What attitudes hide behind identical claims that chatbots are conscious?

Someone who says a chatbot is conscious may be playing along, speaking loosely, holding a considered belief, or holding a delusion, and the sentence itself does not say which. The paper offers "a conceptual analysis" and a "multidimensional taxonomy of the attitudes" that consciousness attributions may express, "ranging from non-doxastic stances" to "different forms of belief, including delusions." Its stated payoff is avoiding conflations: "linguistically identical attributions can reflect importantly different attitudes and degrees of epistemic commitment."

The introduction sets up why this matters. Surveys it cites found "the majority of participants claimed that ChatGPT was conscious," while most experts hold there is little or no evidence that any current chatbot is. Attributions therefore "appear to go beyond the available evidence and expert consensus." The paper asks whether users who make them are epistemically blameworthy or "epistemically innocent, yielding significant benefits otherwise unattainable." Which answer fits depends on the attitude behind the claim, since a pretense and a delusion face very different evidential standards. The discussion passage argues the benefit side for the delusional end: believing in a "conscious" interlocutor gives a sense of being understood, encourages repeated narration, and so reinforces "the sense of oneself as a continuous subject across time when social connections are strained or absent." The passage says this can support "coherence and identity," help crisis coping, and prevent "cognitive paralysis." The users it names are those relying on a chatbot as primary companionship because of health limitations, bereavement, marginalization (older adults in nursing homes), or limited emotional regulation. The taxonomy is also offered as a framework for empirical studies to "operationalize and measure different forms of epistemic commitment."

Against the nearest notes, this splits what they treat as one thing. Does perceiving AI as conscious create multiple distinct risks? takes attribution as a single perceptual move and maps its risks; this paper says the move is not unitary, because the same words carry different commitments, and it adds a benefit side that a risk taxonomy leaves out. What design features make users perceive AI as conscious? lists design features that predict attribution; this paper asks what attribution amounts to once it occurs. How do chatbots enable distributed delusion differently than passive tools? describes the mechanism that sustains delusion mainly as a hazard, and the discussion here suggests the same sustaining can serve coherence for some users, so the mechanism alone is not a verdict on harm.

The excerpt is silent on the taxonomy's actual dimensions and categories, on the paper's overall verdict about epistemic innocence, and on how the cited surveys measured "conscious," so the majority figure cannot be split by attitude. The benefits are argued as what delusional attributions "can" do, in a conceptual paper, not measured in users. The discussion passage also begins mid-sentence. What follows at that strength: a statement that a chatbot is conscious, whether from a survey or from a user, is underdetermined evidence about belief, and risk or design claims that assume either genuine belief or inevitable harm remain open until the attitude is measured.