3 ms·

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, s

by imiric 2y ago

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

dang 2y ago
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 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 https://news.ycombinator.com/item?id=27398725.
- imiric 2y ago
  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-Fastly-Threat-Research-Reveals-91-of-Cyberattacks-Targeted-Multiple-Organizations-Using-Mass-Scanning-to-Uncover-and-Exploit-Vulnerabilities/default.aspx https://investors.fastly.com/news/news-details/2024/New-Fast...
  - intended 2y ago
    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 2y ago
    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&query=by%3Adang%20sufficiently%20smart%20manipulator&sort=byDate&type=comment 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.
    - imiric 2y ago
      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?

by imiric 2y ago

dang 2y ago

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 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 https://news.ycombinator.com/item?id=27398725.

imiric 2y ago
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-Fastly-Threat-Research-Reveals-91-of-Cyberattacks-Targeted-Multiple-Organizations-Using-Mass-Scanning-to-Uncover-and-Exploit-Vulnerabilities/default.aspx https://investors.fastly.com/news/news-details/2024/New-Fast...
- intended 2y ago
  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 2y ago
  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&query=by%3Adang%20sufficiently%20smart%20manipulator&sort=byDate&type=comment 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.
  - imiric 2y ago
    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?