Drive-By Low-Effort Criticism

even low-effort criticism is valuable in terms of truth-seeking

True, but from the perspective of conditioning, it is still a punishment. On a System-2 level this kind of criticism provides valuable information, but on a System-1 level it discourages further writing. So we get a few mistakes corrected... and then the author stops writing.

I think the problem here is with the asymmetry: among nerds, cheap praise is considered low-status, but cheap criticism is high-status, or at least socially acceptable. Therefore the cheap responses are overwhelmingly negative, even if the article is mostly good.

(By the way, I suspect that the prevalence of akrasia among nerds is a similar process, only people are doing it to themselves. Giving yourself negative feelings when you notice a mistake, but no positive feelings when things go right, does not result only in making fewer mistakes, but generally in doing less.)

One thing I do want to note is that while I think you're pointing at a real phenomena, I don't actually think the two examples you gave for my post are quite pointing at the right thing.

This itself serves as an interesting example. Even if a particular author isn't bothered by certain comments (due to an existing relationship, being unusually stoic, etc), it is still possible for others to perceive those comments as aversive/hostile/negative.

This is a feature of reality worth noticing, even before we determine what the correct response to it is. It would suggest you could have a world with many LessWrong members discussing in a way that they all enjoyed, yet it appears hostile and uncivil to the outside world who assume those participating are doing so despite being upset. This possibly has bad consequences for getting new people to join (those who aren't here [LW · GW]). You might expect this if a Nurture-native [LW · GW] person was exposed to a Combat culture [LW · GW].

If that's happening a lot, you might do any of the following:

1) shift your subculture to represent the dominant outside one

2) invest in "cultural-onboarding" so that new people learn to understand people aren't unhappy with the comments they're receiving (of course, we want this to be true)

3) create different spaces: ones for new people who are still acculturating, and others for the veterans who know that a blunt critical remark is a sign of respect.

The last one mirrors how most interpersonal relationships progress. At first you invest heavily in politeness to signal your positive intent and friendliness; progressively, as the prior of friendliness is established, fewer overt signals are required and politeness requirements drop; eventually, the prior of friendliness is so high that it's possible to engage in countersignalling behaviors.

A fear I have is that veteran members of blunt and critical spaces (sometimes LW) have learnt that critical comments don't have much interpersonal significance and pose little reputational or emotional risk to them. That might be the rational [1] prior from their perspective given their experience. A new member to the space who is bringing priors from the outside world may rationally infer hostility and attack when they read a casually and bluntly written critical comment. Rather than reading it as someone engaging positively with their post and wanting to discuss, they just feel slighted, unwelcome, and discouraged. This picture remains true even if a person is not usually sensitive or defensive to what they know is well-intentioned criticism. The perception of attack can be the result of appropriate priors about the significance [2] of different actions.

If this picture is correct and we want to recruit new people to LessWrong, we need to figure out some way of ensuring that people know they're being productively engaged with.

--------------------

Coming back to this post. Here there was private information which shifted what state of affairs the cited comments were Bayesian evidence for. Most people wouldn't know that Raemon had requested Unreal copy the comment moved from FB (where he'd posted it only partially) or that Raemon has been housemates with Qiaochu for years. In other words, Raemon has strongly established relationships with those commenters and knows them to be friendly to him-- but that's not universal knowledge. The OP's assessment might be very reasonable if you lacked that private info (knowing it myself already, it's hard for me to simulate not knowing it). This is also info it's not at all reasonable to expect all readers of the site to know.

I think it's very unfortunate if someone incorrectly thinks someone else is being attacked or disincentivized from contributing. It's worth thinking about how one might avoid it. There are obviously bad solutions, but that doesn't mean there aren't better ones than just ignoring the problem.

--------------------

[1] Rational as in the sense of reaching the appropriate conclusion with the data available.

[2] By significance I mean what is it Bayesian evidence for.

One thing I do want to note is that while I think you’re pointing at a real phenomena, I don’t actually think the two examples you gave for my post are quite pointing at the right thing.

I want to note that, this fact having been pointed out, it is now incumbent upon anyone who thinks that the OP describes a real thing (whether OP himself, or you, or anyone else who agrees) to come up with new examples (see this comment [LW(p) · GW(p)] for details).

I had asked Lauren specifically to crosspost her comment from facebook (where she’d been replying to a shorter version of the post, which I’d deliberately abridged to hit the most important points). And meanwhile, Qiaochu is my roommate and we’ve had a lot of extended discussions about the overall issue.

I think that it would, generally speaking, help if these sorts of facts about a comment’s provenance were noted explicitly. (Don’t get me wrong—there’s nothing at all wrong with this sort of thing! But knowing context like this is just very useful for avoiding confusion.)

The failure mode I’m worried about is something more like “there’s a particular risk of low-effort criticism missing the point, or being wrong, or dragging the author into a conversation that isn’t worth their time.” I don’t have a good principled distinction between that failure mode and “criticism that is actually correctly noticing that the author has made some basic mistakes that invalidate the rest of their post.”

Correct me if I’m wrong, but it seems to me that the karma system is both intended to, and actually (mostly) does, solve this problem.

To wit: if I write a post, and someone posts a low-effort comment which I think is simply a misunderstanding, or borne of not reading what I wrote, etc., I am free to simply not engage with it. No one can force me to reply, after all! But now suppose that just such a comment is upvoted, and has a high karma score. Now I stop and read it again and think about it more carefully—after all, a number of my peers among the Less Wrong commentariat seem to consider it worthy of their upvotes, so perhaps I’ve reflexively sorted it into the “low-effort nonsense” bin unjustly. And, even if I still end up with the same conclusion about the comment’s value, still I might (indeed, should) post at least a short reply—even if only to say “I don’t see how this isn’t addressed in the post; could you (or someone else) elaborate on this criticism?” (or something to that effect).

