note: this post is about gathering data for "what might the archipelago model look like if implemented on LW", which is s separate question from "should the archipelago model be implemented on LW?"

Periodically, I've argued for applying Archipelago-style norms to LessWrong (i.e. give people the tools to establish different norms and culture, then let them experiment however they want, and let the best cultures attract participants)

One cost of this approach is that it's much harder to keep track of which space you're in, and what the rules are. This could be disorienting.

There are various ways to streamline that process. For example, we could (and I expect will, at least on desktop computers) make it so that as soon as you start typing a comment, the moderation norms for the relevant post appear next to your comment box, so it's much easier to see what norms you're expected to follow.

But if there were *hundreds* of different types of norms, this might still get a bit bewildering (in particular if each set of norms had a lot of nuance to it), and place too high a burden on commenters.

One thing I'm wondering is how many different normsets there actually are demand for, among authors. (I do think it's important to ground this out in "what authors want" rather than "what commenters want" since the authors are doing the bulk of the work, and the conversations won't happen at all without them, with the caveat that I think it's a fine outcome for an author to write a post with one set of comment-norms, and someone who prefers other norms to write up a post titled "Discussing Bob's Post X" that sets different norms).

Knobs that I could imagine an author wanting to turn include:

  • Do you want feedback on writing style?
    • (possibly with different answers for typos, high level structure, etc)
  • Is there a particular conversation you're trying to have? Is there a product you're trying to build? Or are your mostly just throwing your ideas out into the wild and see what happens?
  • Should commenters relate to your post as if it's in a playful/creative/exploratory stage, or more of a finished product that you want judged professionally (or somewhere else on a spectrum?
  • Do you prefer that you (and each commenter) be responsible for justifying their ideas, or do you prefer commenters who are helping each other and you to figure out whether an idea is good?
  • How much do you expect people to own their own emotional state?
  • Is this intended to be a 101-discussion (or, a "201" discussion where participants have read the sequences and major-upvoted-posts-since-then, but not necessarily much else), or are you knee deep in a particular frame and you aren't interested in commenters who don't share that frame?

It'd be convenient if this turned out to compress into 2-4 major sets of norms (which could then be made clearly visually distinct)

As an author, or as a person who thinks they'd actively be motivated to write posts akin to "Discussing Bob's Ideas X through Normset Z", what are the commenting norms you'd want to have?

New Answer
New Comment

6 Answers sorted by

Gordon Seidoh Worley

140

I can think of a few natural clusters that might not satisfy everyone's every preference, but I think come close to at least letting you know which kind of space you are in. I'd be willing to compromise on my own norms if there was a nearby enough cluster that gave me 80% or better of what I want. These are based loosely on what I perceive to be the clusters people have asked for or directly expressed in commenting guidelines here and elsewhere.

Suggested norm clusters:

  • Debate/Argue/Combat - This is the cluster for typical ways of arguing things out, where one person says something, another who disagrees tries to refute it, and there's a back and forth of disagreeing. I think of this as the LW 1.0 cluster: very smart people calling you out on things so we can collectively do better by highlighting mistakes and correcting them. This cluster favors direct communication with minimal concern for managing the feelings of others.
  • Dialectic/Inquire/Cooperate - This is the cluster for digging into things by applying maximum charity, trying to understand what someone else is saying, and helping to improve their arguments rather than tearing them down when you notice a mistake. This looks different from the first cluster in that, for example, when you think someone is wrong you don't tell them they're wrong bluntly, you would ask them "hey, this seems to contradict this other thing and I can't figure out how they work together; what do you think?". You could also skewedly think of this as the NVC cluster, where each person is making space to prevent anyone getting defensive or feeling attacked.
  • Discuss/Share/Chatter - This cluster doesn't really existing on LW now because it's not primarily optimized for truth seeking, but things like this exist all over the internet. This is the "let's talk like nice, civilized people at a dinner party" cluster. No one says anything mean, everyone tries to help each other save face, and if you have something to say you bring it in the spirit of sharing only with no intention that you change anyone's beliefs (other than maybe that they now believe that you believe something).

