Why Is Most of What We Know Something We Have Never Seen for Ourselves?

Geoffrey Chen

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

Before leaving home in the morning, a person may already have relied on dozens of things they have never personally verified. They trust the probability of rain in a forecast, the railway timetable, the dosage printed on a medicine packet and the road drawn on a map. They know that Earth has internal layers, that a war occurred before they were born and that an invisible virus can be identified by a test. Very little of this knowledge comes from seeing for oneself. It comes through broadcasts, textbooks, instruments, archives, teachers, doctors, journalists and researchers one has never met.

If genuine knowledge required personal observation, the modern person's world would shrink within a day to something in which it would be almost impossible to live.

This does not make first-hand observation unimportant. It reveals a more basic fact: human knowledge has never been rebuilt independently by every person from raw reality. We can know about distant places, the past, microscopic entities and other people's experience because someone observed, recorded, checked and preserved something, then passed it on through language and institutions. Philosophers use testimony for this broad way of acquiring knowledge from what others tell us. It is not confined to statements in court. “The meeting has moved to three o'clock”, a history book and a scientific report can all enter testimonial relations. The Stanford Encyclopedia of Philosophy's discussion of testimony begins from the scale of this dependence: much of what we know about history, science, politics and one another comes from others, yet exactly how another person's word can give us justified knowledge remains a serious philosophical problem.

“Verify everything yourself” is an impossible ideal

In response to misinformation, people often say: do not believe others; do your own research. The warning contains something right, but it can also produce an illusion.

Suppose I want to establish personally that a medicine contains the ingredients stated on its label. I would need a laboratory, instruments, reagents and training. Even if I performed the test, I would rely on the calibration of the instrument, the purity of the reagents and the validity of the documented method. If I wanted to verify a historical event, I could not return to the past. I would have to compare archives, physical evidence and contemporary records. Even when I check today's temperature, I usually do not perceive “temperature itself”. I read a device made and calibrated according to established standards.

Personal verification usually does not remove dependence on other people's knowledge. It moves the dependence to a different chain. We can shorten some distances, inspect original data or repeat an experiment, but we cannot bring all the labour supporting knowledge back inside a single person.

That is why modern knowledge is both powerful and vulnerable. It is powerful because people can divide epistemic labour. One person collects samples, another designs a method, another reviews the work, another preserves the record, and another translates a technical result into language the public can use. It is vulnerable because error, conflicted interests and distortion can also travel through the chain. An individual need not perform every task, but should know what kind of dependence is involved.

Believing a statement is not merely receiving a sentence

Testimony does not copy a proposition from one mind into another. A speaker must be in some position to know and must express what they know with adequate accuracy. A listener must judge whether the statement is fit to be accepted in the present context.

“The train has been cancelled” might come from someone looking at the platform notice, or from a social-media post published three hours earlier. “This treatment is usually effective” might be a doctor's qualified summary of clinical guidance, or a friend's generalisation from one personal experience. The sentences can sound similar while resting on completely different positions of observation, scopes of application and conditions of correction.

Reasonable trust therefore asks at least four questions. How does the speaker know? Within what scope is the claim meant to hold? Is there independent support for it? If it is wrong, can the error be found and corrected?

This does not mean investigating every ordinary exchange. We adjust the intensity of checking to the consequence. When a friend says a café is closed today, provisional acceptance is normally enough. A claim affecting a major medical decision, a legal duty or public safety needs a more reliable path. Mature knowledge is neither uniform credulity nor uniform suspicion. It calibrates confidence to the risk of the claim, the character of the source and the possibility of review.

Cognitive researchers have used the term epistemic vigilance for the mechanisms by which people evaluate communicated information and its sources. In their interdisciplinary paper on epistemic vigilance, Dan Sperber and colleagues argue that humans depend massively on communication and are therefore exposed to accidental error and deliberate deception. Communication remains advantageous partly because people assess both the credibility of informants and the plausibility of what is said. Vigilance, importantly, does not require treating everyone as a liar. Total distrust would prevent shared knowledge from forming; indiscriminate trust would abandon judgement.

