[dang]

In my experience, that's not true. Rather, people are much too quick to jump to the conclusion that so-and-so is a bot (or a troll, a shill, a foreign agent, etc.), when the other's views are outside the range of what feels normal to them.

I've written a lot of about this dynamic because it's so fundamental. Here are some of the longer posts (mini essays really):

https://news.ycombinator.com/item?id=39158911 (Jan 2024)

https://news.ycombinator.com/item?id=35932851 (May 2023)

https://news.ycombinator.com/item?id=27398725 (June 2021)

https://news.ycombinator.com/item?id=23308098 (May 2020)

Since HN has many users with different backgrounds from all over the world, it has a lot of user pairs (A, B) where A's views don't seem normal to B and vice versa. This is why we have the following rule, which has held up well over the years:

"Please don't post insinuations about astroturfing, shilling, bots, brigading, foreign agents and the like. It degrades discussion and is usually mistaken. If you're worried about abuse, email [e-mail] and we'll look at the data." - https://news.ycombinator.com/newsguidelines.html

[JohnMakin]

In my research and experience, it is. I’m making no comment about bots/shills on this site, either, I’m responding to the plausibility of the original comment.

[diggan]

> I’m making no comment about bots/shills on this site, either, I’m responding to the plausibility of the original comment.

The original comment:

> I wonder the same about HN. Has anyone done this kind of analysis? Me good LLM

Slightly disingenuous to argue from the standpoint of "I'm talking about the whole internet" when this thread is specifically about HN. But whatever floats your boat.

[imiric]

Hey, I would appreciate if you could address some of my questions here[1].

I do think it's unrealistic to believe that there is absolutely zero bot activity, so at least some of those accusations might be true.

[1]: https://news.ycombinator.com/item?id=41711060

[dang]

The claim is not "zero bot activity" - how would one even begin to support that?

Rather, the claim is that accusations about other users being bots/shills/etc. overwhelmingly turn out, when investigated, to have zero evidence in favor of them. And I do mean overwhelmingly. That is perhaps the single most consistent phenomenon we've observed on HN, and it has strong implications.

If you want further explanation of how we approach these issues, the links in my GP comment (https://news.ycombinator.com/item?id=41710142) go into it in depth. If you read those and still have a question that isn't answered there, I can take a crack at it. Since you ask (in your other comment) whether HN has any protections against this kind of thing at all, I think you should look at those past explanations—for example the first paragraph of https://news.ycombinator.com/item?id=27398725.

[imiric]

Alright, thanks. I read your explanations and they do answer some of my questions.

I'm still surprised that the percentage of this activity here is so low, below 0.1%, as you say. Given that the modern internet is flooded by bots—over 60% in the case of ProductHunt as estimated by the article, and a third of global internet traffic[1]—how do you a) know that you're detecting all of them accurately (given that it seems like a manual process that takes a lot of effort), and b) explain that it's so low here compared to most other places?

[1]: https://investors.fastly.com/news/news-details/2024/New-Fast...

[intended]

From what I understand - users accuse others of being shills and bots, and are a largely wrong.

Dang and team use other tools to remove the actual bots that they can find evidence for.

So yes, there are bots, but human reports, tend to be more about disagreements, than actual bot identification.

[dang]

intended's reply is correct.

Most of the bot activity we know about on HN has to do with voting rings and things like that, people trying to promote their commercial content. To the extent that they post things, it's mostly low-quality stuff that either gets killed by software, flagged by users, or eventually reported to us.

When it comes to political, ideological, nationalistic arguments and the like, that's where we see little (if any) evidence. Those are the areas where users are most likely to accuse each other of not being human, or posting in bad faith, etc., so that's what I've written about in the posts that I linked to.

There's still always the possibility that some bad actors are running campaigns too sophisitcated for us to detect and crack down on. I call this the Sufficiently Smart Manipulator problem and you can find past takes on it here: https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que....