Maybe there could be an "asshole" cluster opposite the third "nice" cluster, but I'm not sure that serves any serious purpose here on LW; people already have plenty of places to talk that way to each other and it seems anti-conducive to the general LW project, whereas the "nice" cluster could serve a purpose on some kinds of posts or for some kinds of authors that would not bring down the general quality of LW (it would just be neutral in terms of truth seeking, I think).

Just to add on to this, I ignore some of the other dimensions we could cluster around, like discussions that try to accomplish something in particular or not and feedback on writing style or not. In the case of accomplishing some project I think that's going to always be specific enough to the post to not be a regular norm and something that will have to be spelled out in the post because there's simply too many posts here that aren't about a project.

I generally find most people commenting on writing advice rude and in poor taste except in c... (read more)

So, I would still like to know "what cluster of norms do you actually prefer?" An important part of this question was "what is there actual demand for, and/or supply of?"

8Gordon Seidoh Worley
Personally I only really want to engage with the collaborative norm where people act in good faith, try to discover the source of their disagreement, build context, and then transcend it via synthesizing new understanding. I view the debate/adversarial norm cluster as strictly worse at truth seeking but necessary to support folks who aren't up to handling the greater complexity of the collaborate norm (since it requires much more mind-modeling capability, so much so that I would think imposing it would constitute ableism), so I want an adversarial norm to exist, too. I also want the "nice" norm cluster because it gives an easy onramp for folks who aren't ready for critical discourse norms, but I would also be pretty disappointed if it became too prevalent or started to be used on posts with serious content I wanted to engage with. I don't want the "asshole" norm cluster on LW.
4Gordon Seidoh Worley
Wow, that's some harsh downvoting for a comment giving a directly solicited opinion.
2Raemon
Yeah I was like "wtf bro?".
2Gordon Seidoh Worley
For context of people coming along later, I posted the comment and right away got a single -9 vote, which I found surprising enough to comment on!
4Dagon
I wonder if we should document voting norms in addition to commenting norms. Making a strong downvote early in the life of a comment (so your vote has large impact), unless you think the comment should be removed (which I presume isn't the case here) should require some explanation, IMO.
9Raemon
LW team has discussed a few options for changing the rules for strong votes, with possibilities including: * All strong votes require a (short) explanation * Strong vote power decays if you use it all the time
8sapphire
I would post much more on lesswrong if there was a 'no nitpicking' norm available.
4Raemon
For convenience of skimming, could you post this as a top-level answer?
1Pattern
You could post a new question "What clusters of norms/rules do you want?" and ask people to not worry about the number, just have that be a second phase where people choose/vote/explain their preferences among the suggestions so far.
4Raemon
Ah. What I meant to be clear (but perhaps failed to communicate, is that I wanted each author to list the norms that they personally want, rather than the norms they think other people want. The "how many norms does everyone want" is a fact that should arise emergently from the process of everyone sharing their own individual preferences.

Yeah. One of my hopes with this thread is that basically, there would turn out to be two major types of spaces people wanted, the Debate cluster and the Collaborate cluster (I don't know what names are best for either of them).

There are some fine-tune issues where... well, for example, while I don't expect Eliezer to show up in the comments anytime soon, my model of Eliezer prefers "debate/combat norms, but with people he respects". But that sort of thing might be easily solvable with "who is allowed to comment" being a different axis than "what are the norms and rules if you are commenting?"

