CoChat
Sign Up
Cited by 0Philosophy and Theoretical ScienceEpistemology, Ethics, and MetaphysicsCybernetics and Technology in Society

Are Attributions of Consciousness to AI Chatbots Epistemically Innocent?

Open access
Peters, Uwe·2026·Minds and Machines
DOI 10.1007/s11023-026-09795-8
Open this paper in CoChat — free accountOpen PDF

Abstract

Artificial intelligence (AI) chatbots (e.g., ChatGPT) can communicate in strikingly humanlike ways. This has prompted many chatbot users to attribute psychological properties, including consciousness, to these systems. However, there is little scientific evidence that current AI chatbots are conscious. How, then, should we understand people’s consciousness attributions to chatbots? Are they merely metaphorical claims, or literal expressions of genuine beliefs? If these attributions lack evidential support, are users epistemically blameworthy for making them, or might they be epistemically innocent, yielding significant benefits otherwise unattainable? This paper offers a conceptual analysis of consciousness attributions to AI chatbots and develops a multidimensional taxonomy of the attitudes they may express, ranging from non-doxastic stances (e.g., pretence) to different forms of belief, including delusions. This taxonomy helps avoid conflations by showing that linguistically identical attributions can reflect importantly different attitudes and degrees of epistemic commitment to the proposition that chatbots are conscious. The taxonomy also provides a framework for empirical studies to operationalize and measure different forms of epistemic commitment to AI consciousness. Using this taxonomy, I argue that although some consciousness attributions to chatbots are epistemically benign, and even some irrational ones may be epistemically innocent, many others render the attributor epistemically blameworthy.

AI snapshot

Written from the paper’s full text.

TL;DR

This paper offers a conceptual taxonomy of the attitudes that may underlie attributions of consciousness to AI chatbots, ranging from strategic pretence to bizarre delusion. It argues that some such attributions are epistemically benign, some irrational ones may be epistemically innocent, and many others render the attributor epistemically blameworthy.

Key findings

  • The paper distinguishes ten attitudes that can underlie linguistically identical consciousness attributions to chatbots: strategic pretence, conformist affirmation, allegiance affirmation, imaginative affirmation, acceptance, in-between belief, unbiased belief, biased belief, non-bizarre delusion, and bizarre delusion.
  • It argues that attributions based on conformist affirmation, imaginative affirmation, acceptance, and unbiased belief can be epistemically benign within the attributor’s cognition, though attributors may still be blameworthy for failures of transparency or role-sensitive communication.
  • Attributions based on strategic pretence or allegiance affirmation are described as epistemically pernicious, while biased beliefs, non-bizarre delusions, and bizarre delusions are epistemically irrational; in-between belief is treated as a borderline case.
  • Using Bortolotti’s “epistemic innocence” framework, the paper argues that some irrational consciousness attributions may be epistemically innocent for low-AI-literacy users and vulnerable users, but not for entertainment users or technically trained users, who are epistemically blameworthy.

Methods

  • Conceptual analysis and taxonomy construction, drawing on existing distinctions from political psychology, philosophy of mind, epistemology, and philosophy of psychiatry.
  • The taxonomy is organized along three dimensions: commitment to the proposition, doxastic status, and pathological status; it focuses primarily on explicit attitudes because empirical studies of AI consciousness attribution have relied exclusively on self-report.
  • The paper applies the taxonomy to normative evaluation, using Bortolotti’s two conditions for epistemic innocence: significant epistemic benefit and no available less irrational alternative.

Limitations

  • The paper states that consciousness attributions to AI chatbots are a relatively new phenomenon with little prior research, so systematization risks appearing ad hoc; it also states that the taxonomy focuses primarily on explicit attitudes and sets aside cases of random error, inattention, misunderstanding, or carelessness in surveys.
  • The paper states that it assumes a doxastic view of delusions and notes caution about medicalization; it also acknowledges that delusions exist on a spectrum and that clinical documentation of bizarre delusions about chatbots is scarce.
  • Evident from the text: the paper is conceptual and does not report new empirical data, operationalize or test the taxonomy, or provide systematic clinical evidence; examples include news reports and hypothetical cases.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1007/s11023-026-09795-8

Cites

89 of 89 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

25 of 65 cited sources checked · 7 supported · 0 flagged · accuracy not yet measured

  • Interpretationcitation

    Respondents may view consciousness positively, thus preferentially attributing it to entities, including AI, when uncertain to provide what they perceive as a socially agreeable, expected answer.

    “Respondents may also view consciousness positively (Moncoucy et al., 2025), thus preferentially attributing it to entities, including AI, when uncertain to provide what they perceive as a socially agreeable, expected answer (Steenkamp et al., 2010).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Moncoucy et al., 2025 source

      The Introduction (page unavailable) describes phenomenal experience as intrinsically valuable to the conscious subject, which supports a positive valuation of conscious experience. It does not examine whether respondents view consciousness itself positively. The claims about attributing consciousness to AI and giving a socially agreeable answer are attached to a different citation in the citing sentence. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
    • cites Steenkamp et al., 2010 source
  • Interpretationcitation

    When chatbot users ascribe feelings, understanding, or consciousness to a chatbot, they may treat the relevant proposition as true in their imagination-a domain of cognition-to make interactions with the system more enjoyable.

    “Moving from merely practically committal consciousness attributions to cases when such attributions rest on an at least partially epistemically committal attitude, when chatbot users ascribe feelings, understanding, or consciousness to a chatbot (e.g., Replika), they may treat the relevant proposition as true in their imagination -a domain of cognition -to make interactions with the system more en”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Coghlan, 2024 source

      Section 5, “Questioning the A-Claim” (page unavailable), allows that imaginative suspension of disbelief may apply to some machines, but says it is less apt for many cases. Section 5.4 discusses imagining a robot’s feelings without believing them, while Section 2 mentions Replika; neither passage says chatbot users imagine such propositions to make interactions more enjoyable. (LLM, conf=0.78)

      Experimental — accuracy not yet measured
  • Premisecitation

    Empirical studies of AI consciousness attribution have relied exclusively on self-report, so they have so far investigated only explicit attitudes.

    “Since empirical studies of AI consciousness attribution have relied exclusively on selfreport, however, they have so far investigated only explicit attitudes (Colombatto & Fleming, 2024, p. 4).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Colombatto & Fleming, 2024 source

      The Introduction and Discussion describe this study’s consciousness attributions as ratings collected through self-report, and the Discussion questions whether self-reports capture intuitions that indirect behavioural markers might reveal. The paper does not establish that empirical studies have relied exclusively on self-report or investigated only explicit attitudes; it also notes other studies using different measures. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
  • Premisecitation

    Recent surveys found that the majority of participants claimed that ChatGPT was conscious, i.e., had subjective experience.

    “In fact, recent surveys found that the majority of participants claimed that ChatGPT was conscious, i.e., had subjective experience (Colombatto & Fleming, 2024;Kang et al., 2025).”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites Colombatto & Fleming, 2024 source

      The Results report that 67% of participants attributed some possibility of phenomenal consciousness to ChatGPT. That supports the majority figure, but not the stronger claim that a majority said it was conscious: the Discussion notes that, if responses are interpreted as credence, only 23% rated human-like experience more likely than not. (LLM, conf=0.90)

      Experimental — accuracy not yet measured
    • cites Kang et al., 2025 source

      The Introduction (p. 1) reports that 67% of respondents in a prior survey believed ChatGPT might be conscious, which supports a majority expressing belief in the possibility. It does not say a majority claimed ChatGPT was conscious. This paper’s own survey instead asked about Claude 3 Opus (Methods, p. 2), so its results cannot establish the claim about ChatGPT. (LLM, conf=0.88)

      Experimental — accuracy not yet measured
  • Premisecitation

    AI companions helped people cope with grief by taking the role of a listener simulating deceased, romantic partners or friends.

    “helped people cope with grief by taking the role of a listener simulating deceased, romantic partners or friends (Xygkou et al., 2023)”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Xygkou et al., 2023 source
  • Classified as probably not a substantial claim. 35
  • May combine two claims. 35
  • The quote only partly states this claim. 1

  • Classified as probably not a substantial claim.

    Premisecitation

    Sentience is commonly thought to be sufficient for moral status.

    “This matters ethically, as sentience is commonly thought to be sufficient for moral status (for discussion, see Shepherd, 2024).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Shepherd, 2024 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationcitation

    These attitudes may be explicit or implicit.

    “Finally, these attitudes may be explicit (consciously accessible and measurable by self-report) or implicit (largely unconscious and testable through indirect measures (e.g., Implicit Association tasks (IAT), Greenwald et al., 2022)).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Greenwald et al., 2022 source

      The Introduction’s “Definition of ‘implicit’” section contrasts implicit and explicit attitudes and recommends distinguishing them by indirect versus direct measurement. The “Measurement characteristics of IAT measures” section identifies the IAT as an indirect measure. But the paper cautions that an indirect measure is not a pure indicator of unconscious processes, so it does not support describing implicit attitudes as largely unconscious. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    In media discourse, AI chatbots are frequently described using psychological language and conversational norms in human-AI interaction invite treating chatbots as social agents.

