Why and How to Name Things

Mini sequence: >30min of rationality writing per day for 30 days (21/30)

Related to: "Focusing," for skeptics, Buckets and Bayes

There's a lot of discussion in this subculture about jargon. About the cost of creating and disseminating new terminology; about the problems with having a high barrier-to-entry; about the value of having precise handles for subtle concepts.

I had a friend who once attended the CFAR alumni reunion as somebody's plus-one (he wasn't an alumnus himself). Afterward, I asked him for a download of his thoughts and impressions (of the people, the concepts, the activities, the culture). He paused for a moment, and then said "So, it sounds like eighty percent of what they do is just naming things."

Contrast that with the perspective of Syme, from Orwell's 1984:

"Don't you see that the whole aim of Newspeak is to narrow the range of thought? In the end we shall make thought-crime literally impossible, because there will be no words in which to express it. Every concept that can ever be needed will be expressed by exactly one word, with its meaning rigidly defined and all its subsidiary meanings rubbed out and forgotten. . . . Every year fewer and fewer words, and the range of consciousness always a little smaller. Even now, of course, there's no reason or excuse for committing thought-crime. It's merely a question of self-discipline, reality-control. But in the end there won't be any need even for that."

It's my opinion that the process of expanding our vocabulary—drawing distinctions between things which might otherwise be conflated, drawing attention to things which might otherwise go unnoticed—is absolutely central to the project of rationality. By this I do not mean to imply that it's costless—only that I think it's so high-value that it's worth it under any reasonable projection of its total cost over time.

What you lose by the proliferation of jargon is ease-of-entry and cross-cultural intelligibility and hard-drive space in the brains of people trying to track all of it. But what you gain by it—

  • A dramatic reduction of the amount of time required to communicate complex concepts, such that instead of having to explain that [I suspect Alice is employing a strategy which does indeed solve the immediate problem, but only by switching a bit in the algorithm and replacing one inflexible rule with another, and that this means that we've swung past the ideal and will start to get new (though probably lesser) problems emerging from the new policy], I can just say "yeah, but pendulums."

  • A dramatic increase in people's ability to distinguish between concepts that are near to one another in ideaspace, analogous to having words like cyan, cerulean, cornflower, periwinkle, navy, royal, and midnight rather than just blue. In addition to providing direct value à la knowing-whether-to-ask-and-also-being-able-to-answer whether a particular thought represents an alief or a belief (and thereby how best to debias it), or knowing-that-there's-a-difference-between-and-also-deliberately-choosing between exposure therapy or comfort zone expansion, being embedded in a culture where people are diligently seeking out and popularizing such distinctions makes a given individual far more likely to pay attention to subtle distinctions themselves, accelerating the process of cultural accumulation of nuance and detail.

  • A dramatic increase in people's ability to think explicitly and carefully about specific phenomena. There's a reason people twitch when someone says "So, the Schelling point if you get lost is over by the chapel," and it's because Schelling point means something, and until we had the phrase a lot of us never saw that thing clearly, and once we had the phrase there were suddenly a bunch of new questions and new models and new ideas and new experiments to run and new threads to pull on, and the degradation of the term back into a mere synonym for concepts we already had is an actual loss. When you define a concept rigorously and clearly, you almost always learn new things from playing around with their edges and trying to get them to interface with other rigorous, clear concepts.

To put this another way, I'm fairly certain anyone reading this could identify ten or twenty concepts that they've found utterly useless, and would not mind having erased from their mind, to free up space. But if I started threatening to erase a hundred concepts, or a thousand, you'd push back—because those concepts are you. They represent the lessons you've learned, the experience you've gathered, the thoughts you are able to think.

And those ten or twenty that you found useless are super valuable to someone. They're the cost that we pay, to live in a society where our maps are constantly expanding and constantly gaining resolution—the fact that some of that resolution is irrelevant to you doesn't mean that, on the whole, we should reverse our policy of trying to gain more. This is one of those situations where the liberal, optimistic view is something like "we can reduce all these costs and keep all these benefits, if we just think hard and keep trying," and I subscribe to the conservative, pessimistic view that you can't reduce all the costs to zero, and this is a dear but worthwhile purchase.

* * *

So that's the why. Now for the how.

