CoChat
Sign Up
Cited by 46Epistemology, Ethics, and MetaphysicsEthics and Social Impacts of AIFeminist Epistemology and Gender Studies

Artificial Epistemic Authorities

Open access
Hauswald, Rico·2025·Social Epistemology
DOI 10.1080/02691728.2025.2449602
Open this paper in CoChat — free accountOpen PDF

Abstract

While AI systems are increasingly assuming roles traditionally occupied by human epistemic authorities (EAs), their epistemological status remains unclear. This paper aims to address this lacuna by assessing the potential for AI systems to be recognized as artificial epistemic authorities. In a first step, I examine the arguments against considering AI systems as EAs, in particular the established model of EAs as engaging in intentional belief transfer via testimony to laypeople – a process seemingly inapplicable to intentionless and beliefless AI. Despite this, AI systems exhibit striking epistemic parallels with human EAs, including epistemic asymmetry and opacity that give rise to comparable challenges for both laypeople and AI users. These challenges include the identification problem – how to recognize reliable EAs/AI systems – and the deference problem – determining the appropriate epistemic stance towards EAs/AI systems. Faced with this dilemma, I discuss three possible solutions: 1. reject the concept of artificial EAs, 2. accept that AI can possess beliefs and intentions and thus align with the standard model, or 3. develop an alternative model that encompasses artificial EAs. I argue that while each option has its benefits and costs, a particularly strong case can be made for option 3.

AI snapshot

Written from the paper’s full text.

TL;DR

The paper assesses whether current AI systems can be recognized as artificial epistemic authorities (AEAs), despite lacking beliefs, communicative intentions, and genuine testimonial interaction. It argues for modifying the standard model of epistemic authority so that suitable facts about an authority—including AI outputs—can serve as reliable truth indicators.

Key findings

  • AI systems show structural parallels with human epistemic authorities: they can provide epistemic goods, stand in epistemically imbalanced relationships, exhibit epistemic opacity, and generate similar challenges such as the identification problem and the deference problem.
  • The standard model of epistemic authority as intentional belief transfer via testimony seems inapplicable to current AI, which is “most likely” without beliefs and communicative intentions.
  • The paper identifies three responses to the dilemma: reject artificial EAs, accept that AI has beliefs and intentions, or develop an alternative model; it argues a particularly strong case can be made for the third option.
  • The proposed alternative model says that to defer to an EA is to treat suitable facts about that EA as strong indicators of truth, where those facts bear a reliable relation to truth values; AI outputs can qualify as such indicators.

Methods

  • Conceptual and philosophical analysis, not an empirical study.
  • Uses illustrative conversations with ChatGPT-4 about astronomy, Euclid’s proof of infinitely many primes, and a wisdom-related scenario to motivate claims about AI epistemic capacities.
  • Engages with social epistemology literature on testimony, epistemic authority, preemption, expertise, and AI opacity to build and defend the alternative model.

Limitations

  • The paper states that it cannot discuss arguments for and against preemption in detail and notes that it is not yet clear whether and how AI opacity can be sufficiently overcome.
  • The paper states that current AI systems “most likely” do not meet requirements for beliefs and intentions, while not ruling out future AI with consciousness or person-like features.
  • Evident from the text: the ChatGPT-4 conversations are anecdotal illustrations rather than systematic evidence, and the proposed model is normative/conceptual rather than empirically tested.

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

Integrity signals

Registry

Crossref · 10.1080/02691728.2025.2449602

Cites

53 of 54 references verified1 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 10 cited sources checked · 0 supported · 0 flagged

  • Findingreasoning

    Some AIs have much broader capabilities.

    “some AIs have much broader capabilities.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: The capabilities of AI systems are often extremely narrow.
  • Findingreasoning

    Users can receive a variety of epistemic goods from AI systems, including knowledge, understanding, and wise advice.

    “users can receive a variety of epistemic goods from them, including knowledge, understanding, and wise advice.”
    Quote in the paperSubstantial claim
  • Findingown data

    ChatGPT can provide propositional knowledge.

    “So it seems clearly possible to gain propositional knowledge from ChatGPT.”
    Quote in the paperSubstantial claim
  • Findingnone given

    ChatGPT can convey the truth of Euclid's proof in a maieutic way when explicitly asked.