Provenance is not a footnote to knowledge

On a printed page, a statement often remains attached to an author, a publication and a date. In a digital feed, sentences are easily detached from their origins. Screenshots are forwarded; charts lose titles and axes; somebody's summary of a study displaces the study. Generative AI adds another layer of compression. It can reorganise many texts into a fluent response without necessarily showing which source supports which claim, and it can sometimes produce references that do not exist.

Fluency can then masquerade as testimony. But an AI system does not become the witness of an event, the performer of an experiment or the professionally accountable institution behind a conclusion merely by producing a natural sentence. It is better understood, in this context, as an intermediary capable of rapidly recomposing public expressions. It can help formulate questions, identify lines of inquiry and compare explanations. When factual accuracy bears on an important decision, the reader still needs identifiable sources, dates, jurisdictions, methods and any subsequent corrections.

Research on online evaluation suggests a useful practice: instead of looking only within a page for signs that it is trustworthy, leave it temporarily and inspect how other sources describe the site, author or claim. In a study comparing professional fact-checkers, historians and university students, Sam Wineburg and Sarah McGrew found that fact-checkers more often used this practice of lateral reading. They quickly opened other pages to assess the original site's standing. Their paper, “Lateral Reading: Reading Less and Learning More When Evaluating Digital Information”, does not offer a machine for guaranteeing truth. It redirects attention from “Does this page look credible?” to “What relations of provenance and assessment does it occupy?”

Provenance matters for more than scholarly manners. A traceable claim allows a reader to discover whether it has been taken out of context, superseded or restricted to a particular sample. It also permits an error to travel back through the route by which the claim arrived. An unattributed statement may happen to be true, but it is difficult to maintain as shared knowledge because people cannot collectively inspect or repair it.

Seeing for oneself is not the end of inquiry

The reverse point matters too: first-hand observation is not automatically superior to a report from others. An eyewitness can misjudge distance; memory can be reconstructed later; a person may not understand the significance of what they saw. Professional knowledge often improves on isolated intuition precisely because its institutions do not stop at one observation. They require methods to be recorded, cases to be compared, claims to face criticism and conclusions to be revised when new evidence arrives.

First-hand and second-hand knowledge are therefore not a simple hierarchy. Direct observation can be close to an event but poor in comparison and interpretation. Mediated knowledge can be farther away yet more thoroughly checked. What needs evaluation is the whole epistemic chain: where information originated, how it was transformed, which checks were independent, who can identify an error and what happens when an error is found.

This also shows why “thinking for myself” should not mean “trusting only myself”. Someone who rejects external knowledge confines the world to one field of vision and one lifetime. Someone who delegates all judgement to an authority, platform or AI loses the ability to identify the dependencies shaping belief. Epistemic autonomy is not the independent production of every fact. It is the capacity to explain why trust is warranted here, why judgement should be suspended there, and what evidence would change one's mind.

A world held together by revisable trust

That most of our knowledge was never personally witnessed is not a defect in modern knowledge. It is the condition under which a person can inhabit a world larger than private experience. Language lets an observation leave the observer. An archive lets it cross time. Institutions allow different people's work to continue one another. Education means each generation need not begin at zero.

The extension is not free. Knowledge received through others raises questions of trust, power and responsibility. Who is readily believed? Whose experience is routinely discounted? Which institutions can survive scrutiny? Which platforms merely amplify popular claims? The individual's task is not to imagine an escape from these relations, but to keep the relations visible.

The most reliable knowledge life therefore treats neither personal sight as the only standard nor “everyone says so” as the final one. It acknowledges dependence while preserving traceability. It accepts divisions of labour while requiring correction. It allows others to bring the world into view while retaining responsibility for sources and reasons.

A person does not build a world by personally re-experiencing every fact in it. The more realistic and more difficult task is to establish, wherever direct sight ends, forms of trust that are reliable enough to use and open enough to revise.

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.