The Faults That Have Names

A fallacy is a move that reliably persuades without supporting. Naming them is useful in the way that naming any recurring fault is useful: a thing with a name is easier to see, and easier to point at without having to reconstruct the whole case each time. Naming them is also dangerous, and the danger has become the larger problem.

Start with the useful part. The straw man is the replacement of an opponent's position with a weaker one, followed by a refutation of the weaker one. It usually happens with no intent to deceive, because the weak version is the one that was easiest to understand and hardest to believe, which is exactly the combination that makes something memorable. The discipline that prevents it is to state the opposing view in a form its holder would accept before saying a word against it. Doing so has its own name (the steel man) and it is a great deal rarer than the complaint about straw men.

Now the dangerous part, which is that several of the best-known fallacies are not fallacies in most of the cases where they are alleged. Take the attack on the person. Where a claim rests on somebody's testimony, their reliability is part of the evidence, and doubting a paid witness is not a lapse in reasoning but the whole point of cross-examination. Where the claim is a proof on a page, the author is irrelevant, because the page can be read. The label cannot tell you which situation you are in. Since almost everything anybody believes rests on testimony, assessing the people giving it is not a fault to be eliminated but a skill to be exercised well.

Appeal to authority is worse served by its reputation. Nobody has personally verified the speed of light, the safety record of the aircraft they board, or the pharmacology of the drug they swallow. Deference to relevant expertise is not a weakness in reasoning; it is the only way any person holds more beliefs than they have hours to check. The genuinely fallacious versions are specific and identifiable: a physicist pronouncing on nutrition, a single dissenting voice presented as though it were the consensus, an institution invoked in a domain where it has no standing.

The slippery slope makes the point most sharply. Asserting without evidence that one step leads to a catastrophic end is a fault. Predicting a chain of consequences with a stated mechanism (an incentive created, a precedent courts will follow, a threshold that is easier to lower a second time) is ordinary consequentialist argument, and is frequently correct. The two have the same shape. Only the presence or absence of the mechanism tells them apart, and only examining the case will reveal that.

Which brings us to the modern pathology. In online argument the vocabulary has become a way of ending conversations rather than advancing them: the name is said, and the naming is treated as the refutation. Two things are missing whenever that happens. The first is any demonstration that the move in front of you actually is an instance, which is where most accusations quietly fail. The second is the recognition that even a genuine fault removes only this support for the claim, leaving the claim standing and awaiting a better argument.

In a design review someone says a proposal is bad because its author shipped two failed products. The reflex is to say ad hominem and move on. The better reply is a question: which part of the proposal is the record evidence about? If the proposal rests on an unverifiable estimate from the author, past over-optimism bears on it directly. If it rests on a design that everyone in the room can read, the record is beside the point and the design is on the table. That question does the work the Latin name only pretends to do, and it leaves open the possibility that the objector is onto something and has expressed it badly, which, in meetings, is what is usually going on.