From Self-Report to Judgment: Instability and Interpretive Disagreement in Language Models' Consciousness Claims

This exploratory study examines the stability and interpretation of consciousness-related self-reports generated by large language models. Using 840 elicited responses across three model endpoints, 40 items, and seven experimental conditions, the study investigates how model responses vary under changes in prompting and framing. A second analysis layer uses an LLM judge to classify a subset of responses, followed by a nonblind author audit of those classifications. The findings highlight limitations in treating model self-reports or automated classifications as straightforward evidence about AI consciousness. The study does not attempt to determine whether any evaluated model is conscious; instead, it examines the reliability and interpretive challenges of the methods used to assess such claims.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-17
DOI
https://doi.org/10.5281/zenodo.24296564
Primary Topic
Philosophy and Theoretical Science
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

From Self-Report to Judgment: Instability and Interpretive Disagreement in Language Models' Consciousness Claims

Eswar Kumar Jonnadula
3 links
Zenodo (CERN European Organization for Nuclear Research)
Philosophy and Theoretical Science
preprint

From Self-Report to Judgment: Instability and Interpretive Disagreement in Language Models' Consciousness Claims

Eswar Kumar Jonnadula
preprint en

Abstract

This exploratory study examines the stability and interpretation of consciousness-related self-reports generated by large language models. Using 840 elicited responses across three model endpoints, 40 items, and seven experimental conditions, the study investigates how model responses vary under changes in prompting and framing. A second analysis layer uses an LLM judge to classify a subset of responses, followed by a nonblind author audit of those classifications. The findings highlight limitations in treating model self-reports or automated classifications as straightforward evidence about AI consciousness. The study does not attempt to determine whether any evaluated model is conscious; instead, it examines the reliability and interpretive challenges of the methods used to assess such claims.

Zenodo (CERN European Organization for Nuclear Research)
Philosophy and Theoretical Science
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon