ESSENTIAL PATH Stop 2 / 18

Reality October 31, 2021 Updated: September 11, 2026 16 min read THY-REALITY

How Do We Know What We Know? Evidence, Belief and Certainty

An epistemological map distinguishes belief, knowledge, evidential strength, probability, trust in sources and certainty, showing why good knowledge is not the same as feeling infallible.

In everyday language we use the word know very broadly. Sometimes it means 'I am convinced', sometimes 'I saw it', sometimes 'a reliable source told me', and sometimes 'there is a strong body of evidence for it'. These are not merely linguistic differences. They determine how well we distinguish robust knowledge from guessing, trust, probability and the feeling of certainty.

Prebujenje v Naravni zakon insists on an important distinction between belief and discovery/understanding: something does not become true merely because we believe it. Atlas resničnosti adds a stricter calibration of evidential strength: experience, report, interpretation and broader conclusion are not the same level. This article turns these two starting points into a general epistemological map.

The central question is not only 'Do I believe this claim?' but: 'Why do I accept it, how strong is the evidence, how certain may I reasonably be, and what would change my mind?'

Belief is not the same as knowledge

We can be deeply convinced and still be wrong. A person can sincerely believe a false memory, a poor explanation, an inaccurate source or a conclusion that feels obvious. The strength of conviction is therefore not, by itself, evidence that a claim is true.

Belief is not irrelevant, however. In propositional knowledge — knowing that something is the case — it usually makes sense to say that the person accepts the proposition as true. But knowledge requires more than internal assent. If we merely guess the correct answer, we have a true belief, yet we would hesitate to call it well-acquired knowledge.

The first discipline of epistemology is therefore modest: separate what we believe from the reasons that make it reasonable to believe it.

In analytic epistemology, a long-influential starting point held that knowledge includes at least three components: the proposition must be true, the subject must believe it, and the belief must be appropriately justified. This model remains useful for orientation, but it does not settle the analysis of knowledge.

The important distinction is between the truth of a claim and the quality of our route to that claim. Someone can reach a correct conclusion for a bad reason; another person can follow a very good procedure and still reach a false conclusion because the available data were incomplete. Epistemology therefore evaluates not only the final answer but also how we got there.

For THY-REALITY this matters greatly: the goal is not to reward a belief merely because it happens to agree with our preferred view, but to build procedures that increase the chance of detecting and correcting error.

The Gettier problem: justified true belief can still hit the truth by luck

In 1963 Edmund Gettier used short counterexamples to show that a person can apparently satisfy the conditions of justified true belief while the connection between reason and truth still contains too much epistemic luck. Philosophers have since developed many proposed additional conditions and alternative analyses of knowledge.

For our purposes we do not need to choose one final theory of knowledge. Gettier's lesson is more practical: a correct result does not by itself demonstrate a reliable route. If a stopped clock happens to display the correct time at the moment we look at it, we may obtain a true answer from a defective mechanism.

For important claims we therefore ask not only 'Is this correct?' but also 'Is the link between evidence and conclusion good enough that we would expect similarly reliable answers in comparable cases?'

The word evidence covers different epistemic roles in different domains. In mathematics, proof can mean a deductive demonstration from axioms and rules of inference. In experimental science we usually deal with measurements, observations, repeatable patterns and statistical support. In history, documents, material traces, contemporaneous records and independent testimony matter greatly. In everyday life we rely on memory, perception and the testimony of others.

It is therefore unfair to demand the same kind of evidence for every question. An archival document is not a laboratory experiment, yet it may be very strong evidence that a decision was made. A temperature measurement is not evidence of a person's intention. Personal testimony can strongly support the claim that someone had an experience while supporting a distant metaphysical conclusion much less strongly by itself.

This article therefore uses evidence in a broad epistemic sense: information that legitimately increases or decreases the reasonableness of a claim. How We Prove Hidden Operations: From Suspicion to Document will later specialise this principle for establishing real conspiracies and covert operations.

Evidential strength is not a binary switch

Many debates become confused because we use only two boxes: 'proven' and 'unproven'. Real inquiry often contains degrees of support. One independent document may make a claim more likely; several mutually consistent sources may strengthen it; a direct measurement may exclude some alternatives; new contrary evidence may lower confidence again.

This does not mean that everything is relative. On the contrary, recognising degrees of evidential strength lets us state how much the data warrant. 'There is an indication' is different from 'several independent sources converge', which is different again from 'the result has been independently replicated several times'.

The Atlas idea of calibrating evidence is therefore useful far beyond metaphysics: a conclusion should not be stronger than the evidential base carrying it.

A person can feel completely certain and be wrong. At the same time, we can reasonably know something without possessing absolute certainty. Contemporary epistemology therefore often allows fallibilistic knowledge: knowledge for which error remains possible even though the reasons are sufficiently good for justified acceptance.

This corrects two extremes. The first says: 'If it is not one hundred percent certain, we know nothing.' The second says: 'Because perfect certainty is rare, every opinion is equally good.' Neither follows. Between absolute certainty and complete ignorance lies a large space of well-supported but revisable knowledge.

Epistemic maturity is therefore not the absence of conviction but matching the degree of confidence to the quality of reasons.

Probability lets us reason responsibly under uncertainty

In many cases the question is not 'Has this been proven with absolute certainty?' but 'How probable is it given the information available?' Bayesian epistemology formalises the idea that beliefs can have different degrees — credences — and that those degrees should change in response to evidence.