In a comment response earlier today, user Raemon registered the following hesitation:

I want to be able to link people to this post, and use it as a handle for things, but if I bring it up in conversation and I have to start with "Did You Read Conor's thing on Theodicy?" I have to explain Theodicy before I get to the meat of the question, and if I start with "Did you read Conor's thing on the Hierarchy of How To Interpret People Without Falling Prey to the Fundamental Attribution Error...", well...

In particular, the hesitation was surrounding my implicit claim that theodicy is the right handle to use for this concept.

In a separate comment, Raemon laid out some desiderata for post titles (which doesn't strictly correspond to claims about terms and jargon generally, but is at least partially illuminating):

a) [A title should] contain the handle for each major concept in the post (sometimes this means the post should be broken up if there are too many concepts)

b) [A title should] be easily distinguishable from other similar posts and similar concept-handles

c) [A title should] be a handle that makes it fairly easy to remember the rest of the post if you've read it at least once (and ideally, point you towards the right thing if all you read was the headline, although I admit it's sometimes better to leave people clueless than with a false guess as to what the post is about [BUT it's even worse to have a name that people think points clearly towards a thing when it doesn't])

Raemon was interested in both whether people would be able to rederive concepts from their handles, and also whether the handles would be memorable in the sense that they would actually facilitate surfacing the concept at the moment when the concept is needed.

I have a model of where names come from (/what makes for a good name) that I suspect overlaps with but is not Raemon's model. I claim that he's right to want to optimize for things like resonance and memorability, but that he's too cost-oriented. In particular, I think this because I believe Raemon's applying the rules for one category of Good Names to another, different category that has its own rules.

* * *

The first category is something like "analogies." Names based on analogy are leaning on our prior experiences to tell us that New Thing Is Kind Of Like Old Thing, But Different. General Relativity was exciting because it leaned on Special Relativity, only now it was promising to apply generally. Aversion Factoring (a CFAR class) is a good name because it clues people in to the idea that this will be like Goal Factoring, except we're going to do that Goal Factoring thing on our aversions.

There are analogies which stretch a good deal further. Raemon cited "Reliability: Prophets and Kings" and "Idea Inoculation" as post titles which he thought were particularly strong when compared to others in my portfolio. "Prophets and Kings" was a resonant handle for the thing I was pointing to because we all already know that prophets see the future and kings have power—I was saying "you can be reliable because you're like a prophet or you can be reliable because you're like a king," and this was a direct shortcut to the concept I was talking about, because you can just copy-paste all of the causal arrows. Similarly, idea inoculation is a handle which says "you know that thing that happens with vaccines or childhood illnesses? Think about that same thing happening, but with memes instead."

I pretty much entirely agree with Raemon about the properties which make an analogy name good. You want it to be unique, such that it doesn't clash with other nearby concepts (unless you're deliberately trying to show a connection, such as with special and general relativity). It's not great that we have, in our community, popular activities named Hamming Circles, Doom Circles, and Circling (which has a special subset called Birthday Circles), all of which require 3-10 people sitting down in a circle.

You also want the name to be resonant or evocative, such that someone can sort of reason out the concept from first principles. If I ask somebody to factor their problem into parts, this will probably result in them doing something reasonably close to what I had in mind even if they've never heard of Goal Factoring, since most people learn simple mathematical factoring by middle school and the analogy is fairly apt. If I ask someone to "make quiche," though (which is in fact a phrase that's thrown around sans explanation in at least part of the community), there's not much for a newcomer to grab on to.

And lastly, you want it to be short and memorable, such that it interfaces well with human search-and-retrieval systems. There is a published research paper whose title is "The nucleotide sequence of a 3.2 kb segment of mitochondrial maxicircle DNA from Crithidia fasciculata containing the gene for cytochrome oxidase subunit III, the N-terminal part of the apocytochrome b gene and a possible frameshift gene; further evidence for the use of unusual initiator triplets in trypanosome mitochondria," and this is unlikely to be a handle that fits easily into a normal person's concept web. "The Control Group Is Out Of Control" is a far better title on this axis, even though it has seven words (in fact, it's almost self-rederivable in the sense that, if you remember the first three words, you're likely to be able to regenerate the last four from your felt sense).

