The DOI matches its Crossref record; 53 of the 53 references it cites were found in a registry or an index; 23 of the 55 sources its claims cite were checked for support.
Hallucination as epistemic pollution: epistemic injustice and credibility in futures-oriented science education
Open accessAbstract
In this critical conceptual review, I ask what generative artificial intelligence does to the distribution of epistemic authority in futures-oriented science education. I argue that the phenomenon usually described as hallucination is best understood not as isolated error but as epistemic pollution; that is, as a cumulative degradation of the cues by which learners judge reliability. I draw on four literature domains that have developed largely in isolation, namely technical and critical work on hallucination in natural language generation, philosophical work on epistemic injustice, futures studies scholarship on futures literacy, and science education research on future scaffolding and epistemological access. I first distinguish classes of unsupported output and specify six conditions under which such output counts as pollution rather than ordinary error, namely (1) persistence after correction, (2) circulation at scale, (3) contamination of downstream reasoning, (4) degradation of shared credibility cues, (5) opacity of provenance, and (6) the externalisation of verification costs. I then position the large language model not as an epistemic agent but as a component within a classroom credibility infrastructure, and I identify three mechanisms through which pollution may become injustice: (i) credibility displacement, (ii) hermeneutical narrowing, and (iii) unequal verification labour. Because these are conceptual hypotheses rather than demonstrated classroom effects, I state the conditions under which each would constitute an injustice rather than a harm borne equally, and what evidence would test each. I finally advance five pedagogical design principles, namely (a) provenance first inquiry, (b) collective verification, (c) epistemic role protection, (d) uncertainty discipline, and (e) plural futures safeguards. I offer these principles as regulative ideas alongside a research agenda for their empirical evaluation.
Overview
CoChat Summary
TL;DR
This critical conceptual review argues that hallucination in futures-oriented science education is best understood as epistemic pollution—a cumulative degradation of credibility cues—rather than isolated error. It specifies conditions under which such pollution may become epistemic injustice and proposes five pedagogical design principles, while stating that the mechanisms are conceptual hypotheses rather than demonstrated classroom effects.
Key findings
- The paper distinguishes unsupported output into factual error, fabrication, unfaithfulness to source, unsupported inference, staleness, retrieval failure, and apparent contradiction; its pollution thesis applies to fabrication, unfaithfulness to source, and unsupported inference delivered in the register of warrant, plus unprovenanced factual error and staleness.
- It specifies six conditions for epistemic pollution: persistence after correction, circulation at scale, contamination of downstream reasoning, degradation of shared credibility cues, opacity of provenance, and externalisation of verification costs.
- It identifies three mechanisms through which pollution may become injustice: credibility displacement, hermeneutical narrowing, and unequal verification labour; four further conditions for injustice are differential model performance, differential verification position, naturalisation of dominant framings, and institutional authorisation.
- It advances five design principles: provenance first inquiry, collective verification, epistemic role protection, uncertainty discipline, and plural futures safeguards. No quantitative hallucination figures are reported.
Methods
- Critical conceptual review, not a systematic review or empirical study; no primary data were collected, no classroom observation, learner interview, or hallucination-rate audit was conducted, and no quantitative hallucination figures are reported.
- Synthesises four literature domains: technical and critical work on hallucination in natural language generation, philosophical work on epistemic injustice, futures studies scholarship on futures literacy, and science education research on future scaffolding and epistemological access.
- Searches were conducted in Scopus, Web of Science, and Google Scholar between March and July 2026; foundational philosophical and futures studies texts were not date limited, while technical and educational literature on generative systems was restricted to publications from 2020 onward, with backward and forward citation chaining from bridging works.
Limitations
- The paper states that the relationships proposed in Sect. 5 are conceptual rather than empirically demonstrated; credibility displacement, hermeneutical narrowing, and unequal verification labour are advanced as mechanisms supported by adjacent literatures, not observed together in a futures-oriented science classroom.
- The paper states that the classroom vignette in Sect. 5.4 is constructed, not recorded; the design principles in Sect. 6 are regulative ideas that have not been tested; and the review is critical and selective rather than exhaustive, with science education treated as a demanding test case rather than a representative one.