3Gordon Seidoh Worley
I think including something like the 3rd "nice" cluster is important because it gives a way for people to post things on LW that maybe they don't want to receive critical feedback on, regardless of how it is delivered. These things are rare now, but things I can think of where the "nice" norm might be appropriate: * the short-form feeds * art * posts that people would otherwise be afraid to make because they personally lack the self-esteem to handle people being critical of the content but nonetheless would like to share (think stuff people hide in friends-only Facebook posts or secret Tumblers)
2Gordon Seidoh Worley
I think there's also a way the Debate and Collaborate clusters interact that is worth pointing out. Within a Collaborate space, Debate is unwelcome. If you come and start trying to Debate you're breaking the norms in a destructive way that wrecks what Collaborate is about. Within a Debate space, Collaborate is maybe perceived as "too nice" or "pulling punches" but doesn't wreck Debate. If you Collaborate within a Debate space it's like you're imposing extra rules on yourself that mean you give up some ways you could have expressed yourself (within the debate-as-combat metaphor, it's like bringing a knife to a gun fight, or tying your hands behind your back), and people may think you're less serious about your point than you are because you didn't present it as forcefully as you would have under Debate culture, but you could still get along. Both Debate and Collaborate are problems for Nice. If we go with my 4 clusters and include Asshole it's like: Nice <~ Collaborate <~ Debate <~ Asshole You can always use a lower one in a higher one, but you can't use a higher one in a lower one.
2Raemon
See Abram Demski's Combat vs Nurture and Meta-Contrarianism for (what I think of) as a more fleshed out version of this thought.
1Pattern
How does Collaborate break Nice? (The ordering is a great idea btw.!)
5Gordon Seidoh Worley
People playing nice and being polite don't want their positions analyze, even in a collaborative way; they will view that as adversarial because in that way of interacting any critical consideration is viewed as adversarial. The only correct responses within nice are those that support and accept and don't question for any purpose other than learning more and to not point out contradictions (to create cognitive dissonance in an interlocutor in nice culture would be quite rude!).
2Raemon
(the way this is actually handled in most in-person spaces I know of is that there is a natural progression where strangers are expected to show each other politeness, and people that know each other well are more likely to spar with each other. I think this is in fact good, and part of the problem of online spaces is that there's a lot of ambiguity of who knows each other well enough to spar and have it be net-positive rather than just causing everyone to dig in their heels and get frustrated)
4Dagon
This is interesting, most especially because it's mutable, situational, and different for different participants. There's not a policy about the topic or space, there's a meta-policy that you lurk long enough to understand the fine-grained expectations about which sub-topics are accepted to be discussed by which styles. This matches my experiences with good in-person discussions (incuding business, technical, and personal friendship groups): some points are ok (and expected) to directly contradict if you disagree, some others should be questioned and expanded, but not rejected, and still others are pretty much out of bounds for the current group. And which sub-point is in which group depends BOTH on who you are and what you think. I'm not certain that this translates well to public online discussions. I don't know if we should TRY, even. But it should make us aware that this is harder than just selecting rules we think we want - it's figuring out what behaviors to encourage and discourage, in what situations. It's way more detailed than "pick an island or a subreddit that has rules you like".

Wei Dai

110

I think I don't have any strong object-level preferences along these lines, and if norm clusters develop, will probably end up copying/adopting whatever norm cluster that seems to produce the most vibrant, highest quality discussions.

One thing I would really like though, is a chance to experiment with this idea to see what effects it has on discussions (hopefully positive ones), and would definitely enable it for myself if it was an option that authors could choose for their comment sections.

Would you see Reacts as a viable tool for that idea, or do you mean something else? (and, whether Reacts are quite the right tool, would you want the agree/disagee/etc information to have usernames attached, or not?)

sapphire

50

I would post much more on lesswrong if there was a 'no nitpicking' norm available.

(re-posted as a top level comment at Ray's request)

Raemon

40

Answering for myself on my own posts:

  • I almost always have a product I'm trying to build, and a particular frame for the conversation that I'm most interested
    • I think people showing up with other frames that explore different directions is almost always good. All else being equal more comments is better even if they aren't quite the frame I had.
    • People who make demands on my time that aren't related to the conversation I wanted to have are at least somewhat negative for me, and on the margin make me want to post somewhere else or not at all.
  • I usually prefer getting feedback on writing style/quality
  • Ideas vary in terms of "early stage" vs "late stage", and generally in the early stage I care more about people understanding my frame, and in the later stages I think it makes sense for people to critique it in more detail, from a variety of frames.
  • My sense of "who is responsible for emotional state" is sort of complicated and maybe beyond scope for this comment. But also, I don't think I usually run into this problem on my own posts, in part because I try to design posts such that disagreements play out in a way I consider productive.

Davis_Kingsley

30

I ideally want ~one set of norms, even if it doesn't agree with my preferences on all points. The cost of illegibility and decentralization is not in my view worth the benefit of being able to fiddle with everything.

I do think it's possible for one set of norms to basically include all the stuff you mentioned as different knobs, though.