But what about "Moloch"?

* * *

"Moloch" is a handle that is near-ubiquitous in use within our community in my experience, and which has also taken root in related other communities such that I've had to explain it far less often than I would've naively predicted. It's also a concept that comes up a lot, so the savings of having it are obvious.

But Moloch is not an "analogy" name. It started that way—it was an analogy for Scott in the original post where he coined the current usage.

But—I wager—something like 95% of the people who now frequently and comfortably use the term had basically no referent for the word "Moloch," prior to reading his post. Of those who did have meaning attached to the term, I'm willing to bet the breakdown was fairly even between people who thought of Ginsberg's poem, the Canaanite deity, and the minor Watchmen villain who came to Edward Blake's funeral to pay his respects.

"Moloch" is still a good handle for its brevity, its uniquity, and its general phonetic qualities. But what actually happened in the post was that Scott told people a thing in great detail, and then, both during and after the explanation, said "Call that thing Moloch."

Theodicy is similar. While it literally means "justifying god," most modern users of the term lack the familiarity with ancient Greek that Gottfried Leibniz likely leaned on when he coined it in 1710. For a modern altar boy or seminary student, "theodicy" is a technical handle attached to an explanation. Leibniz's original essay was titled "Theodicy: Essays on the Goodness of God, the Freedom of Man and the Origin of Evil." Or, in other words:

"You know that thing where God is supposed to be all good and all powerful and all knowing, but also there's evil in the world, and that doesn't immediately make sense to us? Yeah—that's the problem of theodicy. We're calling it that now; I get to name it because none of you came up with a word first. Yeah, just memorize it. Yes, for real—people are going to use it and you need to know what they mean by it."

That sort of power move doesn't always work; the concept of implementation intentions has been around since Peter Gollwitzer distilled it in 1999, but we all call them TAPs (trigger-action plans) because of CFAR (and because of shortness/stickiness). CFAR instructor Duncan Sabien has been trying to make the word "coferences" be a thing for a while, with mixed success. And anyone who's seen "Mean Girls" knows that "fetch" is not going to happen.

But there are times when it is the right move to make. In particular, if there isn't a decent and resonant analogy, and you're going to be reduced to a coldhearted technical description anyway, it's often worth the cost to say "and the shorthand for this coldhearted technical description is the Flynn effect."

This does nothing to reduce barrier-to-entry. It does nothing for helping people rederive the concept in question. But it allows those who need to deal with the effect (and lots of other complicated concepts along side it) to save time and energy, and honestly—what sort of analogy could you use that would be rederivable and resonant without losing clarity? I claim if it were easy to find one, it would have already been found—while there might be a snobbish scientist or two who wants to feel above everyone else via impenetrable jargon, most experts are more than happy to use resonant, transparent language if it doesn't cost them time and precision.

When I wrote my post yesterday, I checked to see if I could find any accessible, tropey, analogous handles. I failed (which is not to say that there are none, just that none occurred to me). I then had the choice of giving the post some horribly accurate title like "Operationalizing the Good Faith Principle," or of giving it a technical handle.

(By which, just to reiterate, I mean "syllables you simply have to memorize as the filename for this document.")

I could've called it "spronking" or something, which is a path that people do sometimes take (remember "googol"?). But in this case, I thought I could at least reduce the sheerly technical nature of the term by drawing an obscure analogy, as Scott did when he labeled his race-to-the-bottom, multipolar-trap concept after Ginsberg's Moloch.

The underlying implicit claim is that there was no analogy that would work for a broad swath of people, so the right thing to do was to choose something appropriate and then just turn it into a teachable moment. Theodicy is "actually" the right term, according to me—it's the word that is the closest match for the thing I wanted to gesture toward. The fact that lots of people don't know that word is a shame (I learned it from the SMBC comic linked in the post, myself), but that doesn't mean the word itself should be thrown out. Everybody's gotta learn new words sometime, after all—you can't continually enrich your map unless you're willing to pay the price.

And for people who are going to say "but but but why not just say the thing in a sentence?" the claim is "yeah, but then you just wouldn't remember it at all, any more than early CFAR participants remembered 'implementation intentions.'"

