Evidence, Justification, and Belief

In Lesson 1 we established that knowledge requires justification — not merely true belief, but true belief held for the right reasons. Now we go deeper: what counts as a good reason? What is evidence, precisely? How does evidence connect to justification, and how does justification determine whether our beliefs are rational? These questions sit at the heart of both classical epistemology and the daily problem of navigating an information environment saturated with AI-generated content.

What Is Evidence?

Evidence is information that raises or lowers the probability that a proposition is true. This probabilistic framing, developed formally by the mathematician Thomas Bayes in the eighteenth century and extended through the twentieth, captures something important: evidence is not binary. It does not simply confirm or refute — it shifts probabilities. Suppose you believe a patient has a rare disease. A positive test result is evidence for the disease. But how much evidence depends on the test's sensitivity (how often it correctly detects the disease when present) and specificity (how often it correctly reports no disease when absent), as well as the disease's base rate in the population. A test that is 99% accurate can still produce a result that is more likely false than true if the disease is rare enough. Evidence has to be weighed in context. Strong evidence is information that significantly changes the probability of a proposition. A single anecdote ('my aunt smoked all her life and lived to 95') is weak evidence about smoking's health effects — it shifts population-level probability estimates barely at all. A randomized controlled trial with 50,000 participants is strong evidence. The difference is not that one 'feels' more convincing — it is that one genuinely tells us more about the world.

Not all evidence is equally trustworthy. Epistemologists distinguish between the source of evidence and the content of evidence. First-person evidence comes from your own direct experience: you felt the heat, you watched the reaction, you measured the voltage. This is often the most trustworthy kind — you have direct access to the datum. But it is also limited by your perceptual fallibility, your cognitive biases, and the impossibility of experiencing everything. Testimonial evidence comes from reports by others. Most of what any modern person knows is testimonial — you know vaccines work not because you ran the trials, but because thousands of scientists did and reported their findings through peer-reviewed journals. Testimony is epistemically powerful and practically unavoidable, but it chains your justification to the reliability of your sources. Evaluating testimony well is one of the most important intellectual skills of our era. Instrumental evidence comes from measurement devices — thermometers, seismographs, MRI machines, satellite sensors. Instruments extend our senses but introduce their own error profiles: calibration errors, systematic biases, breakdown. Statistical evidence aggregates many observations. A single coin flip is no evidence about fairness; ten thousand flips with 6,300 heads is strong evidence the coin is biased. Statistical evidence requires understanding of sample size, selection, confounding, and the difference between correlation and causation.

Justification: Internal and External Accounts

What makes a belief justified — what gives you the right to hold it? Philosophers have developed two broad camps. Internalism holds that justification depends entirely on factors internal to the believer's mind — on reasons, evidence, and reasoning processes that the believer is aware of or can become aware of. If you hold a belief for reasons you could articulate and defend, your belief is internally justified. If you hold it on grounds you cannot access, it is not — even if those grounds happen to be reliable. Externalism holds that justification can depend on external factors the believer need not be aware of. The most influential externalist view, reliabilism, says a belief is justified if it is produced by a cognitive process that reliably produces true beliefs. You do not need to know why your perception is reliable — you just need it to be. Under reliabilism, a dog that reliably detects gas leaks through smell has justified beliefs about gas leaks, even though the dog cannot articulate any reasoning. The debate matters practically: an AI system that reliably produces true outputs counts as justified under reliabilism. An internalist would deny this — the system has no accessible reasons, no capacity for self-examination. Choosing between these views has real consequences for how you evaluate AI-generated claims.

Neither pure internalism nor pure externalism fully captures our intuitions about justification. A middle path — responsibilism — focuses on epistemic virtues: the character traits and intellectual habits of a good reasoner. Intellectual humility, open-mindedness, thoroughness, consistency, and willingness to revise beliefs in light of evidence are epistemic virtues. Epistemic vices — closed-mindedness, overconfidence, sloppy reasoning, motivated skepticism — undermine justification regardless of whether your outputs happen to be true. This virtue-epistemology framing is useful when thinking about AI: a well-designed AI system might exhibit some epistemic virtues (thoroughness in retrieving relevant information) while systematically lacking others (genuine open-mindedness, willingness to say 'I do not know'). Identifying which virtues a system has and lacks helps you use it well.