Related to: Why Our Kind Can't Cooperate

Eliezer described a scene that's familiar to all of us:

Imagine that you're at a conference, and the speaker gives a 30-minute talk.  Afterward, people line up at the microphones for questions.  The first questioner objects to the graph used in slide 14 using a logarithmic scale; he quotes Tufte on The Visual Display of Quantitative Information.  The second questioner disputes a claim made in slide 3.  The third questioner suggests an alternative hypothesis that seems to explain the same data...

An outsider might conclude that this presentation went poorly, because all of the people who spoke afterwards seemed to disagree. Someone who had been to a few conferences would understand that this is normal; only the people who disagree speak up, while the the rest stay silent, because taking the mic to say "me too!" isn't a productive use of everyone's time. If you polled the audience, you might expect to find a few vocal dissenters against a silent majority. This is not what you would find.

Consider this situation more closely. A series of people step up, and say things which sound like disagreement. But do they really disagree? The first questioner is only quibbling with a bit of the presentation; he hasn't actually disagreed with the main point. The second questioner has challenged one of the claims from the presentation, but ignored the rest. The third questioner has proposed an alternative hypothesis which might be true, but that doesn't mean the alternative hypothesis is true, or even that the questioner thinks it's likely. If you stopped and asked these questioners whether they agreed with the main thrust of the presentation, they would probably say that they do. Why, then, does it sound like everyone disagrees?

In our community, we hold writing and arguments to a high standard, so when we see something that's imperfect, we speak up. Posting merely to indicate agreement ("me too") is strongly discouraged. In practice, this often translates into nit-picking: pointing out minor issues in reasoning or presentation that are easy to fix. We have lots of practice nit-picking, because we do it all the time with our own writings. We remove or rework weak arguments, expand on points that need clarification and tweak explanations with every draft. Revising a draft, however, does not mean questioning its premise; usually, by the time the first draft is finished, your mind is set, so the fact that you agree with your own paper is a given. When reviewing someone else's work, we transfer this skill, and something strange happens. If we agree, we want to help the author make it stronger, so we treat it as though we were revising our own draft, point out the sections which are weak, and explain why. If we disagree, we want to change the author's mind, so we point out the sections which caused us to disagree, and explain why. These two cases are hard to distinguish, and we usually forget to say which we're doing.

Discussions on the internet are usually dominated by dissent. Conventional wisdom states that this is because critics speak louder, but I think this is amplified by posts that are meant to be supportive but sound too much like dissent. In order to combat this, I propose the following social norm:

When criticizing something in a post other than the main point, authors should explicitly state whether they agree, disagree, or are unsure of the post as a whole.

Imagine the same conference as earlier, except that each questioner starts by saying whether or not he agreed with the presentation. "I agree with your conclusion. That said, the graph on slide 14 shouldn't use a logarithmic scale." "I agree with most of what you said, but there's a claim on slide 3 I disagree with." "I'm not sure whether I agree with your conclusion, because there's an alternative hypothesis that could explain the data." The content of these responses is the same, but the overall impression generated is very different; before, it seemed like everyone disagreed, but now it sounds like there's a consensus and they're resolving details. Since the impression people got of speakers' positions disagreed with what they would have said their positions were, that impression was false. That which can be destroyed by the truth, should be; therefore, if enforcing this rule really does change the tone of discussions, then rationalists should enforce it by asking people to clarify their positions.

New Comment
35 comments, sorted by Click to highlight new comments since:

I agree with your conclusion. That said, your title is missing an "s".

Fixed. I notice that when changing titles, it takes a minute for the change to propagate; committing the edit showed me a page with the original, incorrect title.

I agree with most of your analysis, but, if we're going to be expressing agreement more often in random discussions, we should seriously consider devices for reducing information cascade. We may need speech signals that distinguish between "I have independent evidence that should add to our confidence in the speaker's conclusion" and "based on the evidence others have presented, I now agree, but don't take my agreement as further reason to update".

We may need speech signals that distinguish between "I have independent evidence that should add to our confidence in the speaker's conclusion" and "based on the evidence others have presented, I now agree, but don't take my agreement as further reason to update".

Is it unreasonable to simply suggest that, if someone has independent evidence that adds to confidence, they provide said evidence in their agreeing post?

Lack of novel informational content in agreement strongly indicates the second mode, at least to me.

Person A says "Based on my 20 years of experience in industry X, it seems to me that Y". Persons B, C, D, and E all take the mike and say "IAWYC, and/but __".

