Epistemic Deference to AI
Open accessThe DOI matches its Crossref record; 43 of the 43 references it cites were found in a registry or an index; 1 of the 22 sources its claims cite were checked for support.
Abstract
When should we defer to AI outputs over human expert judgment? Drawing on recent work in social epistemology, I motivate the idea that some AI systems qualify as Artificial Epistemic Authorities (AEAs) due to their demonstrated reliability and epistemic superiority. I then introduce AI Preemptionism, the view that AEA outputs should replace rather than supplement a user’s independent epistemic reasons. I show that classic objections to preemptionism – such as uncritical deference, epistemic entrenchment, and unhinging epistemic bases – apply in amplified form to AEAs, given their opacity, self-reinforcing authority, and lack of epistemic failure markers. Against this, I develop a more promising alternative: a total evidence view of AI deference. According to this view, AEA outputs should function as contributory reasons rather than outright replacements for a user’s independent epistemic considerations. This approach has three key advantages: (i) it mitigates expertise atrophy by keeping human users engaged, (ii) it provides an epistemic case for meaningful human oversight and control, and (iii) it explains the justified mistrust of AI when reliability conditions are unmet. While demanding in practice, this account offers a principled way to determine when AI deference is justified, particularly in high-stakes contexts requiring rigorous reliability.
Overview
CoChat Summary
TL;DR
The paper argues that some AI systems can qualify as Artificial Epistemic Authorities (AEAs), but AI Preemptionism—treating their outputs as replacing one’s own reasons—amplifies classic objections about uncritical deference, epistemic entrenchment, and severed epistemic bases. It proposes a total evidence view in which AEA outputs are contributory reasons integrated with user reasoning under defeasible conditions.
Key findings
- Some AI systems may qualify as AEAs if epistemic authority is understood as superior epistemic positioning and reliability rather than necessarily requiring beliefs and communicative intent; they can provide reliable advice, surpass humans in narrow domains, and exhibit opacity akin to human experts.
- AI Preemptionism, on which an AEA’s output replaces rather than supplements a user’s independent epistemic reasons, is motivated by the track-record argument but inherits amplified objections: opacity and perceived infallibility risk uncritical deference; algorithmic self-reinforcing ecosystems risk entrenchment; and black-box outputs can sever users’ epistemic bases without replacement.
- The proposed Total Evidence AI Deference treats AEA output as a contributory reason, with “Critical Deference with Oversight”: users should withhold or revisit deference if there is domain mismatch, reliability undermining, conflicting authority, or novel evidence; otherwise, they defer.
- The account claims three advantages: it mitigates expertise atrophy, provides an epistemic rationale for meaningful human oversight and control, and explains justified epistemic mistrust when reliability conditions are unmet.
Methods
- Conceptual and normative philosophical analysis in social epistemology; no empirical data, sample, or statistical analysis is reported.
- Defines and compares AI Preemptionism and Total Evidence AI Deference, drawing on literature on epistemic authority, preemption, and testimony, and uses objections and desiderata to motivate the alternative.
- Applies the account to current AI use, especially high-stakes domains such as medical diagnostics, finance, and aerospace.
Limitations
- Stated: few current AI systems will easily satisfy the requirements to qualify as AEAs; meeting benchmarks may demand interpretable architectures, extensive continuously updated domain-specific data, and transparent reporting of errors or biases, which some may consider unrealistic for large-scale or commercial deployments under market pressures.
- Stated: the literature on epistemic deference to AI remains sparse, with Wolkenstein (2024) and Hauswald (2025; forthcoming) among the few notable contributions.
- Evident from the text: no empirical validation, case study, or formal decision procedure is provided; the account remains programmatic and relies on contested philosophical definitions of epistemic authority and deference.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
1 Valid · 3 Not valid · 15 Not checked · 55 Own results and interpretations
Every source the paper cites for this claim was checked and supports it.
In the paper
“Drawing on recent work in social epistemology, I motivate the idea that some AI systems qualify as Artificial Epistemic Authorities (AEAs) due to their demonstrated reliability and epistemic superiority.”
- Also stated in: § 2 Artificial Epistemic Authorities and AI Preemptionism · hedged
“There are good reasons to think that some AI systems might qualify as AEAs (Hauswald, 2025) -especially if we entertain the idea that epistemic authority concerns superior epistemic positioning and reliability rather than necessarily beliefs and communicative intent.”