(Edit: I notice an implicit contradiction that doesn't feel like a contradiction from the inside, à la "but isn't implementation intentions just a technical handle?" The resolution comes from a) it's a technical handle, but it was presented as something that should seem resonant, and failed to meet that bar, and b) TAPs was an even better technical handle.)

The proposed hierarchy, then:

  1. Resonant, tropey analogy names

  2. Short, unique technical handles

  3. Phrase- or sentence-long accessible descriptions

It is a Thing That Is True About Humans, that we often won't bother to hold on to a concept at all unless it has a name (cf. feminism, intersectionality, privilege, microaggression). I interpreted Raemon as making entirely valid requests for better names, but as being inaccurately optimistic about whether there were any such better names out there at all. I agree that "Theodicy in Humans" is not in category 1, but it's a step up from category 3, and I think that the category 2 class that it represents is one we should dignify and respect.

18 comments, sorted by
magical algorithm
Highlighting new comments since Today at 11:32 AM
Select new highlight date
Moderation Guidelinesexpand_more

So, a few thoughts on the example of "theodicy", about which I'm afraid I feel roughly as Raemon does though not necessarily for the same reasons.

For me, "theodicy" already has a meaning firmly attached, and it does kinda map to part of what you're describing in your post with "theodicy" in the title. Unfortunately, that part -- so it seems to me -- already has a perfectly good name: "the principle of charity". PoC says to interpret what people say or do in the way that puts them in the best light, and that's exactly what you're describing.

The extra thing you're adding to that already-existing concept, the thing that (if I understand right) you want the name "theodicy" attached to, is a particular notion of what counts as putting someone in the best light: treat them as (in order of priority) good, smart, perceptive, currently-competent, and put all those ahead of assuming your observations were accurate and representative. And the only connection between this "hierarchy of innocuousness" and the concept usually known as "theodicy" is that there's this one post where you talk about being charitable to people and analogizes it to theodicy, and in that post you propose this specific hierarchy of innocuousness.

So I think this is the thing that bugs me: your post is about three different things: (1) the general idea of making benign assumptions about others, contra the fundamental attribution error; (2) the idea of doing so according to a "hierarchy of innocuousness"; (3) the specific hierarchy of innocuousness you proposed. Of these, the proposed name "theodicy" is a decent match for #1, but (a) we don't need a new name for #1 because we already have "principle of charity" and (b) you're trying to attach it to (I think) #1, #2 and #3 all together.

For me, (b) makes "theodicy" a positively bad name for the concept -- because there's another concept in the same vicinity that it wants to attach itself to. The term "wants" to match to just #1, at least in my head, and trying to use it for all three feels like you're trying to make your preferred form of being-charitable-about-people be correct by definition. As if I wrote a post called "Having good taste" and it turned out that I wanted to define "good taste" to mean liking Bach, dark chocolate and pizza more than Eraserhead, milk chocolate and caviar.

Now, if I'm drawing the right inferences from your response when Raemon suggested "Theodicy and the hierarchy of charity", what looks to me like a bug looks to you like a feature: you don't like Raemon's proposal because it separates these two things, and you want to keep them together. I think that's an error. They are conceptually separate because one can have either without the other; you could try to think well of people without having a prioritized list of kinds-of-merit, and you could (less plausibly) have a prioritized list of kinds-of-merit but hold that it's only rarely that there's any point trying to think well of people. (I guess you could use the list as a guide to how much any given person sucks, by finding the highest level at which you think they fail.) But in practice, saying "Conor's hierarchy of charity" (or whatever) will not only convey the idea of a prioritized list of kinds-of-merit and your particular suggestion for what that list should be, but also (via the term "charity", if nothing else) remind readers/listeners of the context for that list, namely as a thing to use when applying the principle of charity to avoid the fundamental attribution error.

Related thoughts on naming things:

Use unique, non-obvious terms for nuanced concepts by malcolmocean

Nonfiction Writing Advice by Scott Alexander (especially "9. Use strong concept handles")

What is funny about this is that Scott is great at naming concepts but his article names are often very bad. For example, he wrote a piece that outlined his view of morality and titled it "Contra Grant on Moral Offsets." Often I will google for the origin of a Scott-created concept and find that it's buried in an entirely unrelated piece with a title that no longer means anything to anyone.

