Hacker News new | past | comments | ask | show | jobs | submit login

I know that it’s difficult for you to discuss anti-astroturfing methods without disclosing information that could make circumvention easier, but can you give us an idea of how much effort is put in to detecting this kind of activity on HN? You seem very confident that this doesn’t happen here; is that because you’re doing something to prevent it? We are, after all, talking about an entity that is known to use these tactics on pseudonymous forums at extraordinary scale.[1] With that in mind, writing off these concerns as merely “nationalistic” comes across as dismissive.

I guess the real question is: could you really detect well-executed astroturfing here, even if you tried very hard? I worry that authentic discourse on high-traffic pseudonymous forums is basically impossible if someone is determined to sew an opinion and has significant resources at their disposal.

[1] https://gking.harvard.edu/files/gking/files/50c.pdf




Thanks for asking the question and putting so much care into stating it.

I have no idea what kind of methods HN employs. I personally always check the posting history when I'm in doubt about the intentions of a poster. Most of the time, I find an extensive amount of fairly well considered comments on a variety of topics. That leads me to the conclusion that the account is 'genuine'.

To state it a bit more naively: if a Sybil attack would require the attacker to craft so many constructive comments in order to evade detection, it could actually have a net positive effect on the community as a whole.


> I personally always check the posting history when I'm in doubt about the intentions of a poster.

I do the same, and in the cases where it's somewhat obviously a new or 'bad intentioned' it still surprises me how fast the comments are grayed out and/or dead (often within minutes).

I've always been curious how much of that was human curation and how much of it some algorithm. I've never gotten to see or play around with HN-style site data, but personally I'd probably write some code that takes into account valuable 'curator' users (with some checks and safeguards of course, and manually tagged by the moderators).


More succinctly: https://xkcd.com/810/


> You seem very confident that this doesn’t happen here

That's a misunderstanding. I know my posts on this are super repetitive, but I'm careful never to claim such a thing. How could we know? I'm merely saying that the overwhelming majority of the insinuations and accusations that people come up with about it lead to precisely nothing when we investigate. It's like flipping a coin and having it come up tails a thousand times in a row: you start to look for simpler explanations, and there are clear, simple explanations for why this might be.

I've pored over this kind of data on HN so many times that the patterns are blazed into my skull. I'm happy to change my mind as soon as I see a new pattern—if nothing else, it would be a refreshing change of pace. So far, that has almost never happened on political topics [1]. It's more common on business topics, but most of those cases are at the boring end (people trying to promote their startup or whatever). I have to call this as I see it and tell you guys what reality is as far as we can tell. It would be a breach of trust with the community to do anything other than tell the truth, and in this case the truth about what we see is as boring and one-sided as my comments on the issue have been for the last five years.

Is it possible that sophisticated state actors are implanting biased comments into HN threads in ways that are so clever and subtle that they fool us completely, leaving zero traces in the data of the kind we know how to check? Of course. It's possible; how could we say otherwise? But this kind of thinking leads to the wilderness of mirrors, in which people see whatever they think they see. That way lies madness. We need some sanity-preserving heuristics. Fortunately we have at least two.

First: before concluding that there is manipulation, we need something objective to go on. We need some evidence—I'm tempted to say any evidence—that we can point to. And if you apply this rule consistently, which we do, the insinuations all evaporate under it. (Again, I mean on political topics. It's more complex on business topics.) Basically, every time we look, we find nothing. I'm happy to keep looking; the deal with HN users is that if someone is worried about abuse, they're welcome to email hn@ycombinator.com and we'll look. There are a couple such emails in the inbox at the moment that I'm hoping to get to tonight. But I don't know how to communicate to you guys how universal this pattern has been so far. The pattern is: HN users are all over the spectrum on divisive topics, they disagree with each other, often vehemently, and many people have trouble recognizing disagreement as genuine.

Second: any sufficiently well-executed astroturfing is undetectable by definition, so we can't rely just on detection. If sophisticated manipulators are among us, smart enough to evade all detection and fool all the moderators, the only defense the community has to fall back on is good-faith discussion: refuting bad arguments with correct information. That's good news, because that's how what we want HN to function anyway. Going into flamewar serves manipulators just fine, so in the long run our best hope is for HN to get better at what it ought to be doing in the first place. That's the best immune system, and the only one which stands a chance of maintaining a community against sufficiently subtle invasion.

[1] I say "almost" for strict accuracy, but the exceptions I'm talking about are boring and I'm leaving them out for brevity's sake, not because there's anything scintillating there.




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: