The case for using the term 'steelmanning' instead of 'principle of charity'

post by ChristianKl · 2022-06-02T19:24:40.583Z · LW · GW · 7 comments

Contents

7 comments

(Alternative title: “The case for rationalists inventing new jargon”)

Antoine Lavoisier revolutionized chemistry. Did he discover a new substance? If you think that names don't matter, Lavoisier didn't discover a new substance. If you think names matter, he discovered oxygen. 

Joseph Priestley discovered the same substance earlier. Through heating mercuric oxide he produced a gas that allowed a candle to burn brightly. A gas that enabled a mouse to live longer under an airtight glass cover. He called the gas dephlogisticated air.

He showed Lavoisier his experiment. Lavoisier rejected the idea that there should be elements with negative mass like phlogiston and called the same gas oxygen and earned thereby the reputation of inventing modern chemistry. 

The language you use to talk about something influences the way you think about it. If the chemistry you’re talking about is truly something new, then a fight over terminology may be quite an important part of getting to understand that chemistry better. Jay A. Labinger

Long before the discovery of atoms, Lavoisier reordered the ontology for chemistry and named entities in a way that reflected his new ontology thereby founding modern chemistry. 

In the field of rationality, we are today facing a situation where we lack an ontology that’s of the same quality as the ontology we have of chemistry in our times. We are facing a situation where the ontology of rationality is like the ontology of chemistry in Priestley’s time. 

Classical rationalism uses fallacies frequently in its ontology. In Kahnemann’s ontology, the concept of cognitive biases is central. Given CFAR's more recent work it’s fashionable to talk in terms of cruxes while talking about fallacies is less fashionable on LessWrong. New concepts come with new terms and the way we name concepts helps us to relate to our new concepts.

Positivist ideology suggests that non-physical entities that we might talk about in rationality such as biases don’t really exist. Positivists propose that after we get a good ontology for physical entities like atoms, we don’t need to think about ontology anymore and can just focus on epistemology. 

This focus on epistemology leads to us spending less effort on finding good names for concepts and good ontology. When new names for concepts that are already known under other names get introduced those new names often face criticism. 

An example of great naming of our community is the term steelmanning which refers to a concept that’s called the principle of charity in academic philosophy. Jonathan Maloney defines the principle of charity of the philosophers as:

The Principle of Charity demands that one interprets a speaker's statement(s) in the most rational way possible. In other words, when ascribing to this principle, you must consider the strongest possible interpretation of your fellow interlocutor's argument before subjecting it to evaluation.

Academic philosophers borrowed the term from the philosophy of language, where it has a slightly different meaning. Wilson coined the term in his paper Substances without Substrata to deal with the problem of knowing to whom a term refers. If Charles speaks of Caesar and there are multiple people who are named Caesar, Wilson’s principle of charity is a heuristic for knowing which Caesar is meant. According to the heuristic that Julius Caesar is meant by Charles when picking Julius Caesar as designatum for Caesar makes the most claims that a person makes about Caesar to be true. 

We select as designatum that individual which will make the largest possible number of Charles’ statements true. 

In Wilson’s definition, the principle of charity has the goal of figuring out what another person means with the languages they use. This goal is different from improving a poorly argued argument another person made to a stronger form. The concepts that Wilson and Maloney describe differ.

Both of their concepts are about how to interpret a speaker's statement and not about improving on a speaker's statement. Often when we talk about steelmanning in our community we want to improve on the argument that was made in a way that adds reasoning to the case the speaker makes even if the speaker doesn’t know the reasoning we are adding. We speak about steelmanning when we care more about whether the position for which a speaker is arguing is true than caring about whether it’s true for the reasons that the speaker advocates. 

The common idea of a strawman is a way to phrase an argument in a weaker way than it was meant and in contrast to the term strawman the term steelman indicates that the steelman also goes beyond the original meaning. Using the term steelman helps us to be clear that we are indeed attempting to change the meaning.

The philosophers don’t have distinct terms about whether they include or exclude the change in meaning when they speak about using the principle of charity and 

The term principle of charity also comes with additional baggage because the idea of charity is often about assuming good intentions, being respectful, and not hurting other people's feelings. By using the term steelman for the concept Maloney named principle of charity we have a word that maps more directly to a single ontological concept instead of mapping to a mix of different concepts. 

Terms like steelman allow us to speak more effectively about rationality than we can by reusing existing terms. Finding good terms for concepts can help us refine our field of rationality.

7 comments

Comments sorted by top scores.

comment by jessicata (jessica.liu.taylor) · 2022-04-03T15:08:36.053Z · LW(p) · GW(p)

From Wikipedia:

A steel man argument (or steelmanning) is the opposite of a straw man argument. The idea is to help one's opponent to construct the strongest form of their argument. This may involve removing flawed assumptions that could be easily refuted, for example, so that one produces the best argument for the "core" of one's opponent's position.[19][20] It has been advocated as a more productive strategy in political dialog that promotes real understanding and compromise instead of fueling partisanship by discussing only the weakest arguments of the opposition.[19]

This implies that the target of steelmanning is an opponent position. The idea is to show that you can definitely defeat the actual opponent position, by, conservatively, defeating the strongest possible form of it you can think of.

Charity, on the other hand, doesn't imply that the target is an opponent position; it also doesn't imply that you intend to "defeat" the resulting position. It's, instead, implying something like that you intend to interpret the statements such that they are most generally valuable.

I wrote earlier on this topic:

nooo, don't "steelman" a position and then fail to knock down the steelman, that defeats the purpose, you have just put some steel armor onto a thing that seemed bad to you initially, aaa

Replies from: Celarix
comment by Celarix · 2022-06-03T03:01:46.923Z · LW(p) · GW(p)

If you steelman a position and can't knock it down, that indicates that you may be wrong about your point, which, IMO, is valuable. Recognizing error in ourselves has a much higher return than recognizing error in others.

comment by Vladimir_Nesov · 2022-04-03T14:36:38.027Z · LW(p) · GW(p)

There is also a useful concept of charity that Scott Alexander occasionally gestures at, which is more about maintaining curiosity about explanations for others' behavior/claims/beliefs, not settling for implausible caricature stand-ins that don't offer a technical understanding of how certain behavior actually came to be. This is distinct from steelmanning, which is about using observed behavior/claims/beliefs as inspiration for engineering better beliefs/understanding than what is probably behind that behavior/claims/beliefs in reality. Both activities are worthwhile.

Rationality is about methodologies for good/reliable/accurate/useful thinking, so in these terms it's more about steelmanning than charity, though lack of charity is a flaw of curiosity, so it's also rationality's business to appeal for reasonable attention to charity.

Replies from: Pattern
comment by Pattern · 2022-06-04T00:10:49.759Z · LW(p) · GW(p)

lack of charity is a flaw of curiosity,

What?

Replies from: Vladimir_Nesov
comment by Vladimir_Nesov · 2022-06-04T09:20:31.285Z · LW(p) · GW(p)

If charity is taken to mean curiosity about reasons for others' claims/behavior, as I specified in this thread, then lack of charity is a systematic failure to pay some attention to figuring out those reasons. Curiosity is liveness of figuring things out, a rejection of not making progress on any of its subjects. Healthy curiosity keeps the investigation of all relevant topics going, and the actual reasons for someone's ridiculous claims/behavior are relevant to engaging with them.

comment by Keller · 2022-04-03T15:13:49.809Z · LW(p) · GW(p)

These are two different concepts. They're both useful ones. Interpreting an argument charitably has positive arguments in its own right. It can promote healthier conversational styles and more effective reasoning. Actually understanding what your counterpart is trying to say is difficult, and that is why Intellectual Turing Tests are non-trivial to pass and valuable to engage in.

Steelmanning is useful for changing your own mind, but it is at best rude to the person you are in conversation with, and can as a practice close off more substantial shifts in belief. The argument most likely to change my mind is going to tend to be the one that produces the smallest update, since that needs to conflict with fewer other beliefs about the world.

We should keep both terms, not crush the principle of charity entirely.

Replies from: ChristianKl
comment by ChristianKl · 2022-04-03T15:22:22.231Z · LW(p) · GW(p)

Yes, the principle of charity has its uses. When there are multiple possible meanings in what someone wrote reading it in the most charitable way makes sense.