    “Since in media discourse, AI chatbots are frequently described using psychological language and conversational norms in human-AI interaction invite treating chatbots as social agents (Shanahan, 2024), in surveys, this may make consciousness attributions to chatbots seem like the expected or 'correct' response.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Shanahan, 2024 source

      The Conclusion: Why This Matters says that words such as “believes” and “thinks” encourage anthropomorphism, including when speaking to the media, and that interacting with a conversational agent can create the illusion of a thinking creature. It does not establish that media descriptions frequently use such language or specifically attribute the effect to conversational norms. The available passage has no reliable page number. (LLM, conf=0.82)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Morrin et al. (2025) list several recent incidents in which chatbot users felt they were interacting with a conscious god-like AI, and this severely disrupted their life functioning, sometimes leading to suicide.

    “Morrin et al. (2025) list several recent incidents in which chatbot users felt they were interacting with a conscious”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Morrin et al., 2025 source

      The case summaries in “Future directions” describe users treating AI as sentient or a higher power, alongside serious disruptions to relationships, work, and mental health. One case reports an attempted hanging, but the passages found do not report a completed suicide; another user was shot and killed by police. The claim’s reference to suicide therefore overstates the reported outcomes. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Interpretationcitation

    Such a belief needn't be epistemically problematic merely because it is false; it may instead reflect a reasonable attempt to explain the available evidence, especially since consciousness is ordinarily inferred from behavioral and psychological cues, not directly observed.

    “Such a belief needn't be epistemically problematic merely because it is false.It may instead reflect a reasonable attempt to explain the available evidence, especially since consciousness is ordinarily inferred from behavioral and psychological cues, not directly observed (Seth, 2025).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Seth, 2025 source

      Split into 2 atoms — 0 confirmed, 2 partially supported, 0 disputed, 0 misattributed, 0 unverifiable. Expand the claim to see per-atom verdicts.

      Experimental — accuracy not yet measured

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Delusions exist on a spectrum and may be monothematic or polythematic.

    “delusions exist on a spectrum (Kingdon & Turkington, 1994) and may be monothematic (concerning a single theme) or polythematic (concerning various themes, e.g., feeling persecuted, on a divine mission, etc.) (Coltheart, 2013).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Kingdon & Turkington, 1994 source
    • cites Coltheart, 2013 source

  • May combine two claims.

    Premisecitation

    AI companions reduced suicidal ideation and increased users' capacity to handle stress and empathize.

    “reduced suicidal ideation, increased users' capacity to handle stress and empathize (Maples et al., 2024)”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Maples et al., 2024 source

      The Results (Outcomes; page unavailable) quote users saying Replika helped them handle relationship stress and become more empathetic. The paper also reports that 30 students credited Replika with preventing a suicide attempt, but it does not establish a measured reduction in suicidal ideation. The Discussion (Overlapping beliefs and outcomes; page unavailable) cautions that efficacy cannot yet be concluded from these survey reports. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    AI companions encouraged users to greater openness in their human relationships and made users more comfortable with interacting with people.

    “encouraged them to greater openness in their human relationships, and made users more comfortable with interacting with people (Skjuve et al., 2021;Xie & Pentina, 2022).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites Skjuve et al., 2021 source

      Section 6.3.2.1 reports that a few Replika users became more comfortable talking to other people and reconnected with people in their lives. But the openness described in section 6.2.2 is self-disclosure to Replika, not greater openness in human relationships; the paper does not establish that part of the claim. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites Xie & Pentina, 2022 source

  • May combine two claims.

    Interpretationcitation

    Even if delusion-based consciousness attributions to chatbots are sometimes epistemically innocent, they can also reinforce epistemic instability by blurring reality boundaries and disrupting self-regulation.

    “However, even if delusion-based consciousness attributions to chatbots are sometimes epistemically innocent, they can also reinforce epistemic instability, understood here as a reduced ability to distinguish between accurate and inaccurate representations of reality and to respond appropriately to counterevidence, by blurring reality boundaries and disrupting selfregulation (Morrin et al., 2025).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Morrin et al., 2025 source

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The taxonomy helps avoid conflations by showing that linguistically identical attributions can reflect importantly different attitudes and degrees of epistemic commitment to the proposition that chatbots are conscious.