I think I also disagree with something about the Moloch comparison. Like, I agree that the distinction you draw is important. But, something something... feels like there just can't be too many Molochs? If that was a common way-of-naming, it'd quickly become too overwhelming.

There is definitely a sense in which a name like Moloch is 'expensive' because you can't reasonably be expected to re-derive its meaning from the term, or at least most people using the term didn't previously know about the original Moloch. Terms that sort of say what they are are a lot less expensive, since people can be reminded of them and even learn them from context, and someone who doesn't know the term is A Thing doesn't feel as lost.

Moloch is a good name because it is important. If it wasn't important, the name would be bad because, while good for referring to Moloch, it wouldn't be worth a Coined Unique Name Slot, or whatever.

Curious how long you spent attempting to name the thing?

You have (in the past) persuaded me that Jargon is sufficiently important that we should err more (or at least as much) on the side of having more.

But my current thought it something like "I do not know anyone in the rationalsphere who has made much deliberate study of how to name things well, and this probably means we could be much better at naming things than we are." Not sure how long you tried with the theodicy post, but for the average LW-writer-that-I-assume-probably-doesn't-think-enough about naming, I think:

  1. It's worth spending 5 minutes coming up with names for the pure value of "getting the best name in 5 minutes"

  2. It's even more worthwhile to spend extra time thinking of names so that you can improve at doing it so that it can take less time later.

FYI, this is what I output in about 3 minutes.

  • Theodicy and the Hierarchy of Charity

    (for blogposts, I think "Catchy-name plus short sentence-fragment" is a pretty good compromise. In conversation people can say "you're not doing theodicy" and when they go look it up, the post title will help remind them wtf that means. I think, at the very least, this'd have been strictly better)


    • Theodicy and the Hierarchy of Good Faith

    • Operationalizing Good Faith

    • Hierarchy of Good Faith

    • Hierarchy of Charity

  • Avoiding the "Bad Bucket"


    • Bad-Bucketing

    • Badsplaining

    • Sorting the "Bad Bucket"

    • Just-How-They-Are-ism

    • Theodicy, and Sorting the Bad Bucket (I think this is my favorite)

    • (For someone who hates the term "Bucket Errors" I apparently like it as the name for this thing")

Aside: I think "The Fundamental Attribution Error" is one of the worst named things that the LW-o-sphere has tried to popularize, and it could use some kind of better name.

The technical definition of Fundamental Attribution Error is subtly different from the implied definition of "don't just put people in the Bad Bucket", but I think it is easier to say/work-with. ("Just-how-they-are-ism" was an attempt to name the FAE in particular, as opposed to name the post)

I spent more than five minutes, and in general have spent a lot of time making deliberate study of how to name things. I'm not claiming that "Theodicy in Humans" was the best expression of that study, or that I brought my full optimizing power to bear in that specific case, but if your sense of the rationalsphere not attending to this is correct, then I would expect that I'm in the 95th percentile or higher for having put real attention and modeling power behind this domain.

I claim that Theodicy and the Hierarchy of Charity is not only not "strictly better" (which is a very strong phrase) but in practice noticeably worse, because it leaves ambiguous whether the concepts in the post should be stashed under "theodicy" or under "hierarchy of charity." Sort of like if the post from Scott had been "Meditations on Moloch and Multipolar Traps." You end up with double the burden of namespace tracking, and if half your audience resonates with one name and the other half with the other, you've just created a problem where there are two terms floating around for the same thing and newcomers don't know that it's a single conversation.

For reference: I've spent >10 hours of explicit study and modeling on each of the following (and sometimes >100h):

  • Baby names

  • Fantasy names (cities and creatures and individuals)

  • Organizational titles

  • Rationalist house names

  • Phonetics

  • Puns/word inversions/prefix and suffix switches/derived meanings

  • Names for skills and techniques

  • Unique semantic schema that unambiguously link phonemes to morphemes

Woah! That sounds very unusual---it might be valuable for you talk about all that explicitly rather than write more like this post (which was presumably generated from your internalization of all that study, but which doesn't go out of its way to show it).

