Scope: Targeted public research available by each assessment window. This is not a systematic review or proof that no evidence exists.
Search record
Run on 25 September 2026, then filtered by the historical cutoffs: 1 April and 31 May 2025. Search results are a retrospective sample, not a complete index of everything available at those dates.
Subjective experience / qualiasite:arxiv.org 2023 indicators of consciousness in AI systems Butlin subjective experience report
site:nature.com 2025 AI consciousness empirical test subjective experience evidence
site:arxiv.org 2024 artificial intelligence qualia empirical test
Screen: primary reports within the cutoff; distinguish theory indicators and behavioural proxies from a test of experience.
Non-dual awarenessnondual awareness artificial intelligence 2024 2025 empirical test
AI nondual consciousness ego dissolution experimental model 2024
non dual awareness AI system empirical study before May 2025
Screen: primary AI-system tests within the cutoff; distinguish generated language from an experienced state.
Selected primary works
Butlin and colleagues, 2023 ↗Indicator-based assessment under scientific theories. The authors judged the systems they assessed as not conscious. This is not a direct readout of qualia.
Bojić and colleagues, 2024 ↗GPT-3 cognitive, emotional and self-assessment tests. The authors frame their results as signs to investigate, not discovery of machine consciousness.
Saqr, March 2025 ↗Glyph generated language with ego-dissolving features. The study evaluates symbolic output, not whether a system experienced a non-dual state.
Next search protocol
Status: defined, not yet executed. The April and May claims need separate searches using each historical cutoff. The earliest public version date determines which window can cite a study.
Corpus- arXiv AI, machine learning and neuroscience abstracts
- PubMed-indexed empirical studies
- ACL Anthology and major AI conference proceedings
- Lab technical reports and system cards with methods and model identifiers
Screen and extractInclude primary AI-system studies that name the tested model, protocol, observable result and claimed link to experience. Log human-only work, roleplay, unvalidated self-report, later publications and duplicates as exclusions.
For each eligible study, record model version, tools and permissions, intervention, outcome, author interpretation, replication and counterevidence.
The search can report what these sources contain under stated terms. It cannot establish that every AI system lacks experience.
What the scan establishes
The selected works cover indicators, behavioural proxies and simulated language patterns. None supplies a validated direct test of AI qualia or non-dual experience. This limited sample cannot establish a field-wide absence of evidence.
Next: Define the full corpus, inclusion rules, model coverage and operational test before making an exhaustive absence claim. The original red assessments remain visible as authored judgements, with this limitation attached.