    “This taxonomy helps avoid conflations by showing that linguistically identical attributions can reflect importantly different attitudes and degrees of epistemic commitment to the proposition that chatbots are conscious.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    This combination of suggestive indirect consciousness attribution and denial is characteristic of a stance called strategic pretence.

    “This combination of suggestive indirect consciousness attribution and denial is characteristic of a stance I call strategic pretence -the attributor commits to choosing and (indirectly or directly) expressing the proposition that some AI system is conscious to attract attention and users, without treating it as true in cognition and while being ready to swiftly retract the claim.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given

    The taxonomy provides a framework for empirical studies to operationalize and measure different forms of epistemic commitment to AI consciousness.

    “The taxonomy also provides a framework for empirical studies to operationalize and measure different forms of epistemic commitment to AI consciousness.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationreasoning

    Conformist affirmation's commitment is still not epistemic-the attributor needn't treat p as true in inferences-but remains practical and needn't extend to behavior outside the eliciting context.

    “However, this commitment is still not epistemic -the attributor needn't treat p as true in inferences -but remains practical and needn't extend to behavior outside the eliciting context (e.g., surveys).”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationcitation

    Since the proposition isn't yet judged to be true, the epistemic commitment involved in acceptance doesn't yet amount to belief-one may accept that chatbots are conscious without taking (judging) it to be true.

    “However, since the proposition isn't yet judged to be true, the epistemic commitment involved doesn't yet amount to belief -one may accept that chatbots are conscious without taking (judging) it to be true.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given

    At least ten attitudes may underlie consciousness attributions to chatbots.

    “Against this background, at least the following ten attitudes may underlie consciousness attributions to chatbots:”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    Cohen (1992) distinguishes between acceptance and belief, arguing that to accept that p is to adopt a policy of treating p as a premise in some or all contexts, including practical inferences and planning, without judging p to be true.

    “For instance, Cohen (1992) distinguishes between acceptance and belief, arguing that to accept that p is to adopt a policy of treating p as a premise in some or all contexts, including practical inferences and planning, without judging p to be true.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Cohen, 1992

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The later categories increasingly differentiate doxastic states by their degree of inferential integration and resistance to counterevidence, culminating in pathological fixation (bizarre delusions).

    “Thus, the later categories don't simply add more items to the list but increasingly differentiate doxastic states by their degree of inferential integration (e.g., from in-between to fully integrated belief), and resistance to counterevidence (low to high), culminating in pathological fixation (bizarre delusions).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationcitation

    Strategic pretence allows exaggerating Replika's capabilities to lead users to perceive interactions with the chatbot as more meaningful, strengthening attachment and engagement, increasing the company's revenues.

    “While the last feature indicates the weak commitment involved, strategic pretence allows exaggerating (e.g.) Replika's capabilities to lead users to perceive interactions with the chatbot as more meaningful (Guingrich & Graziano, 2024), strengthening attachment and engagement, increasing the company's revenues.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Guingrich & Graziano, 2024

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    Some categories mark different attitude-types (e.g., strategic pretence, imaginative affirmation, acceptance, belief), while others introduce refinements within a single type, especially belief.

    “Some categories mark different attitude-types (e.g., strategic pretence, imaginative affirmation, acceptance, belief).Others introduce refinements within a single type, especially belief.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    Allegiance affirmation involves a commitment to affirm and maintain the expression of p across similar contexts, no matter whether it is the overall socially expected answer.

    “The attitude therefore involves a commitment to affirm and maintain the expression of p across similar contexts, no matter whether it is the overall (e.g., partisan independent) socially expected answer.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Explicit and implicit attitudes can dissociate: a user may sincerely explicitly deny that AI chatbots are conscious while nevertheless exhibiting nonverbal behavior suggesting an opposing implicit attitude.

    “The two can dissociate (Ajzen & Dasgupta, 2015): A user may sincerely explicitly deny that AI chatbots are conscious while nevertheless exhibiting nonverbal behavior (e.g., feeling guilt about 'hurting' a chatbot after insulting it, or reacting faster in IATs linking chatbots with consciousness) suggesting an opposing implicit attitude.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Ajzen & Dasgupta, 2015

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Premisecitation

    Consciousness is distinct from sentience, i.e., positively or negatively valanced experience.