- The paper also states that model capabilities and mitigation strategies change quickly, so specific technical assertions should be treated as provisional, and that the pollution metaphor may understate beneficial uses, learner resistance, and contexts where generative systems widen access. Evident from the text: the four injustice conditions and five principles remain untested, and no systematic protocol or primary data are provided.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
22 Valid · 0 Not valid · 23 Not checked · 123 Own results and interpretations
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“A generative system thus supplies the signals of credibility while bypassing the process that, in the classroom and in inquiry generally, those signals are meant to indicate [2,45].”
§ Hallucination as epistemic pollution: epistemic injustice and credibility in futures-oriented science education > 4 Generative AI within the classroom's credibility infrastructure
What the sources say
- Not checked — no full text of the sourcecites [2] Bender EM, Gebru T, McMillan-Major A, Shmitchell S. On the dangers of stochastic parrots: can language models be too big? In Proceedings of the 2021 ACM conference on fairness, accountability, and transparency. 2021;610–623. https://doi.org/10.1145/3442188.3445922 source
- Supportedcites [45] Shanahan M. Talking about large language models. Commun ACM. 2024;67(2):68–79. https://doi.org/10.1145/3624724. source
In “What about Embodiment” (page not available), Shanahan says an LLM’s responses can take the form of multistep arguments, but that apparent reasoning comes from mimicking well-formed arguments rather than a process faithful to logic. This supports the attributed distinction between signals of credibility and the process they appear to indicate; the classroom application is the citing authors’ framing. (LLM, conf=0.88)
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.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“In this respect, generative artificial intelligence must be understood not merely as instructional technology, but as a participant in the classroom's credibility infrastructure, whose presence rearranges knowledge relations [12,28].”
§ Hallucination as epistemic pollution: epistemic injustice and credibility in futures-oriented science education > 4 Generative AI within the classroom's credibility infrastructure
What the sources say
- Partly supportedcites [12] Heersmink R, De Rooij B, Clavel Vázquez MJ, Colombo M. A phenomenology and epistemology of large language models: transparency, trust, and trustworthiness. Ethics Inf Technol. 2024;26(3):41. https://doi.org/10.1007/s10676-024-09777-3. sourceExperimental — accuracy not yet measured
The Introduction describes LLMs as cognitive artifacts that shift the division of work between users and technology, and its Epistemology discussion examines how their outputs affect users’ trust and knowledge-seeking. It does not examine classrooms or establish that AI rearranges a classroom’s credibility relations. It also cautions against presenting LLMs as epistemic agents rather than instruments. (LLM, conf=0.85)
- Not checked — no full text of the sourcecites [28] Lee JD, See KA. Trust in automation: designing for appropriate reliance. Hum Factors J Hum Factors Ergon Soc. 2004;46(1):50–80. https://doi.org/10.1518/hfes.46.1.50_30392. source
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“A claim may be true and yet unwarranted for the person asserting it, if they cannot trace the path by which it was justified”
§ Hallucination as epistemic pollution: epistemic injustice and credibility in futures-oriented science education > 3 Epistemic injustice as a lens on classroom knowledge relations
What the sources say
- Partly supportedcites [23] Küçükuncular A. Justified true belief + diachronic justification: a contemporary defence. Philosophies. 2025;10(6):126. https://doi.org/10.3390/philosophies10060126. sourceExperimental — accuracy not yet measured
In ‘Diachronic Adequacy (Context-Sensitive)’ (page unavailable), the paper says Smith’s belief was true but lacked a stable chain of justification. That supports the distinction between truth and adequate warrant, but the paper does not make a person’s ability to trace that chain a requirement; it also allows support through memory and testimony without extensive individual checking. (LLM, conf=0.82)
hedgedstates a possible cause- Asserted: passes — The authors put it forward as true.