§ Epistemic Deference to AI > Abstract
What the sources say
- Supportedcites Hauswald, 2025 source
Section 3 ('The Case for AEAs') states the paper's positive case that 'some current AI systems may qualify as AEAs', grounding it in exactly the two features the claim names: epistemic superiority ('AI systems are often significantly better at truth-finding or other tasks than average humans or sometimes even experts' — medical AI, trading AI, AlphaGo; Note 1 makes epistemic superiority the minimalist definition of an EA) and reliability (Section 4: outputs are 'suitable truth indicators insofar as they bear a reliable relationship to the truth values of those propositions', so 'it may well be appropriate to describe the system as an artificial epistemic authority'). The Abstract and Conclusion (Section 5) confirm the paper's stance — it argues option 3, an alternative model that 'encompasses artificial epistemic authorities', is the strongest. The only nuance is that Hauswald voices the qualification with hedges ('may qualify', 'may be a reason to think'), which the citing sentence's own framing ('I motivate the idea') mirrors rather than overstates. (LLM, conf=0.84)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 2 Artificial Epistemic Authorities and AI Preemptionism · hedged
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“AEAs can operate within algorithmically structured epistemic ecosystems that dynamically shape what users see, believe, and engage with, often in ways that are invisible to the users.”
§ Epistemic Deference to AI > 2 Artificial Epistemic Authorities and AI Preemptionism
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“First, many AI systems already provide helpful information, explanations, and reliable advice, functioning in ways that closely resemble traditional human EAs.”
§ Epistemic Deference to AI > 2 Artificial Epistemic Authorities and AI Preemptionism
plainstates an association- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“If these conditions remain unmet -for instance, due to opaque training data, a poor track record, or repeated biases -then it is epistemically warranted for users to withhold deference and maintain scepticism about the system's outputs.23”
§ Epistemic Deference to AI > 3.4 Lack of Epistemic Trust
plainstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“I show that classic objections to preemptionism -such as uncritical deference, epistemic entrenchment, and unhinging epistemic basesapply in amplified form to AEAs, given their opacity, self-reinforcing authority, and lack of epistemic failure markers.”
- Also stated in: § 2 Artificial Epistemic Authorities and AI Preemptionism · hedged
“These objections suggest that while AI Preemptionism might initially appear compelling -especially given AI's potential for superior reliability -it ultimately inherits and amplifies the classic problems associated with epistemic Preemptionism.”
§ Epistemic Deference to AI > Abstract
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § 2 Artificial Epistemic Authorities and AI Preemptionism · hedged
25 more claims
Create a free account to see all 74 claims and to verify the 12 references still open.
- It may state two separate claims as one. 23
- Probably not a substantial claim. 31
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Epistemic Deference to AI > 3.3 Meaningful Human Control
“While normative arguments for human oversight emphasise accountability, responsibility, and the preservation of autonomy (Amoros et al., 2020),”
§ Epistemic Deference to AI > References
“2. Amoroso, D., Tamburrini, G.: Autonomous weapons systems and meaningful human control: ethical and legal issues. Current Robot. Rep. 1, 187-194 (2020)”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Epistemic Deference to AI > 4 Implications for Current Use of AI
“For a similar argument in favour of using human second opinions to resolve disagreements between AI and human experts, see alsoKempt and Nagel (2021).”
§ Epistemic Deference to AI > References
“25. Kempt, H., Nagel, S.K.: Responsibility, second opinions and peer-disagreement: ethical and epis- temological challenges of using AI in clinical diagnostic contexts. J. Med. Ethics 48, 222-229 (2022)”
Citations
Cites
Lackey, J. (2008) · Oxford University Press eBooks · doi:10.1093/acprof:oso/9780199219162.001.0001 · Cited by 688
Epistemic Authority: A Theory of Trust, Authority, and Autonomy in Belief
Zagzebski, L. T. (2012) · doi:10.1093/acprof:oso/9780199936472.001.0001 · Cited by 489 · cited 2×
“In the case of human EAs, some have suggested that we should treat their judgments as preempting our own independent reasoning (Zagzebski, 2012; 2013; 2014; 2016).”
Durán, J. M., & Jongsma, K. R. (2021) · Journal of Medical Ethics · doi:10.1136/medethics-2020-106820 · Cited by 446
Han, R., Acosta, J. N., Shakeri, Z., Ioannidis, J. P. A., Topol, E. J., & Rajpurkar, P. (2024) · The Lancet Digital Health · doi:10.1016/s2589-7500(24)00047-5 · Cited by 336 · cited 1×
“16 This can lead to epistemic losses rather than gains.”
