Validity versus soundness

Scenario

Two arguments, side by side. Judge which one is the better piece of reasoning before reading further.

Argument One: All cats are mammals. All dogs are mammals. Therefore all cats are dogs.

Argument Two: All spiders are insects. All insects have six legs. Therefore all spiders have six legs.

Both conclusions are false: no cat is a dog, and a spider is an arachnid with eight legs, not an insect with six. So neither argument is trustworthy in the sense of delivering a fact you can rely on. But the two arguments fail for entirely different reasons, and the difference matters more than the shared verdict of "wrong."

Argument One fails at the level of connection. Even granting both premises fully (cats are mammals, dogs are mammals, both true), nothing licenses the leap from "cats and dogs share the category mammal" to "cats are dogs." Sharing a category is not the same as being identical to everything else in it: swap in any two other mammals and the same reasoning pattern produces the same false type of conclusion every time. The machinery of the argument is broken, independent of whatever facts you feed it.

Argument Two fails at the level of input. Its machinery is the reverse: take the general fact that all insects have six legs, and apply it to anything the first premise places inside the category "insect." If spiders really were insects, six legs would follow with total certainty; there is no gap in the reasoning to point at. The break is a single false premise, spiders are not insects, smuggled in before the machine even starts running.

One argument has broken machinery and could not be trusted whatever facts you gave it. The other has working machinery fed a false fact. Telling these two failures apart, and naming what each one lacks, is what the rest of this entry does.

Definition

Fix the vocabulary the scenario needs. An argument is a set of premises offered in support of a conclusion. Validity is a property of the argument's structure alone: an argument is valid when there is no way for all its premises to be true and its conclusion false at the same time. Validity says nothing about whether the premises are actually true. It says only that truth, if it is present in the premises, cannot fail to carry through to the conclusion.

Argument Two is valid by this test. Strip out the specific content and look at the shape: all A are B, all B are C, therefore all A are C (the classical "Barbara" form). Feed that shape any three categories you like, and there is no way to make both premises true and the conclusion false, because if every A is inside B, and every B is inside C, then every A is inside C as a matter of logical necessity, not biological fact. The form is airtight even though, in this instance, the first premise happens to be false.

Argument One is invalid by the same test, for a different reason. Its shape is: all A are B, all C are B, therefore all A are C. Here the shared category B (mammal) does not license any relation between A (cats) and C (dogs) at all. You can confirm the invalidity directly, without needing a definition, by finding a case with the identical shape and true premises that still produces a false conclusion: that is exactly what Argument One already is, since "all cats are mammals" and "all dogs are mammals" are both true and "all cats are dogs" is false. A single such case is enough to condemn the form permanently, because validity is a claim about every instance of a shape, and one counterexample breaks it for all of them.

This gives a derived fact worth stating on its own: a valid argument can have a false conclusion only if at least one of its premises is false. Run the contrapositive: if an argument is valid and every premise is true, the conclusion is guaranteed true, because validity is defined as the absence of any true-premises-false-conclusion case. So when a valid argument's conclusion turns out false, as in Argument Two, the premises cannot all have been true. Soundness names the case where they are: an argument is sound when it is valid and every one of its premises is true. A sound argument's conclusion is not merely likely, it is guaranteed, which is the whole reason the two-part test exists rather than one.

Laying the four combinations out removes any temptation to blend the two properties. An argument can be valid with a true conclusion (Barbara, fed true premises), valid with a false conclusion (Argument Two, a true form fed a false premise), invalid with a true conclusion (swap Argument One's categories so the conclusion happens to hold, the reasoning is still broken), or invalid with a false conclusion (Argument One as given). Soundness rules out exactly one of the four: valid-with-false-conclusion is impossible for a sound argument, since soundness demands true premises and a valid form guarantees a true premise set carries to a true conclusion. Only "sound" earns you the conclusion. "Valid" earns you nothing about the world at all, only a guarantee about the relation between two claims you have not yet checked.

Common mistakes

The three errors below all trade on the same shortcut: reading validity as a stand-in for correctness.

A valid argument has a true conclusion. False, and Argument Two is the counterexample sitting above: perfectly valid form, false premise, false conclusion. Validity guarantees only the conditional, that true premises would force a true conclusion; it promises nothing when a premise is false.

A true conclusion means the argument that produced it was good. Also false, in the other direction: an invalid argument can land on a true conclusion by coincidence, the same way Argument One would if its categories were swapped to ones that happen to overlap. A true landing spot does not repair a broken flight path, and treating it as vindication is exactly the mistake of judging reasoning by its luck rather than its structure. This is the one place the distinction is easiest to lose: an invalid argument's conclusion being true tells you nothing about whether the next argument built on the same broken form will be so fortunate.

"Valid" is used as a general compliment for a good argument. In ordinary speech "valid" often just means "reasonable" or "worth considering." In this vocabulary it is a narrow technical claim about form, with a precise failure condition (find one way to make the premises true and the conclusion false) and nothing to say about whether anyone should believe the premises in the first place. An argument can be valid and still not worth a moment's belief, because everything rides on premises nobody has checked.

Build with it

Construct two arguments of your own, on purpose, to demonstrate the distinction rather than just define it.

First, build a valid argument with a false conclusion. Pick a valid form (a Barbara-style syllogism works, or any chain you can confirm has no way to make the premises true and the conclusion false), then deliberately choose at least one false premise so the guaranteed conclusion comes out false. State the form in the abstract (all A are B, and so on) and confirm no substitution of true premises into that same form could produce a false conclusion, which is what makes it valid rather than merely wrong twice over.

Second, build a sound argument for a true conclusion, using the same or a different valid form, this time with premises you can actually defend as true.

Third, write one sentence stating exactly why the first argument is not sound. The correct sentence names the false premise specifically; a sentence that says "the conclusion is false" without locating which premise is false does not count, since that confuses the two-part test with a one-line verdict.

Success on this task looks like: a valid argument whose form you can defend against any substitution, paired with a false conclusion that follows from an identified false premise, plus a sound argument delivering a true conclusion, plus a one-sentence diagnosis of the first argument that names the false premise rather than the false conclusion. Anyone who can produce all three has separated the two tests for good, since the exercise only works by using the properties against each other rather than reciting their definitions.