- Checkable: unclear — It is unclear whether it could be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“UNESCO's [48] global guidance treats generative artificial intelligence as a matter of governance as much as pedagogy, precisely because these systems introduce risks that cannot be addressed by classroom technique alone, including risks to privacy, intellectual integrity, and the conditions of equitable access to reliable knowledge.”
§ Hallucination as epistemic pollution: epistemic injustice and credibility in futures-oriented science education > 4 Generative AI within the classroom's credibility infrastructure
What the sources say
- Supportedcites [48] UNESCO. Guidance for generative AI in education and research. 2023. https://doi.org/10.54675/EWZM9535 source
The Guidance is explicitly framed as both regulation/policy and pedagogy: the summary (p. 48) and Introduction (p. 10) propose steps for governments to regulate GenAI (mandating data privacy protection, age limits) while also proposing a 'human-agent and age-appropriate approach to the ethical validation and pedagogical design processes' and frameworks for instructional design (Foreword, p. 5). The named risks are all present: unprotected user data privacy absent national regulation (p. 3), concerns over safety, data privacy, copyright and manipulation (Introduction, p. 10), and controversies from worsening digital poverty and homogenization of opinions to copyright, with long-term implications for knowledge, teaching, learning and assessment (p. 48, p. 3). Section 3/4 (pp. 21, 27) confirm the governance orientation, i.e. that these issues require policy measures beyond classroom technique. (LLM, conf=0.90)
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.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“the machinery of justification is precisely what a generative system withholds even when its output happens to be correct [23].”
§ Hallucination as epistemic pollution: epistemic injustice and credibility in futures-oriented science education > 3 Epistemic injustice as a lens on classroom knowledge relations
What the sources say
- Partly supportedcites [23] Küçükuncular A. Justified true belief + diachronic justification: a contemporary defence. Philosophies. 2025;10(6):126. https://doi.org/10.3390/philosophies10060126. sourceExperimental — accuracy not yet measured
The Abstract and Section 3 argue that a true belief can be correct by luck without the durable justification required for knowledge. They do not discuss generative systems or establish that such systems withhold justification; that application goes beyond the source. (LLM, conf=0.82)
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.
- Partly supported
51 more claims · 7 with their cited sources checked
Create a free account to see all 168 claims and to verify the 1 reference still open.
- It may state two separate claims as one. 62
- Probably not a substantial claim. 89
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
§ Hallucination as epistemic pollution: epistemic injustice and credibility in futures-oriented science education > 1 Introduction
“UNESCO's [47] call for a renewed social contract for education”
§ Hallucination as epistemic pollution: epistemic injustice and credibility in futures-oriented science education > References
“47. Stroupe D, Suárez E, Scipio D. Epistemic injustice and the "nature of science. " J Res Sci Teach. 2025;62(4):901-41.”
Citations
Cites
Epistemic Injustice: Power and the Ethics of Knowing
Fricker, M. (2007) · doi:10.1093/acprof:oso/9780198237907.001.0001 · Cited by 9,198
ChatGPT for good? On opportunities and challenges of large language models for education
Kasneci, E., Sessler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., Gasser, U., Groh, G., Günnemann, S., Hüllermeier, E., et al. (2023) · Learning and Individual Differences · doi:10.1016/j.lindif.2023.102274 · Cited by 7,326 · cited 4×
“Educational analyses of large language models have already drawn attention to the risk that learners may outsource explanation and reasoning, thereby weakening epistemic agency even when short-term performance appears improved [20].”
InfluentialOn the Dangers of Stochastic Parrots: Can Language Models Be Too Big? 🦜
Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021) · Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency · doi:10.1145/3442188.3445922 · Cited by 6,988 · cited 5×
“On the one hand, the learner may acquire false certainty, mistaking fluency for justification, and persuasion for epistemic entitlement [2].”
InfluentialTrust in Automation: Designing for Appropriate Reliance
Lee, J. D., & See, K. A. (2004) · Human Factors: The Journal of the Human Factors and Ergonomics Society · doi:10.1518/hfes.46.1.50_30392 · Cited by 6,605 · cited 5×
“Trust governs reliance under complexity, but it must be calibrated to actual trustworthiness, lest it become either blind dependence or indiscriminate distrust [14, 28].”