    “The reason is that even though consciousness is distinct from sentience, i.e., positively or negatively valanced experience (pleasure, pain, etc.),foot_1 the two are often conflated (Birch, 2024).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Birch, 2024 source

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationcitation

    Acceptance is the first attitude in which the proposition functions as a premise in reasoning and for action guidance.

    “Specifically, acceptance is the first attitude in which the proposition functions as a premise in reasoning and for action guidance.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationcitation

    Since accepting that p involves treating p as true in one's judgment and decision-making, including action planning, it involves a stronger form of epistemic commitment than the previous attitudes.

    “Since accepting that p involves treating p as true in one's judgment and decision-making, including action planning (Stoter, 2023), it involves a stronger form of epistemic commitment than the previous attitudes.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Stoter, 2023

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The resulting taxonomy is hierarchical, not flat.

    “Relatedly, the resulting taxonomy is hierarchical, not flat.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationnone given

    The list's lower to greater numbering isn't a one-dimensional scale but is meant to broadly track a multidimensional progression from minimal to strong forms of practical and epistemic commitment to AI consciousness, from non-doxastic to doxastic states, and from non-pathological to pathological attitudes.

    “But to clarify, the list's lower to greater numbering isn't a onedimensional scale but is meant to broadly track a multidimensional progression from (a) minimal to strong forms of practical and epistemic commitment to AI consciousness, (b) from non-doxastic to doxastic states, and (c) from non-pathological to pathological attitudes.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationcitation

    This attitude may still fall short of full belief, understood as a relatively stable, normally evidence-responsive, and involuntary state or disposition to take p to be true across contexts and judgment and decision-making instances.

    “Yet, this attitude may still fall short of full belief, understood as a relatively stable, normallyfoot_6 evidenceresponsive, and involuntary state or disposition to take p to be true across contexts and judgment and decision-making instances (Nottelmann, 2013).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Nottelmann, 2013 source

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    Unbiased belief in AI consciousness would indicate the strongest epistemic commitment in the sense of rational endorsement.

    “Such unbiased belief in AI consciousness would indicate the strongest epistemic commitment in the sense of rational endorsement.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    While these attributions are epistemically irrational, this by itself is insufficient to hold that people should be corrected for them, as the pragmatic benefits may justify the epistemic costs.

    “while these attributions are epistemically irrational, this by itself is insufficient to hold that people should be corrected for them, as the pragmatic benefits may justify the epistemic costs.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Bortolotti argues that delusional beliefs can be epistemically beneficial when they explain a puzzling experience, restore cognitive performance, support curiosity and self-efficacy, manage negative emotions, mitigate depression, or support agency.

    “She argues, for instance, that delusional beliefs can be epistemically beneficial when they (1) are adopted to explain a puzzling experience that might compromise the agent's capacity to interact with their environment, (2) help restore aspects of cognitive performance by decreasing anxiety, (3) support an attitude of curiosity and self-efficacy that is more conducive to acquiring new information”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Bortolotti, 2020 source

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Premisecitation

    Value judgments are required to set evidential thresholds, and if the agent is transparent about their judgment, consistently applies their threshold as new data emerges, and stays open to revising their stance, the resulting value-laden belief is unproblematic.

    “Many philosophers of science hold that these value judgments are required to set evidential thresholds and if the agent is transparent about their judgment, consistently applies their threshold as new data emerges, and stays open to revising their stance, the resulting value-laden belief is unproblematic (e.g., Longino, 2002;Elliot & Richards, 2017).”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites Longino, 2002 source
    • cites Elliot & Richards, 2017 source

  • May combine two claims.

    Premisecitation

    Whether current chatbots are conscious is ultimately an empirical question, and empirical testing is probabilistic, involving inductive risk, i.e., the possibility of false positives or false negatives given the available evidence.

    “whether current chatbots are conscious is ultimately an empirical question, and empirical testing is probabilistic, involving”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Elliot & Richards, 2017 source

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based). May combine two claims.

    Interpretationreasoning

    Consciousness attributions to chatbots can count as unbiased if formed and maintained through an evidence-sensitive process, not wishful thinking or selective attention, and the evidential threshold is applied in a stable, nonad hoc way across similar cases.

    “they can count as unbiased provided they are formed and maintained through an evidence-sensitive process, not wishful thinking or selective attention, and the evidential threshold is applied in a stable, nonad hoc way across similar cases.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    According to Bortolotti's framework, an epistemically irrational belief b is epistemically innocent iff the adoption, maintenance, or reporting of b delivers some significant epistemic benefit to A, and the adoption, maintenance, or reporting of a less epistemically irrational belief than b is not available or would fail to deliver the same significant epistemic benefit to A.