I imagine myself assuming B, C, D, and E have additional evidence. I imagine myself increasing my confidence in conclusion Y. Do you expect to respond differently in this case, SA, or do you expect to respond the way I would in this case but believe this case is unrepresentative?

It doesn't seem at all clear to me that persons B through E have additional evidence, no, and I would not adjust my confidence on that basis (unless they then indicate some form of evidence in their later statement after the "IAWYC and/but", of course).

However, I likely would increase my confidence in conclusion Y to some degree, depending on the statements these people made, because each person remarking on the conclusion without pointing out a catastrophic flaw in the reasoning is (weak) evidence to me that such a flaw does not, in fact, exist.

How about metoo or rockon? These may not be quite the terms we want, but I'd bet that natural language already has a solution to both the phantom evidence/updating problem and avoiding jargon.

[-][anonymous]00

Meta: It would be nice if people did this when they were commenting or upvoting/downvoting

edit: just realised you're basically identifying the synthetic/analytic divide, discussed heavily in analytical philosophy. Quinean critiques apply, I don't see a distinction between those two and therefore don't see a need to signal this sort of thing.

Your suggestion encourages too many opinions. It is quite reasonable to not immediately have an overall opinion on a post you just read. You may want to ponder it for a while and even then you might reasonably just not know. You can learn useful things from posts even when you can't decide if you agree with the overall point.

jimrandomh specifically gives this example:

"I'm not sure whether I agree with your conclusion, because there's an alternative hypothesis that could explain the data."

Are you worried that a statement like that would anchor for uncertainty?

My point is much simpler - sometimes you just don't know if you agree or disagree.

I think Robin's concern is that people are reluctant to admit when they're uncertain, so they'll state opinions that they aren't confident in. Perhaps rather than saying IAWYC, it would make more sense for people to say IDDWYC (I Don't Disagree With Your Conclusion)? Not disagreeing means either agreeing or being unsure, without saying which. That would serve to clarify the difference between nitpicking and dissent, but leave a better line of retreat for people who flip from unsure to disagree.

IAWYCTS it might catch on better if we turn it into an acronym.

Actually, "IAWYC but X" would make a perfectly fine idiom for LW.

IAWYC but I worry about jargon creating undue barriers to entry. One solution implemented in some jargon-ridden communities is for the community software to automatically display or link to an explanation of the term when it appears in a comment or post.

For an example, see: http://www.kurzweilai.net/meme/frame.html?main=memelist.html?m=1%23696

Then click to an article.

[-][anonymous]40

IAWYC but I worry about jargon creating undue barriers to entry. One solution implemented in some jargon-ridden communities is for the community software to automatically display or link to an explanation of the term when it appears in a comment or post.

If such a feature were possible I'd love to see it.

Nitpick: "IAWTC" is standard.

[-][anonymous]20

IAWYC and would also provide a link for the convenience of those who are unlikely to comprehend the reference. Conveniently, the URL system makes merely hovering over most such links sufficient. But I may be considered trigger happy when it comes to links.

(I agree with what Steven says also but those extra letters put it over the threshold of practicality.)

On a related note, MarkDown allows you to append a 'title' attribute to your link by putting it in quotes after the URL - (http://thomblake.com "Thom Blake's Website") gives you my website, which makes a tooltip pop up on most browsers.

[-]Roko50

When criticizing something in a post other than the main point, authors should explicitly state whether they agree, disagree, or are unsure of the post as a whole.

Agree upvotes post

Minor suggestion: build this into the comment system. Instead of having a button that says "comment" have three buttons: "(mostly) agree", "(mostly) disagree" and "other". All agreeing comments could be put in the same place, perhaps in a column, and all disagreeing comments in their own place, perhaps another column. Each column would run like the current comments column, and one could even incorporate variable column widths; if almost all comments said "agree", then the "agree" column would occupy most of the screen.

Etiquette for the "agree" column would be different to etiquette for the "disagree" and "other" columns. It would be standard to say things like "that really inspired me" or "I feel really positive that we have someone as clever as you working on these problems" etc. We could even have different Karma counters for "positive (agreeing) karma" and "argumentative (disagreeing and other) karma". Upvoting comments in the "agree" column would based upon how much they seem to promote good emotions and bonding within the community.

EDIT: I'd love to know why this got downvoted. When my comment gets downvoted, it means that there is the tantalizing possibility of me excising a false belief from my mind. If you don't tell me what you think this false belief is, I can't do anything about it... this is especially true for comments that get voted to below 1.

There are too many different blogging/commenting systems as it is. For someone interested in finding useful or interesting content rather than in "communing", it is seriously annoying to keep track of how they work.