My current best guess [epistemic status: likely to change my mind about this] is something like "politeness norms, that are tailored for epistemic culture."

The point of politeness norms is to provide a simple API that lets people interact reliably, without having to fully trust each other, and without spending lots of cognitive overhead on each interaction.

Default politeness norms aren't really optimized for epistemic culture. So nerds (correctly) notice "hmm, if we limit ourselves to default politeness norms we seem epistemically screwed."

But the naive response to that causes a bunch of problems a la why our kind can't cooperate [LW · GW] and defecting by accident [LW · GW] (hey, look, another post by Lionhearted)

I think it's important that people be encouraged to write posts (both low and high effort) without stressing too much – as part of a longterm strategy causing them to grow and the site to flourish. It's also important that people be encouraged to point out errors in the post (without stressing too much), as part of a longterm strategy that includes "make sure the people writing posts get feedback, and grow, as well as correct object-level errors they may have been making so readers don't come away from the post with misinformation."

The failure mode is "people writing low effort criticism that turns out to be wrong, or missing the point, etc". (And sometimes, this being due to fairly deep differences in frame).

I think something like "epistemic status tags for both posts and comments" may help here.

A post might open with "Epistemic status: exploratory, low confidence" which sets the tone for what you should expect from it.

A comment might open with "Didn't read the post in full, but a quick first impression: This point seems wrong" or "not sure if I quite understood your point, but this seemed incorrect" or something. [Note: not necessarily endorsing those particular phrasings]. The phrasing hedges slightly against the downside of misinterpretation, frame difference, or the critic-being-the-wrong-one.

I think "accidentally suppressing criticism" is a real concern, but it seems like there should be fairly standardizable conversation tools that make it easier to write criticism without as much downside risk.

It is good to discourage people from spending a lot of effort on making things that have little or no (or even negative) value.

Would you care to distinguish a means of discouraging people from spending effort on low-value things, from a means that simply discourages people from spending effort in general? It seems to me that here you are taking the concept of "making things that have little or no (or even negative) value" as a primitive action--something that can be "encouraged" or "discouraged"--whereas, on the other hand, it seems to me that the true primitive action here is spending effort in the first place, and that actions taken to disincentivize the former, will in fact turn out to disincentivize the latter.

If this is in fact the case, then the question is not so simple as whether we ought to discourage posters from spending effort on making incorrect posts (to which the answer would of course be "yes, we ought"), but rather, whether we ought to discourage posters from spending effort. To this, you say:

But there is no virtue in mere effort.

Perhaps there is no "virtue" in effort, but in that case we must ask why "virtue" is the thing we are measuring. If the goal is to maximize, not "virtue", but high-quality posts, then I submit that (all else being equal) having more high-effort posts is more likely to accomplish this than having fewer high-effort posts. Unless your contention is that all else is not equal (perhaps high-effort posts are more likely to contain muddled thinking, and hence more likely to have incorrect conclusions? but it's hard to see why this should be the case a priori), then it seems to me that encouraging posters to put large amounts of effort into their posts is simply a better course of action than discouraging them.

And what does it mean to "encourage" or "discourage" a poster? Based on the following part of your comment, it seems that you are taking "discourage" to mean something along the lines of "point out ways in which the post in question is mistaken":

If I post a long, in-depth analysis, which is lovingly illustrated, meticulously referenced, and wrong, and you respond with a one-line comment that points out the way in which my post was wrong, then I have done poorly (and my post ought to be downvoted), while you have done well (and your comment ought to be upvoted).

But how often is it the case that a "long, in-depth analysis, which is lovingly illustrated [and] meticulously referenced" is, not only wrong, but so obviously wrong that the mistake can be pointed out via a simple one-liner? I claim that this so rarely occurs that it should play a negligible role in our considerations--in other words, that the hypothetical situation you describe does not reflect reality.

What occurs more often, I think, is that a commenter finds themselves mistakenly under the impression that they have spotted an obvious error, and then proceeds to post (what they believe to be) an obvious refutation. I further claim that such cases are disproportionately responsible for the so-called "drive-by low-effort criticism" described in the OP. It may be that you disagree with this, but whether it is true or not is in a matter of factual accuracy, not opinion. However, if one happens to believe it is true, then it should not be difficult to understand why one might prefer to see less of the described behavior.

Be a human gatherer tribe. Go hunt rabbits for a week. Tribe member sits around campfire for a week. You both have 0 rabbits at the end of the week. It would seem that atleast you tried is meaningful. I get that someone that hunts for only one day and gets 1 rabbit is better than both. And it might be that you are getting "expected rabbit" even if not actual rabbit.

I agree that length should be considered a cost not a benefit. (My own posts seem to average around 2 minutes of read time.) When the situation calls for it, I prefer to signal effort [LW(p) · GW(p)] (which I think is often useful for passing people's initial filters and getting their attention) by doing things like being meticulous about syntax/spelling, using more formal/academic language, and not by increasing length.

I'm curious if you see a serious problem with people considering length as a benefit on LW, or confusing effort with quality (beyond using effort as an initial filter). For example, can you cite a post that is not very good but has apparently been highly upvoted because it's long or otherwise seems high-effort?

Start with lower-effort posts, to get a sense of how people react to the headline and thesis statement.

Shortform [LW · GW] seems like a great way to do this.