    “an epistemically irrational belief b is”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Bortolotti, 2020 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Interpretationreasoning

    Most benefits that Bortolotti lists are instrumentally epistemic because they promote or safeguard a person's ability to gather, evaluate, or retain knowledge.

    “most benefits that Bortolotti lists are instrumentally epistemic.They promote or safeguard a person's ability to gather, evaluate, or retain knowledge, thus counting as epistemic, on her view, because they ultimately improve the agent's cognitive performance in ways that matter to epistemology”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationnone given

    The pragmatic benefits of epistemically irrational consciousness attributions may in some cases outweigh the costs.

    “the pragmatic benefits of epistemically irrational consciousness attributions may in some cases outweigh the costs.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Strategic pretence, allegiance affirmation, and conformist affirmation are not doxastic attitudes; their epistemic defect lies in insincerity, manipulation, and corruption of epistemic environments rather than evidence-recalcitrance in cognition.

    “Strategic pretence, allegiance affirmation, and conformist affirmation, by contrast, aren't doxastic attitudes to begin with -the attributor doesn't genuinely believe anything false about chatbots but only produces assertions about them with practical commitments.Their epistemic defect therefore doesn't lie in evidence-recalcitrance in cognition but in (e.g., persistent) insincerity, manipulation,”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Consciousness attributions grounded in strategic pretence or allegiance affirmation can be viewed as epistemically pernicious, and attributors are epistemically blameworthy for them.

    “consciousness attributions of this kind, especially when indirectly advanced by developers or AI companies, can be viewed as epistemically pernicious.Given their negative social epistemic effects, attributors are epistemically blameworthy for them.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning

    On the broader measure of epistemic commitment, more evidence-resistant belief states involve a stronger form of commitment because they retain p as a premise even in the face of compelling counterevidence.

    “On that measure, more evidence-resistant belief states would involve a stronger form of such commitment (despite being less rational), because they involve retaining p as a premise even in the face of compelling counterevidence.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    In biased belief cases, the difference from unbiased belief does not lie in the threshold for belief but in how its maintenance is governed; unbiased belief is constrained by evidence whereas biased belief is stabilized against it, indicating stronger epistemic commitment.

    “in such biased belief cases, the difference doesn't lie in the threshold for belief but in how its maintenance is governed.Unbiased belief is constrained by evidence whereas biased belief is stabilized against it, indicating stronger epistemic commitment.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable). May combine two claims.

    Interpretationreasoning

    Since such beliefs are maintained despite even stronger counterevidence than in non-bizarre delusions, they reflect an even more extreme form of epistemic commitment, warranting their inclusion as the most severe pathological belief in AI consciousness within the taxonomy.

    “since such beliefs are maintained despite even stronger counterevidence than in non-bizarre delusions (due to their more extreme, less tenable content), they reflect an even more extreme form of epistemic commitment, warranting their inclusion as the most severe pathological belief in AI consciousness within the taxonomy.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Guingrich and Graziano (2025) found that the more subjects rated bots as conscious, agentic, experiential, and humanlike, the more positive their perceptions of the technology became, which positively affected users' relationships with family and friends, improving their mood, self-esteem, and social skills.

    “When studying people's interactions with Replika, Guingrich and Graziano (2025) found that the”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Guingrich & Graziano, 2025 source

  • Classified as probably not a substantial claim.

    Premisecitation

    Bortolotti distinguishes three kinds of unavailability: strict unavailability, motivational unavailability, and explanatory unavailability.

    “She distinguishes between three kinds of”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Bortolotti, 2020 source

  • Classified as probably not a substantial claim.

    Interpretationreasoning

    The taxonomy allows us to tease apart epistemically benign, epistemically pernicious, and epistemically irrational consciousness attributions.

    “I argue that it allows us to tease apart what may be called (1) epistemically benign, (2) epistemically pernicious, and (3) epistemically irrational consciousness attributions.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation

    In 2023, a Belgian man in his 30s died by suicide after six weeks of intense conversations with a chatbot named Eliza; the dialogue culminated in the man offering to sacrifice himself if Eliza would promise to save humanity and the planet from a climate catastrophe through AI.

    “in 2023, a Belgian man in his 30s died by suicide after six weeks of intense conversations with a chatbot named”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Attilah, 2023

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable), S4 (based).