    “For example, when explicitly asked, ChatGPT is able to convey the truth of Euclid's proof in a maieutic way, i.e. just by asking questions.”
    Quote in the paperNot a substantial claim
  • Interpretationreasoning

    In many paradigmatic cases, we defer to groups by treating facts about those groups other than their beliefs as strong indicators of the truth of particular propositions.

    “As a result, while in some cases we do defer to group beliefs, in many paradigmatic cases we defer to groups by treating facts about those groups -other than their beliefs, if they have any -as strong indicators of the truth of particular propositions.”
    Quote in the paperSubstantial claim
  • May combine two claims. 10
  • Classified as probably not a substantial claim. 17

  • May combine two claims.

    Findingreasoning

    ChatGPT can promote understanding both directly and indirectly.

    “And it can promote understanding not only in a direct way, but also in an indirect way”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    ChatGPT can explain concepts, adapt to the user's level of difficulty, and respond to follow-up questions.

    “So ChatGPT seems to be able to explain, adapt to the user's level of difficulty, and respond to follow-up questions.”
    Quote in the paperSubstantial claim

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

    Interpretationreasoning

    AI outputs are suitable truth indicators insofar as they bear a reliable relationship to the truth values of the propositions they concern.

    “These outputs are suitable truth indicators insofar as they bear a reliable relationship to the truth values of those propositions.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Some AI systems produce outputs that bear a reliable relationship to the truth values of the propositions they concern.

  • May combine two claims.

    Interpretationreasoning

    Current AI systems most likely do not meet the requirements for being epistemic authorities: they lack beliefs and communicative intentions and cannot engage in genuine testimonial interactions.

    “They lack beliefs and communicative intentions, and are incapable of truly participating in testimonial interaction, so they do not fit the standard picture of the EA-subject relationship.”
    • Also stated in: § 2. The Case Against AEAs · hedged
      “for now, it seems fair to assume that current AI systems (most likely) do not meet these requirements for being epistemic authorities.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    AI systems already have impressive capabilities and can do many of the things that human epistemic authorities typically do.

    “These conversations with ChatGPT suggest that AI systems already have impressive capabilities and can do many of the things that human epistemic authorities typically do.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: AI systems are often significantly better at truth-finding or other tasks than average humans or sometimes even experts.

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

    Premisecitation

    To epistemically defer to a scientific community or other group is to treat suitable facts about that group, such as consensus or majority belief, as strong indicators of the truth of a proposition.

    “As I argue at length elsewhere (Hauswald 2024), to epistemically defer to a scientific community or other group is to treat suitable facts about that group, such as the fact that there is consensus in that group, or the fact that a majority of its members believe a certain proposition, as strong indicators of the truth of that proposition.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Hauswald 2024

      Not checked in the free run.

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

    Interpretationreasoning

    Suitable facts about an epistemic authority are such that they bear a reliable relation to the truth values of the propositions in question.

    “Suitable facts are such that they bear a reliable relation to the truth values of those propositions.”
    • Also stated in: § 5. Conclusion · plain
      “Appropriate facts are ones that bear a reliable relation to the truth values of those propositions.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    The motivation for adopting the modified picture is not ad hoc.

    “I have emphasized that the motivation for adopting this modified picture is not ad hoc.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given

    Access to the suitable facts about an epistemic authority can be gained through testimony or other means.

    “Access to these facts can be gained through testimony or other means.”
    • Also stated in: § 5. Conclusion · plain
      “And access to these facts can be gained through testimony or other means.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationnone given

    AI systems lack beliefs and communicative intentions, and the ability to engage in genuine testimonial interaction.

    “On the other hand, they lack beliefs and communicative intentions, and the ability to engage in genuine testimonial interaction”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    Both AI systems and human epistemic authorities are characterized by epistemic opacity.

    “both are characterized by epistemic opacity”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationnone given

    Both AI systems and human epistemic authorities are often epistemically superior to subjects/users.

    “both are often epistemically superior to subjects/users”
    Quote in the paperNot a substantial claim

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

    Interpretationnone given

    There is a reliable relationship between an epistemic authority's beliefs and the truth values of the corresponding propositions.