Hmm – can you talk more about what it'd mean to have one set of rules and norms with lots of knobs? (Like, how is that distinguished from having multiple sets of knobs? Are you saying that you expect clarity over the knobs to reduce the confusion to the point that we can all agree that there's just one set of meta-rules? That seems plausible but it'd be helpful for you to spell out more detail of what that looks like)

-1Davis_Kingsley
Hmm, perhaps another way to put it would be that I would like there to be one "standard of discourse" across the site, and multiple types of conversations that can be had within that standard.
2Raemon
What distinction would you draw between that and "there is a meta-standard that includes either 2, or possibly hundreds, of conversation-types?" (I have come around to "'hundreds of conversation' types are pretty confusing and not a good idea", but not sure which principle you're drawing.

Bendini

20

I generally prefer norms that look like sparring - anything that's relevant is fair game, anything on the boundary of personal attack is fair game so long as you can make the case for its relevance.

Personal preferences aside, the biggest norm problem I've encountered is when people make an assertion based on priors that are taboo to discuss but you can't make a solid counterargument without addressing them.

10 comments, sorted by Click to highlight new comments since:

I'm not sure I'd place the emphasis on authors as strongly as you did. In terms of words written, hours spent, and intellectual progress generated, commenters might be equal to post authors. Further, it's frequent commenters who pay the high cost of there behind being multiple norm sets.

I think it might be valuable to ask a separate question about commenter preferences. But this question is mostly oriented around "if LessWrong leaned hard into the archipelago model, what would that look like?" which is a separate question from "Is the archipelago model good?", and it comes with some background assumptions, which include:

  • For a given comment section, there can only be one set of norms. (Or at least, it'd require a lot of additional work to create multiple comment section norms per post, and that sounds really confusing and hard)
  • I think authors and commenters are jointly necessary for LessWrong being healthy, but I still think authors are dramatically upstream of commenters in terms of value. Slatestarcodex has a vibrant commenter community, but seems obvious (to me) that this is because Scott is a good writer that attracts people to the blog. In LessWrong's healthy heyday, there were prominent authors who were skilled at writing who wrote constantly and when those authors left the good commenters driveled up.
  • There are relatively few candidates for "who gets to set the norms for a given comment section."
    • I don't think "aggregated preferences of commenters" is a viable candidate – I think that's mostly a choice not to go the archipelago route, and instead go either "we figure out the single best set of norms and implement that universally" or "norms are always sort of confusing and depend on context". (I'm not necessarily opposed to either of those, but the latter one in particular has some obvious downsides)
    • The three other schelling candidates I can see are "the author", "the LessWrong moderation team", and "freelance moderators that the author appoints, and/or opts into a particular moderation space." (i.e. authors might submit to subreddits that have their own norms, but in this case the choices are still rooted in author preference)

There is a skillset, separate from authorship, which is "willingness to cultivate* and/or enforce* norms". One of the problems with my initial conception of archipelago-on-LW was that it required those two skillsets be found in the same person (but, actually, they are quite distinct).

So it may be particularly valuable to ask "of people who would be interested in running a subreddit and enforcing/cultivating norms there, what clusters of norms would they be interested in cultivating?". But I still think the value of this question is downstream of "which authors would actually want to post in that subreddit?"

But I'd guess/hope that this is addressed by the "you count as an 'author' for this question if you'd be motivated to actively create spaces with particular normsets."

*there's two fairly different conceptions of norms, one of which is where they're something you enforce, another is they're something you cultivate. Enforcement looks more like saying "hey, you're violating the norm, please don't" (which is backed up by force). Cultivation looks more like just being a "good participant" who helps steer conversations in useful ways.

In many ways I think cultivation is better, except that it's much more costly in terms of time, and sometimes emotional energy.

For compeleteness, not sure if I endorse it: another conception of norms is that they're something you discover. They're an emergent feature of very complex social expectations, and even the attempt to formalize them causes ripple effects that can make the norms less effective. There are certainly ways to influence norms, both to filter for participants more amenable to norms you prefer, and to nudge those participants toward behaviors that reinforce those norms, but they remain fundamentally illegible and chaotic.

I guess that's "lightweight cultivation", more 'tending' than 'planning'.