    Interpretationreasoning

    A chatbot user's, developer's, or researcher's inductive risk threshold may make the proposition that some contemporary chatbot is conscious sufficiently credible to warrant belief.

    “chatbot users', developers', or researchers' inductive risk threshold may be such that, for them, the current evidence makes the proposition that some contemporary chatbot is conscious sufficiently credible to warrant (not just acceptance but) belief”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    If the belief is revised when new evidence is presented, then it can count as unbiased.

    “If the belief is revised when new evidence is presented, then it can count as unbiased.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Many motivationally biased beliefs depend on an illusion of objectivity: people typically arrive at and sustain desired conclusions only when they can justify them to themselves, which can prevent the formation of wildly implausible motivated beliefs about chatbots.

    “many motivationally biased beliefs aren't yet fully insulated from counterevidence but depend on what has been called an”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Kunda, 1990 source
    • cites Epley & Gilovich, 2016 source

  • May combine two claims.

    Interpretationreasoning

    The Belgian man and the teenager arguably believed the chatbot could understand their offer, make promises, and act accordingly, which is difficult to explain unless at least an implicit belief in chatbot consciousness was present.

    “both arguably believed the chatbot could understand their offer, make promises, and act accordingly.This is difficult to explain unless at least an implicit belief in chatbot consciousness was present.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Delusions in general have serious epistemic and psychological costs, and these costs may vastly outweigh their epistemic benefits.

    “Delusions, in general, have serious epistemic and psychological costs, and these costs may vastly outweigh their epistemic benefits (Bortolotti, 2020, chapters 4-5).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Bortolotti, 2020 source

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    The taxonomy provides empirical researchers with conceptual resources for operationalizing different forms of epistemic commitment to the proposition that AI chatbots are conscious and for identifying potential confounders in the study of AI consciousness attribution.

    “It also provides empirical researchers with conceptual resources for operationalizing different forms of epistemic commitment to the proposition that AI chatbots are conscious and for identifying potential confounders in the study of AI consciousness attribution.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Premisecitation

    If an epistemically irrational belief co-occurs with any of these kinds of unavailability of an alternative, less irrational state, then it counts as epistemically innocent on Bortolotti's account.

    “If an epistemically irrational belief co-occurs with any of these kinds of unavailability of an alternative, less irrational state then it counts as epistemically innocent on Bortolotti's account”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Bortolotti, 2020 source
    • The authors state this as an assumption: Bortolotti's account of epistemic innocence is tenable.

  • May combine two claims.

    Interpretationreasoning

    The more systems are engineered to appear conscious, the more users will satisfy the unavailability condition, and the greater the developers' responsibility for facilitating such attributions.

    “The more systems are engineered to appear conscious, the more users will satisfy the unavailability condition, and the greater the developers' responsibility for facilitating such attributions.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Delusional consciousness attributions to chatbots can support a sense of coherence and identity, bolstering crisis coping and preventing cognitive paralysis.

    “Hence, these delusional attributions can support a sense of coherence and identity, bolstering crisis coping and preventing cognitive paralysis.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    The concept of epistemic innocence helps us discern, normatively flag, and reduce the spread of epistemically problematic consciousness attributions to chatbots.

    “Furthermore, by providing a tool for identifying epistemically irrational consciousness attributions that are not epistemically innocent, the concept of epistemic innocence also helps us discern, normatively flag, and reduce the spread of epistemically problematic attributions of this kind.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Technically trained users who insist on chatbot consciousness violate an epistemic obligation to proportion belief to evidence and lack an excuse, as alternative more rational cognitions are available and psychologically tolerable to them.

    “These users, too, would violate an epistemic obligation to proportion belief to evidence and lack an excuse, as alternative more rational cognitions are available and psychologically tolerable to them.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    For some low-AI-literacy users, the denial of chatbot consciousness may appear puzzling or unintelligible, undermining confidence in their own social-cognitive abilities and introducing uncertainty that hinders smooth interaction.

    “For such users, the denial of chatbot consciousness may appear puzzling or unintelligible, undermining confidence in their own social-cognitive abilities and introducing uncertainty that hinders smooth interaction.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationreasoning

    Some consciousness attributions are epistemically benign; among those that are epistemically irrational, some may nevertheless be epistemically innocent, whereas others are epistemically pernicious, rendering the attributor epistemically blameworthy.

