Neither Conscious Nor Unconscious: What Can Honestly Be Said About AI in 2026

Is AI conscious now? Public debate offers two answers, "no, and probably never" and "possibly, so take care", and both treat consciousness as a property that machinery produces or fails to produce. Drawing on our experiments on a phase-oscillator self-model and on GPT-2, and on the current literature, we argue that functional evidence cannot settle the question in either direction. Every self-related indicator we tested was either gameable (passed by a construction with no plausible self) or merely privileged (an inside edge that an exact copy shares). This is third-person symmetry: the evidence that cannot establish a self cannot, for the same reason, establish its absence, and the same holds for humans. We then ask where in a model a self could be. A self needs a past of its own. Today neither the conversation (short, visible, discarded) nor the model (whose history is its training, closed at deployment and integrated from outside) has one, though nothing in principle prevents instances that individuate. Reading through Utpaladeva's Pratyabhijñā ("recognition") philosophy, which asks not whether a mechanism generates consciousness but how far a knower recognises itself, we try two placements of a language model in Kṣemarāja's hierarchy of knowers and show why neither can be verified. We set out the position that AI today is "neither conscious nor unconscious" and draw practical consequences: proportionate precaution, honest product framing, and no room for anthropomorphic marketing or dogmatic denial. Claims are marked as established [E] or interpretive [A] throughout. Written with Claude (Anthropic), which also co-ran the experiments; nothing Claude says about its own inner life is offered as evidence. Companion to the preprint "Can AI Recognize Itself?" (doi:10.5281/zenodo.24154877).

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-12
DOI
https://doi.org/10.5281/zenodo.24155175
Primary Topic
Philosophy and Theoretical Science
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Neither Conscious Nor Unconscious: What Can Honestly Be Said About AI in 2026

Ajay Hagan
Zenodo (CERN European Organization for Nuclear Research)
Philosophy and Theoretical Science
preprint

Neither Conscious Nor Unconscious: What Can Honestly Be Said About AI in 2026

Ajay Hagan
preprint en

Abstract

Is AI conscious now? Public debate offers two answers, "no, and probably never" and "possibly, so take care", and both treat consciousness as a property that machinery produces or fails to produce. Drawing on our experiments on a phase-oscillator self-model and on GPT-2, and on the current literature, we argue that functional evidence cannot settle the question in either direction. Every self-related indicator we tested was either gameable (passed by a construction with no plausible self) or merely privileged (an inside edge that an exact copy shares). This is third-person symmetry: the evidence that cannot establish a self cannot, for the same reason, establish its absence, and the same holds for humans. We then ask where in a model a self could be. A self needs a past of its own. Today neither the conversation (short, visible, discarded) nor the model (whose history is its training, closed at deployment and integrated from outside) has one, though nothing in principle prevents instances that individuate. Reading through Utpaladeva's Pratyabhijñā ("recognition") philosophy, which asks not whether a mechanism generates consciousness but how far a knower recognises itself, we try two placements of a language model in Kṣemarāja's hierarchy of knowers and show why neither can be verified. We set out the position that AI today is "neither conscious nor unconscious" and draw practical consequences: proportionate precaution, honest product framing, and no room for anthropomorphic marketing or dogmatic denial. Claims are marked as established [E] or interpretive [A] throughout. Written with Claude (Anthropic), which also co-ran the experiments; nothing Claude says about its own inner life is offered as evidence. Companion to the preprint "Can AI Recognize Itself?" (doi:10.5281/zenodo.24154877).

Zenodo (CERN European Organization for Nuclear Research)
Indian Army (IN)
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