I can't say whether or not this exists (that follows by definition—"sufficiently" means smart enough to evade detection). All I can tell you is that in specific cases people ask us to look into, there are usually obvious reasons not to believe this interpretation. For example would a sufficiently smart manipulator be smart enough to have been posting about Julia macros back in 2017, or the equivalent? You can always make a case for "yes" but those cases end up having to stretch pretty thin.

[JohnMakin]

Dang, I agree with and appreciate your moderation approach here and I completely agree with most of what you said. IME the last 18 months or so I’ve been here, this has been a welcome bastion against typical bot/campaign activity. Nowhere in the web seems safe the last ~dozen years. Most of what I’ve written here applies to my research of foreign bot activity on social networks, particularly in election years, in which you can far more easily piece together associations between accounts and narratives and writing style and piece together a lot more dots than on a site like this - and conclude very definitively that yes, this is a bot.

My original comment was just meant to chime in that, in the wild the last ten years, I’ve encountered an extraordinary amount of this kind of activity (which I confirmed - I really do research this stuff on the side and have written quite a lot about it) - that would support credibility to anyone that felt they experienced bot activity on this site. I haven’t done a full test on this site yet, because I don’t think it’s allowed, but at a glance I suspect particular topics and keywords attract swarms of voting/downvoting stuff, which you alluded to in your post. I think the threshold of 500 upvotes to downvote is a bit low, but clearly to me what you are doing is working. I’m only writing all of this out to make it very clear I am not making any criticisms or commentary about this site and how it handles bots/smurfs/etc.

Most of my research centers around 2016,2020 political cycles. Since the invention, release, and mass distribution of LLM’s I personally think this stuff has proliferated far beyond what anyone can imagine right now, and renders most of my old methods worthless, but for now that’s just a hypothesis.

Again, I appreciate the moderation of this site, it’s one of the few places left I can converse with reasonably intelligent and curious people compared to the rest of the web. Whatever you are doing, please keep doing it.

[pixl97]

I think that HN may in general be an outlier here. Typically outright political content is not allowed, along with religious which is quite often intertwined with politics. Because of the higher quality of the first pass filter here (users flagging this stuff), you don't see the campaigns here that you do on typical social media.

For example in Reddit you'll see accounts that are primed, that is they reuse other upvoted/mostly on topic older existing user replies on new posts of the same topic to build a natural looking account. Then at some point they'll switch to their intended purpose.

[imiric]

Thank you. I appreciate your positive outlook on these things. It helps counteract my negative one. :)

For example, when you say "The answer to the Sufficiently Smart Manipulator is the Sufficiently Healthy Community", that sounds reasonable, but I see a few issues with it.

1. These individuals are undetectable by definition. They can infiltrate communities and direct conversations and opinions without raising any alarms. Sometimes these are long-term operations that take years, and involve building trust and relationships. For all intents and purposes, they may seem like just another member of the community, which they partly are. But they have an agenda that masquerades as strong opinions, and are protected by tolerance and inclusivity, i.e. the paradox of tolerance.

2. Because they're difficult to detect, they can easily overrun the community. What happens when they're a substantial percentage of it? The line between fact and fiction becomes blurry, and it's not possible to counteract bad arguments with better ones, simply because they become a matter of opinion. Ultimately those who shout harder, in larger numbers, and are in a better position to, get heard the most.

These are not some conspiracy theories. Psyops and propaganda are very real and happen all around us in ways we often can't detect. We can only see the effects like increased polarization and confusion, but are not able to trace these back to the source.

Moreover, with the recent advent of AI, how long until these operations are fully autonomous? What if they already are? Bots can be deployed by the thousands, and their capabilities improve every day.

So I'm not sure that a Sufficiently Healthy Community alone has a chance of counteracting this. I don't have the answer either, but can't help but see this trend in most online communities. Can we do a better job at detection? What does that even look like?

[tptacek]

If you come up with good ideas on this problem you should share them, but the core of this thread is that having commenters on thread calling out other commenters as psyops, propaganda, bots, and shills doesn't work, and gravely harms the community, far more than any psyop could.