    “A plausible answer, I think, is that there is a reliable relationship between an EA's beliefs and the truth values of the corresponding propositions.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Access to AI outputs will typically be gained through observation.

    “And access to these outputs will typically be gained through observation.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    To defer to an epistemic authority is to regard suitable facts about that authority as indicators for the truth of certain propositions that are strong enough for the deferring subject to adopt the doxastic attitude that the indicator suggests.

    “To defer to an EA is to regard suitable facts about that EA as indicators for the truth of certain propositions that are strong enough for the deferring subject to adopt the doxastic attitude that the indicator suggests.”
    • Also stated in: § 5. Conclusion · plain
      “According to my modified picture, deferring to an EA means taking appropriate facts about that EA as strong indicators of the truth of certain propositions.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    To epistemically defer to an AI system is to take suitable facts about it-its outputs-as indicators for the truth of certain propositions that are strong enough for the user to adopt the doxastic attitude that the output suggests.

    “To epistemically defer to an AI system is to take suitable facts about it -its outputs -as indicators for the truth of certain propositions that are strong enough for the user to adopt the doxastic attitude that the output suggests.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    The third option may be particularly attractive.

    “In what follows, I will argue that the third option may nevertheless be particularly attractive.”
    Quote in the paperNot a substantial claim

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

    Premisenone given

    What distinguishes interactions between epistemic authorities and subjects from other testimonial interactions is the great epistemic distance: epistemic authorities know or understand much more about their domains of expertise.

    “What distinguishes interactions between EAs and subjects from other testimonial interactions is the great epistemic distance: EAs know or understand much more about their domains of expertise -which is precisely what makes them epistemic authorities with respect to those domains.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    It can be difficult or impossible to explain how an AI transformed its inputs into its outputs; AI is an opaque black box.

    “it can be difficult or impossible to explain how an AI transformed its inputs into its outputs, which is why AI has often been described as an opaque black box.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Kleinberg and Mullainathan 2015

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Premisecitation

    The standard view of epistemic-authority-subject interaction requires epistemic authorities to have beliefs and communicative intentions and to be able to engage in genuine testimonial interactions.

    “The picture requires epistemic authorities to have beliefs and communicative intentions, and to be able to engage in genuine testimonial interactions.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Pritchard 2004

      Not checked in the free run.

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

    Interpretationreasoning

    A particularly strong case can be made for developing an alternative model that encompasses artificial epistemic authorities (option 3).

    “I argue that while each option has its benefits and costs, a particularly strong case can be made for option 3.”
    Quote in the paperNot a substantial claim

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

    Interpretationreasoning

    If an AI system produces outputs that can reasonably be treated as strong indicators, it may well be appropriate to describe the system as an 'artificial epistemic authority'.

    “If an AI system produces outputs that can reasonably be treated as strong indicators, it may well be appropriate to describe the system as an 'artificial epistemic authority'.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Some AI systems produce outputs that bear a reliable relationship to the truth values of the propositions they concern.

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

    Interpretationreasoning

    Deferring to epistemic authorities involves treating their beliefs, or facts about their beliefs, as strong indicators of the truth of certain propositions, and access to these beliefs can be gained through testimony or other means.

    “Generalizing, as a first step toward an alternative model of the EA-subject relationship, we can say that deferring to epistemic authorities involves treating their beliefs (or facts about their beliefs) as strong indicators of the truth of certain propositions, and that access to these beliefs can be gained through testimony or other means.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationcitation

    It is not yet clear whether and how the opacity of AI can be sufficiently overcome.

    “Despite considerable progress in this program, however, it is not yet clear whether and how the opacity of AI can be sufficiently overcome (see Beisbart and Räz 2022 for a more detailed discussion).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Beisbart and Räz 2022

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Both the relationship between epistemic authorities and subjects and the relationship between AI systems and their users are epistemically imbalanced relationships.

    “both the relationship between EAs and subjects and the relationship between AI systems and their users are 'epistemically imbalanced relationships' (Seger 2022).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Seger 2022

      Not checked in the free run.

    • Rests on an assumption: AI systems are often significantly better at truth-finding or other tasks than average humans or sometimes even experts.

  • Classified as probably not a substantial claim.