Everyday reasoning does not require calculating Bayes' theorem in every conversation. The useful discipline is simpler: consider the prior plausibility of a claim, the quality of the new information, and how much more expected that information would be if the claim were true than if it were false.

This helps us avoid two common errors: allowing one dramatic datum to jump confidence straight to 100 percent, or rejecting it merely because it conflicts with our initial view. Good evidence changes confidence; it need not always create certainty.

Science does not become weaker when it admits uncertainty. Metrology attempts to quantify and express it. NIST emphasises that measurement is an experimental process and that a result carries uncertainty associated with method, instrument, calibration, sampling and other sources of variation.

This reveals an important distinction between uncertainty and ignorance. Saying that a value is estimated within a defined range with well-characterised uncertainty is not the same as saying 'we have no idea'. The ability to state the limits of a result is often a mark of better knowledge.

The same habit is useful outside the laboratory: if the data support an interval, do not report a point; if they support a probability, do not speak as though it were certainty; if we only know the direction of an effect, do not invent an exact magnitude.

Most of what we know is social: we rely on testimony

No one personally repeats every experiment, reads every archive or builds every instrument on which their knowledge depends. Much of what we know about history, medicine, astronomy, technology and other people comes through testimony — speech, books, articles, databases, expert institutions and documents.

This is not an epistemic defect but a necessary feature of human knowledge. The question is when reliance on testimony is reasonable. Relevant factors include not only a speaker's sincerity but also competence in the specific domain, access to information, independence, traceability of the claim and the possibility of checking it.

So 'I do not trust authorities' is not a sufficient epistemology — and neither is 'an authority said it, therefore it is true'. A better question is: what does this source actually know, how does it know it, and can we trace the path back to the underlying evidence?

A person may be an outstanding expert in one field and an ordinary or poor source in another. A respected institution may measure one thing exceptionally well while having no special competence regarding another. Trust is therefore better understood relationally: whom do we trust about what, and on what basis?

How to Evaluate a Source already offered a practical source audit. This article adds the epistemological layer: a source is not a substitute for an evidence chain. Where possible, we seek the primary document, method, dataset or original study to which a secondary source refers. For expert questions we also ask whether there is broader expert convergence and what quality of evidence supports it.

This also helps when sources conflict. Two opposing quotations are not epistemically equal merely because there are two of them. Access to data, domain expertise, methodology, independence and consistency with other evidence all matter.

Independent confirmation is stronger than repetition from one origin

Ten websites copying the same original mistake are not ten independent pieces of evidence. Likewise, a scientific study does not become independently confirmed merely because it has been cited many times. Citation counts do not replace verification.

This is why concepts such as replication, reproducibility, triangulation and convergence of methods matter. The National Academies and NIST emphasise that recomputation, procedural transparency and independent repetition can expose errors and increase confidence in results — although no single control makes a result infallible.

History has analogous forms of convergence: independent archives, documents produced by different parties, material traces and testimony that does not derive from the same chain. Methods differ, but the principle is similar: the more independent evidential routes converge, the less likely the conclusion depends on one shared error or interest.

When a competent person with access to similar evidence disagrees with us, it is reasonable to re-check our confidence. The philosophy of epistemic disagreement studies exactly how much a peer's contrary judgement should affect our beliefs.

But the mere existence of two sides does not imply a 50:50 probability. People may differ in access to data, expertise, methods, conflicts of interest or quality of argument. Popularity is not a guarantee of truth either. Disagreement is a reason for reassessment, not an automatic equalisation of all claims.

A useful question is: if someone with comparable competence and the same evidence reached the opposite conclusion, what would I need to re-check? That stance is stronger than either blind surrender to a majority or automatically treating every dissenter as deceived.

A good belief should know the conditions of its own correction

One of the strongest differences between inquiry and dogmatism is not that inquiry has no beliefs, but that it can say what would force revision. If no possible document, measurement, prediction or counterexample can affect a conclusion, the claim may not be framed in a way that evidence can seriously test.

This does not mean that every philosophical or metaphysical claim must have a simple laboratory test. It means we should state honestly what kinds of reasons support it and what would weaken that support. Where the answer is 'nothing', public evidential force should be calibrated accordingly even if the idea remains personally important.

Truth Sets Us Free — in Practice expressed this practically as willingness to revise a conclusion when better information arrives. Epistemic strength lies not in never being wrong, but in having a mechanism for detecting and correcting error.

Before an important conclusion we can run a short epistemic audit: 1. What exactly am I claiming? 2. Am I expressing belief, probability or knowledge? 3. What is the strongest evidence? 4. Is it direct or indirect? 5. Which alternatives could explain the same datum? 6. Are supporting sources genuinely independent? 7. How reliable is the source in this domain? 8. What is the uncertainty of the measurement or reconstruction? 9. What would lower my confidence? 10. Does my conclusion exceed the strength of the evidence?

These questions are not a recipe for perfect certainty. They are a way to align the strength of our language with the strength of our reasons. Sometimes the best answer will be 'I know'. At other times it will be 'very likely', 'there are good indications', or simply 'for now, I do not know'.

Where Does Fact End and Interpretation Begin? makes the next step more operational: where does fact end and interpretation begin? There we will separate observation, source, reconstruction, interpretation, hypothesis and speculation. This article leaves us with the foundational rule: belief should be as strong as the evidential route supporting it — and always flexible enough to change when the evidential picture changes.