The question contains several questions
A system may detect a signal, use it in a decision, describe that decision, or represent itself as the cause. These are distinguishable capacities. Whether any of them is accompanied by experience is a further question. Our public vocabulary separates task performance, access to information, self-description, and phenomenal experience.
This distinction prevents two shortcuts: treating impressive competence as proof of consciousness, or treating unfamiliar implementation as proof that experience is impossible. Neither shortcut substitutes for an account of what would count as evidence.
Several theories, not one settled blueprint
Butlin and colleagues surveyed recurrent processing, global workspace, higher-order, predictive-processing, and attention-schema approaches, deriving computational indicator properties for evaluating AI systems. Their 2023 report is a theory-guided assessment framework, not a definitive machine-consciousness test or a verdict on systems developed after that assessment. Butlin and colleagues · Consciousness in Artificial Intelligence (2023) ↗
In a 2025 adversarial collaboration, the Cogitate Consortium compared preregistered predictions of integrated information and global neuronal workspace theories in human visual experience. The results supported some predictions and challenged important claims of both. This does not establish a single winning theory or determine whether Luna is conscious. Cogitate Consortium · Adversarial testing of consciousness theories (2025) ↗
What Neurasoft can investigate
We can intervene on a mechanism, compare alternate explanations, trace an action to its consequences, and test whether a retained history changes later behavior. We can distinguish a fixed prompt from a persisted state, and a reported preference from a preference that actually affects choices.
Each finding has a local scope. A successful replay supports a reproducibility claim. A selective intervention may support a causal claim. A combination of such results could inform a broader assessment, but no attractive diagram or isolated score bridges that gap automatically.
What would make the evidence more serious?
Before an experiment, identify what rival explanations predict. Preserve the conditions and the original outcomes. Include controls that distinguish the mechanism from additional compute, extra context, or a more helpful prompt. Repeat the comparison in settings where the explanation risks being wrong.
We call the spark a research aspiration, not a substance we claim to have extracted. The work is to understand and build an organization that might support experience, while remaining accountable for what has actually been observed.
Public editorial synthesis · 16 September 2026. Internal source observations are dated separately; this is not a live operational report. See the reading room and publication standards for source classes.
The question is extraordinary. The evidence must be ordinary enough to examine.