# Revision and source notes

**Work:** Learning from an AI Claimant: Scientific understanding and the justification of artificial consciousness claims  
**Author:** Hongju Liu  
**Report:** TA-TR-2026-09, version 1.2, 19 September 2026  
**Previous edition:** https://doi.org/10.5281/zenodo.22844928

## Method and status

This is a conceptual philosophical preprint using stipulated thought experiments. It reports no empirical or computational research experiments. The English paper contains 19 references; the Chinese companion is an argument guide, not a full translation or a separate paper. Substantial generative-AI assistance in research, drafting, criticism, revision and preparation is disclosed. AI-assisted review is not independent scholarly peer review, journal acceptance, or an exhaustive originality assessment. The author directed and authorized revision and publication; no separate final human line-by-line review is claimed.

## Changes from the published version 1.1

The revision strengthens the opponent's case: an AI may teach genuinely better assessment standards, identify errors in existing human criteria, and justifiably exclude mistaken dissent. A diagram-redrawing case separates a warranted correction of organizational assessment from acceptance of a consciousness bridge. Correcting one assessment does not automatically validate an endorsement requirement in another.

The selection argument is now explicitly conditional. If an observer already knows both the endorsement-based admission rule and that the certified group is nonempty, unanimity on that required endorsement supplies no additional support. Group existence, membership, size, reasons and other new information may still be evidence. The paper does not claim that learning adds no information or that every selected consensus is worthless.

The public reporting norm rests on an additional premise about presenting certification as a further check. It does not follow from the selection constraint alone. Mastery, derivation and belief are distinguished; a competent dissenter can derive a conclusion while questioning a strong but defeasible bridge. Cases also cover justified deference, opaque expertise, correction favoring AI, and symmetric owner-driven exclusion.

Two AI-assisted critical reviews compared the revised argument with version 1.1. The final revision addressed their concerns about information added by the existence of a group, the reporting norm's additional premise, and the distinction between competence and endorsement. This review narrows and clarifies the contribution; it does not guarantee originality, correctness or acceptance.

## Selected primary-source anchors

- Krenn et al. (2022), *On Scientific Understanding with Artificial Intelligence*, https://doi.org/10.1038/s42254-022-00518-3. AI-mediated understanding and the teacher-student-referee proposal; the source explicitly sets aside truth of the explanation.
- Gonzalez Barman et al. (2024), *Towards a Benchmark for Scientific Understanding in Humans and Machines*, https://doi.org/10.1007/s11023-024-09657-1. Learning, assessment standards and the teacher's understanding, especially sections 6–7.
- Nguyen (2020), *Cognitive Islands and Runaway Echo Chambers*, https://doi.org/10.1007/s11229-018-1692-0. Expert recognition and self-reinforcing expert selection are established precedents, not discoveries claimed here.
- Weatherson (2019), *Normative Externalism*, chapter 9. A substantive counterposition on higher-order evidence; the manuscript gives the publisher locator in its bibliography.
- Longino (1990), *Science as Social Knowledge*, chapter 4, https://doi.org/10.2307/j.ctvx5wbfz. Established role of effective criticism in scientific objectivity.

These selected anchors do not imply uniform full-text access to every reference. The manuscript's reference list is authoritative for the complete bibliography. The contribution remains an application and development of established epistemic resources to learning from an AI that is itself the consciousness claimant. Human understanding is not made a prerequisite of moral standing; political authority requires separate justification.

## Version and preservation integrity

This package is a revision of one paper, not a tenth research paper. Version 1.1 remains available at its own record. The version 1.2 DOI and all 14 public file hashes must be verified against the actual public record. SHA256SUMS.txt covers the other 13 files. OTS and Arweave completion is established only by separate version-specific receipts, never by this document or by a prior edition's proof.

Copyright © 2026 Hongju Liu. New manuscript and companion material: CC BY 4.0, https://creativecommons.org/licenses/by/4.0/. Cited third-party works retain their own rights.