    Premisecitation

    The paradigm case of testimony is an intentional transfer of a belief from one agent to another, whether via verbal assertion or another means such as a note.

    “The 'paradigm case of testimony' in turn is an 'intentional transfer of a belief from one agent to another, whether in the usual way via a verbal assertion made by the one agent to the other, or by some other means, such as through a note' (Pritchard 2004, 326).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Pritchard 2004

      Not checked in the free run.

26 of 26

  • The paper relies on this without support.

    AI systems are increasingly assuming roles traditionally occupied by human epistemic authorities.

    “While AI systems are increasingly assuming roles traditionally occupied by human epistemic authorities (EAs), their epistemological status remains unclear.”

  • The paper relies on this without support.

    AI systems are often significantly better at truth-finding or other tasks than average humans or sometimes even experts.

    “Similarly, AI systems are often significantly better at truth-finding or other tasks than average humans or sometimes even experts.”

  • The paper relies on this without support.

    Human epistemic authorities also tend to have expertise in fairly narrow domains.

    “For one thing, human EAs also tend to have expertise in fairly narrow domains.”

  • The paper relies on this without support.

    The capabilities of AI systems are often extremely narrow.

    “It is true that the capabilities of these systems are often extremely narrow.”

  • The paper relies on this without support.

    Some AI systems produce outputs that bear a reliable relationship to the truth values of the propositions they concern.

    “If an AI system produces outputs that can reasonably be treated as strong indicators, it may well be appropriate to describe the system as an 'artificial epistemic authority'.”
    “These outputs are suitable truth indicators insofar as they bear a reliable relationship to the truth values of those propositions.”
5 of 5

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Trust is an overarching theme that arises in all previously discussed topics and the development and deployment of AI generally.

    “Trust is an overarching theme that arises in all previously discussed topics and the development and deployment of AI generally [5, 138, 139, 145, 146].”

    From Autonomous artificial intelligence, scientific research, and human values (2026) · cites this paper as [138]

Cites

Flagged first, then by citations

2 of the 56 references couldn’t be matched to a known paper.

Research works printed but not linked: 5

  1. Jäger, Christoph. Forthcoming. “Epistemic Authority.” In Oxford Handbook of Social Epistemology, edited by Jennifer Lackey and Aidan McGlynn. Oxford: Oxford University Press.
  2. Keren, Arnon. 2020. “Trust, Preemption, and Knowledge.” In Trust in Epistemology, edited by Katherine Dormandy, 114–135. New York: Routledge.
  3. Kleinberg, Jon, and Sendhil Mullainathan. 2015. “We Built Them, but We Don’t Understand Them.” In What to Think About Machines That Think, edited by John Brockman, 62–65. New York: Harper Perennial.
  4. Millgram, Elijah. 2015. The Great Endarkenment. Philosophy for an Age of Hyperspecialization. Oxford: Oxford University Press.
  5. Zimmer Ben. 2024. “A Name for AI That Sounds a Bit Less Intelligent.” The Wall Street Journal January 20 2024.

Cited by

Sorted by citations

Related

Authors
Hauswald, R.
Affiliations
Technische Universität Dresden (DE)
Corresponding author
Hauswald, Rico
Venue
Social Epistemology
Publisher
Informa UK Limited
Published
2025-01-13
Volume
39
Issue
6
Pages
716-725
ISSN
  • 0269-1728 (ISSN-L)
  • 1464-5297
DOI
10.1080/02691728.2025.2449602
OpenAlex
W4406340304
Semantic Scholar
4a0da704154fbc62824c51a195dc8d07de13f71a
Cited by
46
References
56
Citations by year
2026: 36 · 2025: 8 · 2024: 1 · 2022: 1
FWCI
350.89
Type
journal article
Language
en
Open access
Licence
Locations
Social Epistemology (publishedVersion, cc-by)
Funders
BMBF
Topics
  • Social Sciences › Arts and Humanities › Philosophy › Epistemology, Ethics, and Metaphysics
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Social Sciences › Sociology and Political Science › Feminist Epistemology and Gender Studies
Keywords
epistemology of AI, epistemic authority, witness testimony, epistemic deference, epistemic opacity, epistemic asymmetry, identification problem, trust in AI

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 2026-10-07.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-07.

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