Do you model different threads on reddit as each thread having their own norm, which can't be entirely codified/put into words?

Please distinguish between "norms", which are generally at least somewhat illegible and enforced informally at multiple levels, from "rules", which are expressed more clearly and judged/enforced by a coordinated entity. Moderation is about rules. Voting, reacts, and comments are controlled by norms.

The site owners, mods, and authors have pretty good control over rules, and can perhaps design nudges toward norms. The readers and commenters are the ones who'll actually determine the norms (and, in turn, be influenced by each other's ideas of the norms).

As more a reader/commenter than an author, I'm unlikely to pay much conscious attention to more than 2 or 3 variants of the guidelines. Possibly a few more if there is a strong separation between the islands (visually distinct, different post/comment/curation lists), so my brain is in a different context for each of them.

[edit: This probably should be a comment, not an answer. Is there any way for me to change it? ]

I agree with this distinction. Will try to keep it in mind in the future and possibly rewrite the post when I have time to clarify.

[meta: I think you should be able to move comments back and forth between answers and comments by using the triple-dot menu in the top-right. It might not be there if you're using GreaterWrong]

More complicated alternatives:

(Borrowed ideas from current answers marked with [1], inspiration marked with [2], clarifying something I wrote marked with [3])

1) norms consisting of simple defined pieces, of which all or some combinations of pieces exist as "norms/normsets".

Example pieces:

Ways to disagree: A) Saying "This is obviously wrong." B) Asking "what am I missing?"

Spelling[1]: A) If I spelled something wrong, tell me. B) Say nothing. C) Only let me know if it really sticks out like a sore thumb and it's not clear what I mean. D) Editing commentary takes place in an comment which just says "Errata" which D1) Only the author can make, D2) Anyone can make.

2) Comment sections come in different flavors, or sections. The author might "host"/moderate one of these, and maybe people who are interested in participating in other sections have a way of indicating it, and if there's enough interest A) some people (number of mods based on number of people interested) who are interested in participating instead serve as moderators. B) People who are allowed to moderate (or are allowed to moderate a given norm set) can indicate their willingness to moderate a (currently) un-hosted, or insufficiently hosted, section of a post.

This one sounds complicated, and like it could go different ways itself - the OP's idea for this sounds better (in that it's simpler):

with the caveat that I think it's a fine outcome for an author to write a post with one set of comment-norms, and someone who prefers other norms to write up a post titled "Discussing Bob's Post X" that sets different norms).

Writing this made me ask "who is going to moderate normsets[3], who/how will they be chosen, and how are normsets made/who makes them?" (It also made me feel like an an economist writing about anarcho-capitalism.)

3) Outright Groups/Channels/Threads[2]. A Group is a set of people, and a set of rules for entering (invite only, anyone can enter, or "all may read, only members may comment")

A) Groups choose a normset (and moderators?)

B) Groups come up with their own normset (and possibly the moderation - if no one (who can moderate) wants to moderate a "asshole* group", then they either cough up moderation, or do without)

How exactly a group comes into being isn't clear, or how they relate to each other. Everyone on the site could be in a giant group to start with, which has the option of organizing into schisms, with the possibility of the original group sticking around and operating by compromise rules.

4) People choose a normset. (Or multiple normsets.)

A) Each normset has an associated color. When you choose that normset, your name, and words, are displayed in that color. A complicated set of rules is layed out for how to talk with people based on both your colors. (Or a simple set, like "Never Talk to people with other colors.") This is sitewide

A1) Normsets can be changed, A2) They can't, A3) They're algorithmically assigned.

B) Like A, except Colors are confined to a comment section.

B1) Like, except whenever someone writes their first comment in a comment section, they choose a color.

B2) People can post comments with different colors, but keep it the same for the same conversation.

B3) You can change it at any time.



[3] When I wrote this I meant "who is going to moderate comments sections of posts tagged "Normset Z"."

The question associated with it's literal form is still worth asking (if anyone can make norms/groups).

Can we have different settings for subthreads than for a main post (or for the parent comment)? It seems like it might be helpful for a commenter making a semi-humorous comment, or diving into a side-point, to be able to signal that any responses to the comment have different expectations than responses to the main post.