AI ETHICS

What Do We Know About AI Consciousness?

Linguistic self-reports and convincing behavior count as evidence, but alone they do not establish subjective experience.

Turquoise signal lines run across a bright, translucent surface before a dark, layered interior. The boundary between what is observable and the unknown interior remains open.
An abstract image of the difference between observable performance and a question of experience that cannot be directly accessed.

The inquiry question

What can language, behavior, self-reports, and technical architecture warrantedly support about possible subjective experience—and what statement remains responsible while a certain diagnosis is unavailable?

A fictional SME case

A small learning software provider tests a text-based practice assistant. In a demonstration, the system explains a pun and changes its answer after additional context is entered. The team can describe the performance, but does not know whether subjective experience follows from it. The case is entirely fictional and does not describe a particular product.

Keep four levels distinct

Observation

Which task, input, version, and response were actually recorded?

Interpretation

Which ability or inner state is inferred from it? Which counterhypothesis explains the same data?

Theory-guided evidence

Which theory of consciousness makes a feature relevant, and what technical evidence is still missing?

Observable points on a bright plane pass through transparent filters and a branching dark model; part of the space remains hidden in mist.
The illustration is a metaphor for thinking, not a measurement scale or an AI consciousness test.

Phenomenal experience must be distinguished from cognitive access. Intelligence, emotion, self-description, and responsibility are likewise not interchangeable terms. A lack of evidence establishes neither consciousness nor its absence.

Three models with distinct tasks

F · Questioning Model

A clarifies terms, aim, and assumptions. B opens up competing explanations. D marks uncertainty, missing data, and limits. C formulates a narrower, action-guiding question. The sequence A → B → D → C replaces a hasty yes/no diagnosis.

K · Consultation Model

Exploration identifies role and knowledge gaps. Reflection and Analysis set evidence alongside the opposing position. Decision and Recommendation establish a provisional rule. Feedback and Evaluation state when it will be reassessed.

R · Advisory Model

R is used only for a specific specialist question as an expert handoff, for example to review technical documentation or a measurement design. It structures the request and feedback. It does not replace K and does not measure consciousness.

45-minute course exercise

0–5Read the fictional case and record the system description
5–12Distinguish ability, self-report, architecture, and experience
12–22Use F-B to develop at least two competing explanations
22–32Use F-D to mark missing evidence and limits on transfer
32–40Use the evidence map to develop a limited statement and triggers for revision
40–45Peer review: examine the strongest counterhypothesis and the scope of the statement

Practical decision framework

ObservationDocument the task and test condition.
CounterhypothesisExamine at least one other explanation.
Limited statementState abilities; do not infer experience from a text output.
RevisionSpecify the responsible role, evidence, and review trigger.

Limit of the overview

The research sources in the article support different philosophical or empirical subquestions. Human studies are not presented as AI tests. F, K, and R organize questions and decisions; their descriptions do not establish methodological effectiveness. This overview is not a consciousness diagnosis, a validated audit standard, or legal advice.