Artificial intelligence in Finance: a comprehensive review through bibliometric and content analysis
Bahoo, S., Cucculelli, M., Goga, X., & Mondolo, J. (2024) · SN Business & Economics · doi:10.1007/s43546-023-00618-x · Cited by 218
Artificial Intelligence Index Report 2023
Maslej, N., Fattorini, L., Brynjolfsson, E., Etchemendy, J., Ligett, K., Lyons, T., Manyika, J., Ngo, H., Niebles, J. C., Parli, V., et al. (2023) · arXiv (Cornell University) · doi:10.48550/arxiv.2310.03715 · Cited by 188
Trust and belief: a preemptive reasons account
Keren, A. (2014) · Synthese · doi:10.1007/s11229-014-0416-3 · Cited by 156
Nguyen, C. T. (2021) · Philosophy and Phenomenological Research · doi:10.1111/phpr.12823 · Cited by 152
EPISTEMIC AUTHORITY, PREEMPTIVE REASONS, AND UNDERSTANDING
Jäger, C. (2016) · Episteme · doi:10.1017/epi.2015.38 · Cited by 144
Autonomous Weapons Systems and Meaningful Human Control: Ethical and Legal Issues
Amoroso, D., & Tamburrini, G. (2020) · Current Robotics Reports · doi:10.1007/s43154-020-00024-3 · Cited by 144
Research works printed but not linked: 3
- Rico, H.: Forthcoming. “AI and the philosophy of expertise and epistemic authority. In: A Companion to Applied Philosophy of AI, edited by Martin Hähnel and Regina Müller. Wiley, Hoboken
- Christoph, J.: (forthcoming). Intellectual authority and education. In: Eder, A.-M., Brössel, P., Grundmann, T. (eds.) The Epistemology of Expert Judgement, Routledge
- Christoph, J.: (forthcoming). Epistemic authority. In: Lackey J., McGlynn A. (eds.) The Oxford Handbook of Social Epistemology. Oxford University Press, Oxford
Cited by
Milella, F., & Cabitza, F. (2026) · Machine Learning and Knowledge Extraction · doi:10.3390/make8020036 · Cited by 11
“Experts, indeed, are typically epistemic agents seen as epistemically superior to us concerning assertions within their areas of competence [27–29].”
Mühlhoff, R. (2026) · AI & Society · doi:10.1007/s00146-026-03281-6 · Cited by 4
Calibrated Epistemic Deference to Conversational AI in Mental Healthcare
Lange, B. (2026) · The American Journal of Bioethics · doi:10.1080/15265161.2026.2645593 · Cited by 2
Self-citationKnowledge Immunity in AI-Driven Organizations: Combating Misinformation and Epistemic Risk
Islam, M. A. *. (2026) · Qualitative Research Review Letter · doi:10.5281/zenodo.19482257
Cole, C. (2026) · LUTPub (LUT University)
The Captured Oracle: Authorship and Agency in the Ethics of Answer-Engine OptimizationPreprint
Walton, L. F. (2026) · Zenodo (CERN European Organization for Nuclear Research) · doi:10.5281/zenodo.20676327
From Kafka to the rabbi-machine: Artificial intelligence and the return of moral authority
Gazit, L. (2026) · AI and Ethics · doi:10.1007/s43681-026-01245-6
Mueller, M. R. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7276359
Rethinking AI in clinical decision support: a framework for reciprocal human-AI interaction
Greengrass, C. J. (2026) · Frontiers in Digital Health · doi:10.3389/fdgth.2026.1887161 · Cited by 2
Who decides after AI predicts? A framework for authority in treatment decisions
Mueller, M. R. (2026) · AI and Ethics · doi:10.1007/s43681-026-01462-z
Gozum, I. E. A. (2026) · Journal of Religion and Health · doi:10.1007/s10943-026-02818-9
Metadata
- Authors
- Lange, B.
- Affiliations
- Munich Center for Machine Learning (DE)
- Corresponding author
- Lange, Benjamin
- ORCID
- Lange, Benjamin
- Venue
- Lecture Notes in Computer Science
- Publisher
- Springer Nature Switzerland
- Published
- 2025-10-01
- Pages
- 174-186
- ISSN
- 0302-9743 (ISSN-L)
- 1611-3349
- OpenAlex
- W4414638843
- Semantic Scholar
- 83f85400b0b7595b7ebbc93e5c18103eba4a577a
- Cited by
- 12
- References
- 38
- Citations by year
- 2026: 12
- FWCI
- 148.69
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Lecture notes in computer science (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion, cc-by)
- PhilPapers (PhilPapers Foundation) (submittedVersion)
- arXiv (Cornell University) (cc-by)
- Topics
- Social Sciences › Arts and Humanities › Philosophy › Epistemology, Ethics, and Metaphysics
- Physical Sciences › Computer Science › Artificial Intelligence › Ethics and Social Impacts of AI
- Social Sciences › Sociology and Political Science › Feminist Epistemology and Gender Studies
- Keywords
- epistemic deference, epistemic authority, AI reliability, total evidence, epistemic entrenchment, AI opacity, human oversight of 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.
A language model drew the claims from the paper on 2026-11-14, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-14 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (qwen3.8-omni-flash).
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
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.