(Also, for what it's worth, I thought the title "Theodicy in Humans" was good---good enough for me to generate an approximation of the post before even reading it, although with slightly different context I'd have expected "theodicy" to be a derogatory analogy. And to bikeshed a bit, I might have used "theodicy for humans" [or maybe "of"], as you do in the text; it seems more accurate, and for your purposes it would make sense to use the title verbatim at least once.)

I echo this. People don't think enough about naming things so unpacking your additional thoughts will be helpful. I too am being deliberate with my naming of things but there's a good chance you have good advice for how to do it better.

I agree that "in humans" is wrong, but I don't like "for humans" either. "Theodicy in humans" means theodicy as practised by humans. "Theodicy for humans" means theodicy as it should be practised by humans. "Theodicy of humans" means theodicy applied to humans, and this is the one that matches Conor's intention. I don't think this is bikeshedding; the preposition Conor used is, for me at least, actively misleading.

Yeah, my autocorrect guessed what he meant easily enough, but I'm convinced. I think I just needed to see someone else say this.

You end up with double the burden of namespace tracking, and if half your audience resonates with one name and the other half with the other, you've just created a problem where there are two terms floating around for the same thing and newcomers don't know that it's a single conversation.

Yeah. Point conceded.

I would expect that I'm in the 95th percentile or higher for having put real attention and modeling power behind this domain.

Yeah, I did phrase the initial paragraph carefully because I expected you to be fairly attentive to this already.

I'm emphasizing this topic because I think post-names is the single biggest area in your sequence that would benefit from increased attention (and, however hard it is to make further progress, in at least a few cases, lack-of-it-being-better is the main obstacle from some of the posts being something I will proactively link people to all the time)

The world is full of unfair facts and I believe one of them is "even if naming things optimally is very hard, we need to get really good at it anyway. Especially if naming things turns out to be the dominant way we're providing value."

I will admit that I haven't been putting much strength behind naming the posts in this sequence, so your wish that it receives more attention is basically granted for the last nine posts.


Also, hopefully goes without saying (but maybe even if it does, beneficial for *other* people to hear it said?) that the sequence has been generally great and full of valuable nuggets.

(I perceive Conor has having opted into a more direct [i.e. slightly harsher] style of feedback that I would not give to people who did not seem to have opted into that)

I agree jargon is useful. And I think you missed one of the biggest reasons it is useful, which is to allow you to do more complicated mental operations now that your concepts take up less working memory. (I think cognitive psychologists refer to this as "chunking". Chunking is covered in this Coursera class.)

However, I think in some cases, people in the LW community may be too eager to create new jargon for stuff that can already be described using existing jargon. Using existing jargon leaves your conceptual map more compact and better connected. It's easier to see if two phenomena are two sides of the same coin, and it's easier to see if you are compartmentalizing/not propagating beliefs. (Example: I'd argue that Scott Alexander would benefit from uniting his thinking on "motte and bailey" and "dog whistles" in to a single conceptual handle. Or if not a single conceptual handle, an orthogonal basis for the subspace spanned by those two concepts.) And if the existing jargon is in common use, or the total jargon load is kept small, it becomes easier to make connections to insights outside the community/communicate with outsiders.

There's an analogy to computer programming here: it's not good to write new methods that do almost the same thing that existing methods do, or new, rarely-used methods which only consist of a few lines of code that call existing methods. It's not good to write your own method that duplicates functionality you could import via a library call. And of course you should give your new method a good name!

I would not go so far as to say that every new term that someone proposes should receive the same level of scrutiny that additional lines of code receive. I think in practice, people will probably not adopt new terms if they aren't actually useful. (Although they might do this as a way to do ingroup signalling--I'm still not clear on what "Ra" adds over existing concepts such as "status" and "prestige".) But anyway, my intent is more to encourage paying off the ontological equivalent of "technical debt" vs discourage greenfield ontologizing.

I agree with basically all of this. I disagree with the claim that I missed that reason, though? But that's a not-very-useful disagreement outside of point-scoring, so whatev. =)

(considered deleting/not making this comment, but left it for people's models of my brain's process)

This model predicts that e.g. Zvi's current frontpage post "The Fish-Head Monk" is well-named if and only if there's no resonant, tropey analogy name for what he's trying to say, and also a longer, phrase-based or sentence-based title would be accurate but too slow to say and too unlikely to stick.