One of the best achievements of the LessWrong community is our high standard of discussion. More than anywhere else, people here are actively trying to interpret others charitatively, argue to the point, not use provocative or rude language, apologise for inadvertent offenses while not being overtly prone to take offense themselves, avoid their own biases and fallacies instead seeking them in others, and most importantly, find the truth instead of winning the argument. Maybe the greatest attribute of this approach is its infectivity - I have observed several newcomers to change their discussing habits for better in few weeks. However, not everybody is susceptible to the LW standards and our attitude produces somewhat bizarre results when confronted with genuine trolls.
Recent posts about epistemology1 have all generated large number of replies; in fact, the discussions were among the largest in the last few months. People have commented there (yes, I too am guilty) even if it was clear that the author of the posts doesn't actually react to our arguments. After he was rude and had admitted to do it on purpose. After commiting several fallacies, after generating an unreasonable amount of text of mediocre to low quality, after saying that he is neither trying to convince anyone nor he is willing to learn anything nor he aims for agreement. In short, perhaps all symptoms of trolling were present, and still, people were repeatedly patiently explaining what's wrong with the author's position. Which reaction is, I must admit, sort of amazing - but on the other hand, it is hard to deny that the whole discussion was detrimental to the quality of LW content and was mostly a waste of time.
So, here is the question: why didn't we apply the don't feed the troll meme, as would probably happen much sooner on most forums? I have several hypotheses on that.
1. We are unable to recognise trolls for lack of training. The first hypothesis is quite improbable, given that the concerned troll was downvoted to oblivion2, but still possible. There are not many trolls on LW and perhaps it is difficult to believe that someone is actively seeking that sort of confrontation. I have never understood the psychology of trolls - I try to avoid combative arguments instinctively and find it hard to imagine why somebody would intentionally try to create one. Perhaps a manifestation of the typical mind fallacy combines with compartmentalisation here: although we consciously know that there are trolls out there (as this is hard to ignore), when meeting one our instict tells us that the person cannot be so much different from us.
2. We are unwilling to deal with trolls. The second theory is that although we know that a person isn't sincere, we cherish our standards of discussion so strongly that we still try to respond kindly and maintain a civil debate, or at least one side of the debate. If it is the case, it is not automatically a bad policy. Our rationality is limited and we always operate under the threat of self-serving biases. Some quasi-deontological rule of kindness in debates, even if it is an overkill, may be useful in the same way presumption of innocence is useful in justice.
3. Sunken costs. Once the debate has started, our initial investments feel binding. It is unsettling to quit an argument admitting that it was completely useless and we have lost an hour of our life for nothing. Sunken costs fallacy is well known and widespread, there is no reason to expect we are immune.
4. Best rebuttal contest. An interesting fact is that not only the number of replies was fairly large, but also lot of replies were strongly upvoted. It leads me to suspect that those replies weren't in fact aimed at the opponent in the discussion, but rather intended to impress the fellow LessWrongers. Once the motivation is not "I want to convince my interlocutor" but rather "I can craft an extraordinarily elegant counter-argument which until now didn't appear", the attitude of the opponent doesn't matter. The debate becomes an exercise in arguing, a potentially useful practice maybe, but with many associated dangers.
5. Trollish arguments are fun. I include this possibility mainly for completeness since I don't much believe that significant number of LW users enjoy pointless arguments. But still, there is something fascinating in fallacious arguments. They are frustrating to follow, for sure, especially for a rationalist, but I cannot entirely leave out of consideration the appeal of seeing biases and fallacies in real life, as opposed to mere reading about them in a Kahneman and Tversky paper.
Whatever of the above hypotheses is correct, or even if none of them is correct, I don't doubt that on reflection most of us would prefer to have less irrational discussions. The karma system works somehow, but slowly, and cannot prevent the trollish discussions from gaining momentum if people continue in their present voting patterns. One of the problems lies in upvoting the rebuttals which gives additional motivation for people to participate. There seem to be two main strategies of voting: "I want to see more/less of this" and "this deserves more/less karma than it presently has". The first strategy seems marginally better for dealing with trolls, but both strategies should work better when applied in context. Even a brilliant reply should not be upvoted when placed in an irrational debate: first, it is mostly wasting of resources, and more, we certainly want to see less irrational debates. I don't endorse downvoting good replies, if only because the troll could interpret it as support for his cause. But leaving them on zero seems to be a correct policy.
1 I am not going to link to them because I don't want to generate more traffic there; one of those posts figures already on the 4th place when you Google lesswrong epistemology. Neither I write down the precise topic or the name of the author explicitly, which I hope decreases the probability of his appearing here.
2 In fact, the downvoting, even if massive, came relatively late, with the person in question being able to post on the main site after several days.
To your question, my own question: Why would we apply the "don't feed the troll" meme? There is no reason to believe it works. There's no academic study of least trollish places on the web, nor are there reams of data with which we can analyze best responses. Interestingly, the most trollish places on the internet tend to rely entirely on the "don't feed the troll" meme. While the least visibly trollish places on the web all use distributed moderation systems along with an active defense by all the posters there to combat trolling.
Using 4chan's method of ignoring trolls when we could be following Stackoverflow's method of obliterating trolls is insane. It's like asking for advice on passing our driving test from the girl who failed the test 12 times. Sure she's got more experience than anyone, but that's only because she so bad at it. We have no reason to believe that "don't feed the trolls" works. Going further, I'd say we have good evidence to believe it doesn't work. We should first determine what deters and eliminates trolls best, instead of aspiring to an untested ideal.
"Don't feed the trolls" is just a meme that may or may not work. Nothing more and nothing less.