Truth, Consensus, and Disagreement

Is something true because everyone agrees with it, or do people agree because it is true? This question — the relationship between truth and consensus — is not merely academic. It has immediate consequences for how you navigate scientific disputes, political disagreements, and the flood of conflicting claims that define information-age life. The answer turns out to be: consensus can be strong evidence for truth, but the two are not the same thing, and knowing when to trust consensus — and when not to — requires careful epistemic work.

Theories of Truth

Philosophers have proposed several accounts of what truth is. The correspondence theory holds that a proposition is true if and only if it corresponds to a fact about the world. 'Water is H2O' is true because there is something in the world — the molecular structure of water — that the proposition correctly describes. This is the most intuitive and widely held view: truth is about getting the world right. The coherence theory holds that a proposition is true if it coheres with a body of beliefs — if it fits consistently with everything else we believe. This view is more common in mathematics and formal systems, where 'truth' is relative to an axiom system. Pragmatist theories hold that truth is what works — what is useful for navigating experience and achieving goals. A proposition is true if believing it reliably leads to successful action. This view is associated with philosophers William James and John Dewey and has been influential in AI research, where systems are often evaluated by their practical performance. For most empirical questions — science, history, current events — the correspondence theory is the operative standard. A claim is true or false based on what the world is actually like, independent of what anyone believes. This independence of truth from belief is crucial: the Earth was not flat even when virtually everyone believed it was.

If truth is independent of consensus, why does consensus matter epistemically? The answer is that consensus, when properly structured, is powerful evidence about truth — not the same thing as truth. Scientific consensus is the convergence of expert judgment after extended evaluation of shared evidence under conditions designed to minimize bias. The consensus on climate change, evolution, vaccine efficacy, and the age of the universe does not make these things true — they are true (or well-justified) because of the evidence. The consensus is evidence that the evidence has been evaluated carefully and repeatedly by many independent parties and consistently points in the same direction. This matters enormously in practice. You cannot personally evaluate all the evidence for evolution, climate change, or quantum mechanics. But you can evaluate the quality of the process that produced the consensus: Did it use peer review? Has it been replicated? Are those who agree independent of each other? Have competing hypotheses been seriously tested and ruled out? A consensus that passes these process checks is strong evidence — even if not certainty. But consensus can also be wrong, and the history of science is full of examples. Twentieth-century physicians widely believed that stomach ulcers were caused by stress; Barry Marshall and Robin Warren demonstrated in 1984 that most are caused by the bacterium H. pylori — a claim initially dismissed by the consensus. The correct epistemic response is not to distrust consensus but to understand what makes consensus reliable and to remain appropriately open to well-evidenced challenges.

Types of Disagreement and What They Mean

Not all disagreement is the same. Distinguishing types of disagreement is critical for knowing how to respond. Empirical disagreement: disagreement about what the facts are. This is in principle resolvable by gathering more data, running better experiments, or examining primary sources. When two scientists disagree about whether a drug reduces mortality, more rigorous trials can settle the question. Methodological disagreement: disagreement about the right procedures for investigating a question. Two economists may use different models of the same data and reach different conclusions. The disagreement is not about the data but about what counts as valid inference from it. This is harder to resolve but often amenable to formal comparison. Value disagreement: disagreement about what we should do or what matters. Even if two people agree on all the facts about a policy's effects, they may disagree about whether those effects are worth the costs — because they weight different values differently. No additional evidence can resolve this type of disagreement; it requires negotiation and compromise. Deep disagreement: disagreement so fundamental that the parties do not share enough common ground to adjudicate the dispute through standard means. Arguments about first principles — whether human life has intrinsic value, whether there are natural rights — often cannot be resolved by evidence because the disagreement is about what counts as evidence. Misidentifying the type of disagreement leads to bad epistemic moves: treating a value dispute as if more data will resolve it, or treating an empirical question as if it is merely a matter of values that no evidence can touch.

AI introduces new dynamics into epistemic disagreement. On one hand, AI tools can accelerate empirical research — running analyses, reviewing literature, identifying patterns — in ways that could help resolve empirical disputes faster. On the other hand, AI systems trained on large text corpora inherit the biases and errors present in that text, and when many AI systems are trained on similar corpora, they tend to reproduce the same distributions of claims. This creates a new epistemic risk called epistemic homogenization: if millions of people route their questions through AI systems that all learned from the same underlying data, the diversity of perspectives and interpretations in circulation may narrow. Historically, disagreement — even wrong disagreement — has been epistemically productive. It forces defenders of positions to articulate their evidence, tests arguments against the best counterarguments, and occasionally produces a minority view that turns out to be correct. A world where AI flattens all queries toward a single mode of response may be epistemically impoverished in ways that are not immediately visible.