checks and training experiments are reported separately. Empirical predictions address the bridging hypothesis, whose necessity and sufficiency remain conjectural.
Publication and version history
All 18 public files passed anonymous complete-file byte-count and SHA-256 readback; the DOI resolves to this version. The nine training runs exactly reproduced all 171 reported numeric fields in the recorded environment. These are small computational organizational tests, not evidence that an artificial system is conscious. No arbitrary consciousness threshold or single aggregate consciousness score is introduced.
Version 1.1 is a revised version of v1.0 DOI 10.5281/zenodo.22852885, within concept DOI 10.5281/zenodo.22852884. The original v1.0 files and v1.0 receipt remain unchanged. This is still the tenth distinct paper; the series contains ten papers, not eleven.
CC BY 4.0 applies within the author’s rights. Third-party rights remain intact. Publication integrity does not establish philosophical truth, peer review or indexing. Version-specific OTS / Arweave preservation status is tracked separately; earlier proofs do not cover these revised bytes.