05 / Reading list
A short route into original work, with enough context to know what each source offers and what it leaves open.
A small starting shelf Begin with Birch for the ethical framework, Butlin and colleagues for assessment methods, and Seth for a substantial challenge.
Start here / Precaution
Published book · Oxford University Press
A framework for reasoning about sentience and proportionate precaution under uncertainty. Investigation priorities, sentience candidature, and policy decisions have different thresholds.
Keep in view: Its proposals are contestable; it is not a diagnostic standard.
Source checked 9 October 2026
Start here / Assessment
arXiv report · v3, 22 August 2023 · Peer review not verified
A theory-led framework for investigating possible consciousness in machine intelligence.
Keep in view: Provisional indicators; 2023 case studies; not a direct assessment of welfare.
Source checked 9 October 2026
Start here / A substantial objection
Journal article · Behavioral and Brain Sciences 49, e315 · First online 2025
A developed challenge to computational sufficiency, examining whether the processes of living systems matter for consciousness.
Keep in view: Biological dependence is not established. Read alongside his 2026 author response.
The stuff matters: Consciousness, computation, and biology (2026 response)
Source checked 9 October 2026
Go deeper / Felt value
Journal article · Philosophy and Phenomenological Research 97(3), 658–680 (November 2018) · First online 25 March 2017
An account of valence as a motivating representation of value. Useful for asking what separates value processing from an unpleasant experience.
Keep in view: A theoretical account, including unconscious processing; no machine-intelligence experiment.
Source checked 9 October 2026
Go deeper / Self-reports
Preprint · arXiv v1, 14 November 2023 · Peer review not verified
A proposal for studying when self-reports might be informative about moral status.
Keep in view: Its proposed tests do not yet provide a validated assessment of experience. Some authors and arguments overlap with other sources here.
Source checked 9 October 2026
Go deeper / Internal representations
Preprint · arXiv v2, submitted 25 September 2026 · First version 14 September 2026 · Peer review not verified
A mechanistic and behavioral study of pain-related representations in language models. The revised controls do not support reliable relief-seeking, and some harmful choices persist without an offer of relief in fine-tuned Qwen models.
Keep in view: Representations and behavioral changes do not establish suffering; controls and generalization need further work. This annotation was updated from v1 to v2, which revises the relief-seeking interpretation.
Source checked 9 October 2026
This is a curated introduction, not a comprehensive bibliography. Inclusion means a source is worth examining; it does not imply agreement, independent corroboration, or endorsement by its authors.