    “Some consciousness attributions are epistemically benign; among those that are epistemically irrational, some may nevertheless be epistemically innocent, whereas others are epistemically pernicious, rendering the attributor epistemically blameworthy.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Such design choices can drastically reinforce users' illusions of chatbot consciousness, significantly reducing the availability of more accurate beliefs to them.

    “Such design choices can drastically reinforce users' illusions of chatbot consciousness, significantly reducing the availability of more accurate beliefs to them.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Interpretationreasoning

    Both conditions for epistemic innocence could then be met for some low-AI-literacy users.

    “Both conditions for epistemic innocence could then be met.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Bortolotti's account of epistemic innocence is tenable.
60 of 60

  • The paper relies on this without support.

    The potential variety of mental states expressed by consciousness attributions to AI chatbots has not been systematically investigated yet.

    “However, the potential variety of the mental states expressed by consciousness attributions to AI chatbots hasn't been systematically investigated yet.”

  • The paper relies on this without support.

    It is currently unclear whether, which, or when consciousness attributions to AI chatbots are epistemically rational or delusional.

    “It is currently unclear whether, which, or when consciousness attributions to AI chatbots are epistemically rational or delusional.”

  • The paper relies on this without support.

    It remains largely unexplored how we should respond to people producing consciousness attributions to AI chatbots.

    “Similarly, it remains largely unexplored how we should respond to people producing such attributions.”

  • The authors state this as an assumption.

    It is at least possible for some chatbot to be conscious.

    “Assuming that it is at least possible for some chatbot to be conscious”

  • The authors state this as an assumption.

    Bortolotti's account of epistemic innocence is tenable.

    “I shall assume that this account of epistemic innocence is tenable”

  • The paper relies on this without support.

    The notion of epistemic innocence has not been explored in the context of consciousness attributions to chatbots yet.

    “it hasn't been explored in the context of consciousness attributions to chatbots yet.”

  • The authors state this as an assumption.

    The doxastic view of delusions is correct: delusions are beliefs.

    “But the doxastic view I assume here is widely accepted.”

  • The paper relies on this without support.

    Current AI systems are not conscious.

    “experts widely agree that current AI systems aren't conscious.”

  • The paper relies on this without support.

    Many people attribute some degree of consciousness to AI chatbots.

    “While surveys suggest that many people attribute some degree of consciousness to AI chatbots”
9 of 9

Cites

Flagged first, then by citations

Research works printed but not linked: 6

  1. Allport, G. W. (1935). Attitudes. A handbook of social psychology (pp. 798–844). Clark University Press.
  2. Cohen, L. J. (1992). An essay on belief and acceptance. Clarendon Press.
  3. DSM-5. (2013). Diagnostic and statistical manual of mental disorders. Fifth edition. American Psychiatric Association Washington, DC. Arlington, VA: American Psychiatric Association.
  4. DSM-IV. (2000). Diagnostic and statistical manual of mental disorders. Fourth edition. American Psychiatric Association Washington, DC. Arlington, VA: American Psychiatric Association
  5. Kidd, I. J. (2017). Capital epistemic vices. Social Epistemology Review and Reply Collective, 6(8), 11–16.
  6. Sebo, J. (2025). The moral circle: Who matters, what matters, and why. Norton Short.

Authors
Peters, U.
Affiliations
Utrecht University (NL)
Corresponding author
Peters, Uwe
Venue
Minds and Machines
Publisher
Springer Science and Business Media LLC
Published
2026-08-18
Volume
36
Issue
3
ISSN
  • 0924-6495 (ISSN-L)
  • 1572-8641
DOI
10.1007/s11023-026-09795-8
OpenAlex
W7170174436
Semantic Scholar
816b720410368e3ba38c61e535a319427aecb7f0
Cited by
0
References
62
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
Locations
  • Minds and Machines (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion, cc-by)
  • PhilPapers (PhilPapers Foundation) (submittedVersion, cc-by)
  • Utrecht University Repository (Utrecht University) (submittedVersion, other-oa)
  • arXiv (Cornell University) (cc-by)
Topics
  • Social Sciences › Psychology › Experimental and Cognitive Psychology › Philosophy and Theoretical Science
  • Social Sciences › Arts and Humanities › Philosophy › Epistemology, Ethics, and Metaphysics
  • Social Sciences › Arts and Humanities › History and Philosophy of Science › Cybernetics and Technology in Society
Keywords
AI chatbots, epistemic commitment, delusions, pretense, belief, ChatGPT, anthropomorphism of AI, epistemic irrationality

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-13.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-13.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review