Why Do Real Disagreements So Often Begin with the Same Word?

Geoffrey Chen

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

Two people argue about whether a workplace arrangement is “fair”. One says that fairness means applying the same rule to everybody. The other says that when people begin in different circumstances, an identical rule can produce unfairness. After a long conversation, they discover that they differ not only in their evaluation of the arrangement. They are using fair to select different kinds of comparison.

Two hasty conclusions commonly follow. The first says, “So it was merely a difference in definition; there is no real disagreement.” The second says, “My definition is the correct one; you do not even understand the word.”

Either can be mistaken. People sometimes do use different senses and find that the dispute dissolves once the usage is explained. At other times, contest over a word's use is part of the substantive disagreement. Who counts as a worker? What qualifies as a risk? Which activities constitute care? What system deserves to be called intelligent? The boundaries of terms can alter statistics, responsibilities, resources and self-understanding. Definitions do not sit harmlessly on the surface of reality. They help organise what can be seen in it.

Three possible disputes need to be separated

When one word becomes the focus of conflict, at least three situations should be distinguished.

First, the parties may point to the same facts while following different conventions of usage. One person calls a drink “hot” only above a certain temperature, while another draws the line slightly lower. Once the temperature and practical preferences are explicit, continued battle over the word may have little value.

Second, the parties may use approximately the same sense but classify the facts differently. They agree that deception involves intentionally causing a false understanding, but disagree about whether a particular omission had that intention. A dictionary cannot settle this. Evidence is needed.

Third, the parties may be arguing about how the word ought to be used because competing uses have normative consequences. Calling an activity “work” can bring it within systems of payment and protection. Describing an outcome as “harm” changes questions of responsibility. Calling a system “autonomous” may weaken perceived responsibility on the part of designers and users. Such disputes concern language and also the public structure we are prepared to build.

In his work on verbal disputes, David Chalmers proposes a useful diagnostic: temporarily forbid the disputed expression and ask whether the parties can state a residual substantive disagreement in other terms. His paper “Verbal Disputes” develops this method in detail. If removal of the expression makes the dispute disappear, it may be substantially verbal. If different factual, evaluative or practical claims can still be stated, the word was obscuring a deeper issue.

The method does not declare linguistic questions unimportant. It separates the tasks so that participants can tell whether they are establishing a fact, selecting a classification or contesting the social consequences of a classification.

A definition sometimes manages a boundary

Dictionaries record usage. They cannot determine every boundary required by every institution. Law, medicine, statistics and public policy need operational definitions because reality changes continuously while administration must decide inclusion and exclusion. Definitions therefore have purposes.

“Unemployment”, for example, can refer in ordinary life to lacking adequate work and income. In labour-force statistics it is a classification governed by executable conditions. The definitions answer different questions. The error is not their difference. It is treating the statistical classification as a complete description of a person's situation.

Likewise, when a community debates who counts as a member, it is not merely discovering the natural essence of a word. It is deciding who may speak, who receives protection and who incurs obligations. When a technical team decides what counts as an “explanation” from an AI system, it affects whether a user receives reasons, variable importance or simply a plausible paragraph.

David Plunkett and Timothy Sundell use metalinguistic negotiation for a class of exchanges in which speakers appear to make contrary statements using the same evaluative term while also advocating different ways of using it. Their paper on disagreement and normative or evaluative language shows why a dispute about use is not necessarily trivial. A proposed meaning may also be a proposal about how the world should be divided.

What matters is making the proposal visible. Instead of “This is the one true meaning”, a speaker can say: “I favour this usage because it reveals these differences, protects these people and supports this institutional purpose; its cost is that it may obscure those other differences.” Attached to reasons, a definition becomes open to comparison rather than a struggle for dictionary authority.

A shared word can create false consensus

Disagreement does not always sound like conflict. A more dangerous possibility is that everyone uses the same word and assumes they agree.

An organisation promises to improve “quality”. Management means fewer errors. Customers mean easier access to help. Employees mean enough time to complete the work. The policy shows strong consensus while implementation produces conflict, because the common word concealed different objects of measurement.

“Safety”, “innovation”, “transparency” and “human-centred” can all produce this false consensus. They have positive value, so few participants want to oppose them openly. For the same reason, each participant can fill the term with different content. A meeting quickly reaches agreement in principle. The conflict reappears later when resources are allocated and responsibility is assigned.

The most useful test of a shared expression is often not another abstract definition but examples and boundaries. Which specific case qualifies, and which adjacent case does not? When two goals conflict, which takes priority? What evidence would show that the promise has been met? Who gains or loses under this definition?

Examples return a term to reality. Borderline cases reveal its hidden criteria. Questions about consequences prevent values inside the definition from remaining invisible.

AI makes meanings more fluent and easier to mix

Ask a generative AI “What is fairness?”, “What is consciousness?” or “What is risk?” and it will usually produce a smooth and balanced definition. It may combine ordinary usage, academic theories, policy language and common online expression. Synthesis can be useful for orientation. It can also place incompatible senses in a single paragraph, producing text that looks complete but cannot support a determinate judgement.

AI also adjusts meanings to conversational context. The same word can acquire different qualifications across prompts without the user noticing that its boundary has moved. Continuity of language tempts us to believe that continued use of one word guarantees continued discussion of one concept.

On an important question, the first definition should therefore be followed by further questions. Is this ordinary, legal, technical or philosophical usage? What purpose does it serve? Which influential alternatives exist? How would adoption of another definition alter the conclusion in this case? Which sources support each usage?

AI can generate a map of concepts. It cannot decide for a community which boundary should carry practical consequences. A language model can display usages; a normative choice still requires people to give reasons and face questions from those affected.

Resolving a dispute about words does not begin by choosing a winner

When an argument repeatedly circles one term, a more patient sequence helps.

First remove the expression temporarily and restate each claim in concrete sentences. Then list separately the factual disagreement, the evaluative disagreement and the choice of language. Compare candidate definitions across typical and borderline cases. Finally state what the current setting actually requires: ordinary understanding, technical measurement, legal executability or moral evaluation.

This process does not guarantee consensus. People who understand one another fully may continue to hold different values. But the disagreement becomes more honest. “You do not understand the word” no longer conceals “I want the world classified in this way.”

A common language is not a finished container. It is a public structure maintained through use, contest and revision. Words enable a world to be shared and can flatten its differences. They preserve inherited judgements and may carry old boundaries into new circumstances.

Real disagreement often begins with the same word not because language has interrupted reality, but because language is already participating in it. To build a world is partly to learn which terms reveal differences, when to resist a ready-made classification and how to let others inspect the reasons for one's naming.

The best outcome is not always a single final definition. More important is knowing what work the word is doing—and being willing to answer for the boundary it draws.

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.