InfluentialSurvey of Hallucination in Natural Language Generation
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., & Fung, P. (2023) · ACM Computing Surveys · doi:10.1145/3571730 · Cited by 5,025 · cited 7×
“The technical literature on hallucination shows that plausible text may arise without grounding, and that the surface marks of coherence do not suffice for epistemic entitlement [18].”
InfluentialHuang, L., Yu, W., Ma, W., Zhong, W., Feng, Z., Wang, H., Chen, Q., Peng, W., Feng, X., Qin, B., et al. (2025) · ACM Transactions on Information Systems · doi:10.1145/3703155 · Cited by 4,688 · cited 4×
“Recent taxonomic work distinguishes at least seven cases [15, 18].”
InfluentialMisinformation and Its Correction: Continued Influence and Successful Debiasing
Lewandowsky, S., Ecker, U. K. H., Seifert, C. M., Schwarz, N., & Cook, J. (2012) · Psychological Science in the Public Interest · doi:10.1177/1529100612451018 · Cited by 2,920
Trust in Automation: Integrating Empirical Evidence on Factors That Influence Trust
Hoff, K. A., & Bashir, M. (2015) · Human Factors: The Journal of the Human Factors and Ergonomics Society · doi:10.1177/0018720814547570 · Cited by 2,817
Tracking Epistemic Violence, Tracking Practices of Silencing
Dotson, K. (2011) · Hypatia · doi:10.1111/j.1527-2001.2011.01177.x · Cited by 2,150 · cited 2×
“Dotson’s [6] analysis of silencing adds a further caution, because a speaker may be silenced not only by being prevented from speaking, but by a failure of uptake that renders their contribution inert within the exchange.”
Jasanoff, S., & Kim, S.-H. (2009) · Minerva · doi:10.1007/s11024-009-9124-4 · Cited by 1,911 · cited 1×
“…futurising becomes still more visible: collective visions of desirable futures do not merely describe possibilities, they organise expertise, allocate legitimacy, and guide public investment, often by binding technological expectations to moral narratives of national or institutional purpose [17].”
Research works printed but not linked: 1
- Lewis P, Perez E, Piktus A, Petroni F, Karpukhin V, Goyal N, et al. Retrieval-augmented generation for knowledge-intensive NLP tasks. Adv Neural Inf Process Syst. 2020;33:9459–74.
Metadata
- Authors
- Küçükuncular, A.
- Affiliations
- Near East University (CY)
- Corresponding author
- Küçükuncular, Ahmet
- ORCID
- Küçükuncular, Ahmet
- Venue
- Discover Artificial Intelligence
- Published
- 2026-09-23
- Volume
- 6
- Issue
- 1
- ISSN
- 2731-0809 (ISSN-L)
- OpenAlex
- W7214080237
- Semantic Scholar
- 88e86f6dcbe8609647d3cdc3bdaa81605f7ab4b8
- Cited by
- 1
- References
- 50
- Citations by year
- 2026: 1
- FWCI
- 5.67
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by
- creativecommons.org/licenses/by/4.0 (tdm)
- creativecommons.org/licenses/by/4.0 (vor)
- Locations
- Discover Artificial Intelligence (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion)
- Topics
- Social Sciences › Psychology › Developmental and Educational Psychology › Educational Strategies and Epistemologies
- Social Sciences › Education › Science Education and Pedagogy
- Social Sciences › Education › Digital Education and Society
- Keywords
- hallucinations, epistemic injustice, generative AI, large language models, credibility, epistemic authority, provenance, uncertainty, futures literacy
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-05, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-05 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and 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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Hallucination as epistemic pollution: epistemic injustice and credibility in futures-oriented science education. https://app.cochat.ai/papers/10.1007/s44163-026-02318-5
- The paper
- Küçükuncular, A. (2026). Hallucination as epistemic pollution: epistemic injustice and credibility in futures-oriented science education. Discover Artificial Intelligence. https://doi.org/10.1007/s44163-026-02318-5
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.