[-]Roko10

It strikes me that such an attitude will stifle innovation.

someone had to be the first one to implement voted commenting... for that matter, someone had to be the first one to implement commenting!

Also, what's the point of discussing new methods of group rationality if we're not going to put them into practice, because we want to keep doing things the way everyone else does?

I don't have anything against innovation - provided it's more useful than the inconsistency it introduces. Tools, including software, are used for other ends, they are not ends in themselves except for a few people who specialize in them, or are otherwise particularly interested in them. As I put it earlier, I am interested in finding interesting or useful content, not in learning to manage a dozen different software systems.

[-]Roko10

I don't have anything against innovation - provided it's more useful than the inconsistency it introduces.

Agreed. So I guess we differ on a complicated question: my pet idea for a new commenting system will (hopefully!) improve the quality of debate, but it will also introduce more complexity.

Can you think of a way of rationally deciding which way the tradeoff ought to be calculated, e.g. "3 extra minutes of learning time is worth X improvement in community quality"?

Standardize somewhat on the blogging/commenting systems. Reducing the number of different systems will lessen the complexity a lot more than adding features to one or another would increase it. Reduce the number of systems by making it easier for current sites to transfer to another system. Reduce forking of projects by making it easy to patch systems to a consistent standard.

When my comment gets downvoted, it means that there is the tantalizing possibility of me excising a false belief from my mind.

A downvote doesn't increase the possibility of your losing a false belief. It just means that someone, somewhere, decided to hit one button over another.

You have no idea who voted you down, and as we don't have clear standards as a group, you can't even make a stab in the dark as to why some individuals disapproved. It's a popularity measure, nothing more.

[-]Roko40

A downvote doesn't increase the possibility of your losing a false belief. It just means that someone, somewhere, decided to hit one button over another.

My working assumption is that votes on Less Wrong provide greater than zero information, i.e. that the hitting of the "down" button correlates positively with a poor comment.

(shrug)

My own experience is that positive and negative scores have little if any relation to the quality of a comment.

Also, total signal strength matters. The few massively-negative comments I've seen I agree are bad, but the ones with only a few negatives often don't strike me as deserving their scores.

As most comments don't receive any feedback, positive or negative, having a score of -1 only means that one person cared enough to make the comment and two cared enough to vote it down, for whatever reason.

It would be nice to be able to see total number of positive and negative votes, to see which comments are the most contentious, but that's not currently a feature.

[-]taw30

I agree with what you're proposing, I also enjoyed comments which go meta about it. I hope this becomes a common practice on OB/LW at least. Not just for the sake of arguments which we agree on, but also to make occasional genuine total disagreement stand out more strongly against the usual nitpicking background.

I agree with the overall post, but this sentence does not follow at all:

"Since the overall impression was changed by adding true information, the original impression must have been false."

If there is a big pool of considerations supporting Hypothesis A, and a smaller but still large pool supporting Not-A, then repeatedly receiving different sets of true facts could take your estimate of the truth of Hypothesis A all over the map.

You're right; that argument there was invalid. I have removed it and replaced it with a valid argument for the same conclusion: "Since the impression people got of speakers' positions disagreed with what they would have said their positions were, that impression was false."

It's interesting how this article is acting as a testbed for the norm it proposes. It'll be interesting to see whether it carries over to other threads, and how much of an effect it has. For a proper test, we'll have to see how people respond to posts that they fully disagree with; if people just habitually start their posts with "I agree with your conclusion, however", then eventually someone will post something like "I agree with your conclusion, however, I disagree with your conclusion".

Why not just vote the topic up, and comment what you like? The score on the topic or comment will be high, even if there aren't a lot of people saying "you rock" in the comments.

Isn't that the same signal?

Voting the topic up can send signals other than agreement--I vote up articles which I find interesting but with which I disagree. Explicit (dis)agreement would reduce the noise in the up/down metric.

Also, stating agreement prior to a negative remark helps with interpretation without nonverbal clues.

Voting up fails to distinguish between the overall point, the individual arguments, the novelty of the point made, and other possible factors.

There's a big difference between "I agree with your entire point", "I agree with most of your arguments, but I dispute point 3 which may undermine the point being made", and "I think you've made a novel and plausible point, but I disagree with it", all of which are (for me at least) valid reasons to upvote something.

Simply voting up and commenting makes it unclear which of the above, if any, a given commenter feels.

Blast... I apparently shouldn't have gotten distracted while writing my comment. [I agree with SA.]