When We Cite Authority, Are We Borrowing Knowledge or Giving Up Judgment?

Geoffrey Chen

How a Person Builds a World · Season Two: “Others, Dialogue, and Shared Knowledge” · Article Ten

One passage says, “Harvard research has proved…” Another says, “A Nobel laureate believes…” A third is shorter still: “Science tells us…” Once an authoritative name appears, the sentence seems to acquire weight automatically. The reader may know neither the research question, sample nor conclusion, but further doubt already feels like ignorance of expertise.

No person, however, can inspect every item of knowledge independently. Citing researchers, institutions and professional consensus is a necessary practice within a modern division of epistemic labour. If every appeal to authority is dismissed as a logical fallacy, knowledge will be handed back to the most confident private intuition.

The issue is not whether authority is cited but what function it performs in the argument. It can be a bridge along which the reader enters knowledge beyond individual capacity. It can also be a wall, using status to prevent another question.

A citation should perform three tasks

First, a citation supplies provenance. A fact, figure or explanation is no longer only a sentence in the present author's voice. A reader can identify who proposed it, at what time and from which material.

Second, a citation compresses epistemic labour. A systematic review may integrate dozens of studies. An official statistic rests on complicated sampling and adjustment. The present text cannot repeat the full process. A reliable reference lets the reader provisionally borrow completed work.

Third, citation establishes a relation of accountability. If the author misrepresents a study, the reader can compare the original. If the research is corrected later, the old claim can be traced. The epistemic value of a reference is not simply “someone supports me”. It makes the supporting relation inspectable.

Together these functions show that citing authority need not surrender judgement to a name. Proper citation increases the conditions under which judgement is possible. It tells us where to inspect, why trust may be warranted and how disagreement can continue.

“Many experts say”, without field, source or date, performs almost none of this work. It uses the social impression of authority while withholding its epistemic structure.

Authority supplies a second-order reason within a relevant scope

An outstanding physicist does not thereby gain equal professional weight on education policy. A medical body can assess evidence about treatment but cannot decide, by technical status alone, what burden a patient should accept in life. Authority is always indexed to a domain and a type of question.

When an ordinary reader cites an expert, the expert's statement is generally not a first-order proof. It supplies a second-order reason: this person or institution is better equipped than I am to evaluate the relevant first-order evidence, so its judgement deserves greater weight. That reason is entirely legitimate, but conditional. The expertise must be relevant. The expert must have access to appropriate material, lack severe undisclosed conflicts and avoid speaking beyond the evidence.

When experts substantially disagree, selecting one person who supports one's conclusion turns expertise into decoration. The distribution of opinion, the source of disagreement and the reason for weighting one assessment above another become important. Authority is not a vote-counting game, but an outlying voice should not be presented as “science has established”.

A more useful formulation is not “Authority A says X, therefore X”, but: “On this question, relevant researchers or institutions arrived at X through these methods; this is the scope and remaining uncertainty.” The name remains relevant, and the structure of reasons remains present.

Citation chains can turn cautious findings into settled facts

Knowledge can be distorted even when every paper contains a genuine reference. A second paper cites the first paper's summary. A third reads only the second. Later authors repeat the statement. An initial observation with qualifications gradually becomes “widely proven”.

Steven Greenberg reconstructed the citation network surrounding a particular biomedical claim, analysing how supporting material was selectively amplified, contrary data neglected and a hypothesis transformed into an apparent fact. His study of how citation distortions create unfounded authority examined one scientific controversy and cannot establish that every literature behaves this way. It clearly demonstrates that citation quantity and evidential quality are not identical.

A systematic review and meta-analysis also found that studies reporting statistically significant results tended to be cited more than studies with non-significant findings, giving positive results greater representation in the literature. The research on scientific citations favouring positive results shows why “highly cited” may indicate influence, visibility and network position without independently establishing truth.

Ordinary readers cannot reconstruct a citation network each time. A principle of proportion helps. For general background, an accurate account from an authoritative review may be enough. When a claim carries the central argument, has serious practical consequences or appears unusually certain, one should return where possible to the original research or formal method document and check whether it supports the current sentence.

The decisive issue is not the existence of a reference but support alignment. Did the source study the same question? Do its sample and scope match the current claim? Did the paper establish association, prediction or causation? Has the citer removed a qualification on which the conclusion depends?

“I read the source” does not guarantee that judgement was preserved

Returning to an original paper can also become ritual. A person may find one sentence in the abstract that supports a prior position while ignoring the design and limitations. Technical appearance can itself be used to discourage readers from asking further questions.

Borrowing knowledge requires a minimum ability to restate it. The citer should be able to explain in ordinary language what the source claims, what material supports it, what it does not establish and why it is relevant to the present argument. If the person knows only a title and an author's name, authority may be substituting for understanding.

Demanding mastery of every technical detail is equally unrealistic. Understanding remains layered. We can rely on peer review, professional institutions and methodological transparency while identifying which layer we did not verify independently. “I rely on this review's integration of the specialist literature” is more judgemental than pretending to command the entire field.

Judgement does not require proving from zero. It requires describing dependence accurately.

AI separates the appearance of a reference from a real source

Generative AI can produce a perfectly formatted bibliography in seconds. Perfect format does not show that the works exist. A study checking bibliographies generated by ChatGPT documented fabricated and erroneous references and noted that false citations can look highly plausible. See the paper on fabrication and errors in AI-generated citations. Other research in medical literature search has recorded conversational systems spontaneously inventing references; one study examined citation fabrication in psychiatry searches.

Even when a link is real, the source may not support the sentence placed before it. An AI answer “with citations” therefore needs two checks: bibliographic identity and evidential correspondence. Clicking the link is the beginning of verification, not its completion.

A safer workflow asks the system to retrieve and cite accessible originals where possible; checks title, author, year and persistent identifier; opens decisive sources at the relevant section; seeks independent support for consequential claims; and states limitations when the full text is unavailable. AI should not be asked merely to “generate some references” and then rewarded for typographic plausibility.

AI can lower the cost of finding sources and multiply pseudo-authority. A bibliography that once required time to fabricate can now appear almost instantly. Personal judgement must therefore ask not only whether references are present, but whether they resolve, open and support.

Authority should extend judgement, not terminate it

In a mature knowledge life, citing authority is a constrained dependence. I acknowledge that others can perform some epistemic work better than I can and permit their result to enter my argument. At the same time, I retain checks on relevance of domain, the chain of sources, conflicts of interest, scope and correction.

When does citation approach a surrender of judgement? When the name replaces a specific claim; the authority moves outside its field; disagreement is concealed; sources cannot be traced; the writer cannot explain how the document supports the sentence; or any question is classified as opposition to science or knowledge.

When is it borrowing knowledge? When the reader can travel back through the reference; the scope of dependence is stated; the authoritative conclusion is compared with other evidence; later correction can alter the text; and the writer remains responsible for the use made of the knowledge.

Without authority, a person's world is confined by private experience. With authority alone, the world becomes a hierarchy supported by uninspectable names. Shared knowledge needs the more difficult route: trusting completed specialist labour while keeping its provenance, boundary and return path.

The most important punctuation in a citation is not the parenthesis at the end of a sentence. It is the path the citation opens. A good reference says: you do not have to repeat this whole work, but you can still know where it came from and inspect it again when necessary. That is not the abandonment of judgement. It is an acknowledgement of how judgement remains possible in a world of knowledge larger than a person.

Primary sources and further reading

Continue reading: Explore the How a Person Builds a World series.

Discover more from Geoffrey Chen

Subscribe to get the latest posts sent to your email.