3 ms·

I knew it. I've been saying this is going to become an issue for awhile eg [1][2]. For context Section 230 avoids strict liability for an "interactive computer

by jmyeet 5mo ago

I knew it. I've been saying this is going to become an issue for awhile eg [1][2]. For context Section 230 avoids strict liability for an "interactive computer service" hosting third-party content. For comparison, a first-party site (eg a media site) does have strict liability. Here's the problem and what (IMHO) will become a big issue: if you, as a platform, only push out content with a particular view, how is that any different to just publishing that view and thus being liable? Put another way: modern tech companies want to have their cake and eat it too. They want the protections from strict liability and, while hiding behind that shield of third-party content, they analyze the content of posts and they push or suppress posts based on the content. How is that any different to being a first-party publisher? Example: if 10,000 people write posts for or against South African apartheid and I suppress only those posts that argue against it, aren't I, in effect, publishing a pro-apartheid position? I think this is a landmine waiting to go off. Tech companies have successfully sold us on the idea that "the algorithm" is somehow agnostic or neutral. Nothing could be further from the truth, particularly with modern AI tools that can deeply classify posts based on their content, politics, etc. Meta's policy chief bragged about doing this [3]. People on HN seem to have this weird, third-rail stance on Section 230. It's nuanced. Legally, I think making algorithmic decisions based on content should be treated as a first-party act of publishing. And I believe courts are going to get there. [1]: https://news.ycombinator.com/item?id=47066527 https://news.ycombinator.com/item?id=47066527 [2]: https://news.ycombinator.com/item?id=46148529 https://news.ycombinator.com/item?id=46148529 [3]: https://theintercept.com/2024/10/21/instagram-israel-palestine-censorship-sjp/ https://theintercept.com/2024/10/21/instagram-israel-palesti...

pdonis 5mo ago
> Legally, I think making algorithmic decisions based on content should be treated as a first-party act of publishing. That was basically the direction the ruling in Anderson v. TikTok was pointing, yes? The question is going to be where you draw the line. For example, if HN's automated mechanisms make a post invisible (to anyone who doesn't have showdead enabled) after enough downvotes, is that "making algorithmic decisions based on content"? I don't think so, or at any rate I don't think it's at all the same as the kind of things the big tech giants do algorithmically. But the line needs to be carefully drawn.
- krapp 5mo ago
  >I don't think so, or at any rate I don't think it's at all the same as the kind of things the big tech giants do algorithmically. It's an algorithmic decision based on content designed to influence the end user, it's exactly the same thing. Section 230 is scale invariant in that it applies to all sites hosting third party content, there is no reason to assume any regulation which replaces it would be any different.
  - pdonis 5mo ago
    > It's an algorithmic decision based on content designed to influence the end user, it's exactly the same thing. That's certainly one way to view it, yes. Indeed, court rulings (prior to Anderson v. TikTok) basically took exactly this viewpoint. Anderson v. TikTok, however, did not: that ruling found that TikTok could not claim Section 230 protection for its algorithms that recommended certain videos to users. But that ruling said nothing about moderation actions in general, such as the kind I described that HN routinely takes. Those actions are "algorithmic", yes, but the algorithm is very simple, and its key input comes from users, not the platform: basically "if a post gets x or more downvotes, put it into this different category that means users won't see it unless they opt in to seeing posts in this category". In other words, HN is not using its own data (like what kinds of posts a user has looked at, or viewed longest, etc.--I don't think HN even collects such data) to decide what posts to "recommend" (or perhaps in this case "disrecommend") to users; it's just using user input that already exists (downvotes). I think that difference is significant, but of course I'm not a lawyer. But I think it would be bad if that kind of moderation action got lumped in with the kind TikTok is being sued for, if the ruling in that suit, that TikTok can't claim Section 230 protection for its algorithms, holds up. I think sites like HN should have that protection, even if we decide to take it away from sites like TikTok for those particular kinds of algorithms. But, as I said, that line has to be carefully drawn. One possible way of drawing the line might be to look at the particular language of Section 230: it talks about sites not being liable for good faith actions to restrict access to material they deem objectionable; but it does not talk about sites not being liable for actions taken to encourage access to material they think a user would like. HN's moderation actions are the former; but the TikTok algorithms that they're being sued for were the latter.
mchusma 5mo ago
I have long wondered why these companies have only one model? Why not many models, letting the user choose? Why not let users tune what they want to see? It seems like a better user experience and safer from this type of (IMO valid) concerns. I’d want nothing to do with picking a users feed if I were them.
- Timon3 5mo ago
  Because better alignment with user interests would hurt the companies own interests. Their goal is to maximize profits, so even if they have multiple models available, they would choose the one predicted to maximize profits generated by you.

by jmyeet 5mo ago

pdonis 5mo ago

> Legally, I think making algorithmic decisions based on content should be treated as a first-party act of publishing. That was basically the direction the ruling in Anderson v. TikTok was pointing, yes? The question is going to be where you draw the line. For example, if HN's automated mechanisms make a post invisible (to anyone who doesn't have showdead enabled) after enough downvotes, is that "making algorithmic decisions based on content"? I don't think so, or at any rate I don't think it's at all the same as the kind of things the big tech giants do algorithmically. But the line needs to be carefully drawn.

krapp 5mo ago
>I don't think so, or at any rate I don't think it's at all the same as the kind of things the big tech giants do algorithmically. It's an algorithmic decision based on content designed to influence the end user, it's exactly the same thing. Section 230 is scale invariant in that it applies to all sites hosting third party content, there is no reason to assume any regulation which replaces it would be any different.
- pdonis 5mo ago
  > It's an algorithmic decision based on content designed to influence the end user, it's exactly the same thing. That's certainly one way to view it, yes. Indeed, court rulings (prior to Anderson v. TikTok) basically took exactly this viewpoint. Anderson v. TikTok, however, did not: that ruling found that TikTok could not claim Section 230 protection for its algorithms that recommended certain videos to users. But that ruling said nothing about moderation actions in general, such as the kind I described that HN routinely takes. Those actions are "algorithmic", yes, but the algorithm is very simple, and its key input comes from users, not the platform: basically "if a post gets x or more downvotes, put it into this different category that means users won't see it unless they opt in to seeing posts in this category". In other words, HN is not using its own data (like what kinds of posts a user has looked at, or viewed longest, etc.--I don't think HN even collects such data) to decide what posts to "recommend" (or perhaps in this case "disrecommend") to users; it's just using user input that already exists (downvotes). I think that difference is significant, but of course I'm not a lawyer. But I think it would be bad if that kind of moderation action got lumped in with the kind TikTok is being sued for, if the ruling in that suit, that TikTok can't claim Section 230 protection for its algorithms, holds up. I think sites like HN should have that protection, even if we decide to take it away from sites like TikTok for those particular kinds of algorithms. But, as I said, that line has to be carefully drawn. One possible way of drawing the line might be to look at the particular language of Section 230: it talks about sites not being liable for good faith actions to restrict access to material they deem objectionable; but it does not talk about sites not being liable for actions taken to encourage access to material they think a user would like. HN's moderation actions are the former; but the TikTok algorithms that they're being sued for were the latter.

mchusma 5mo ago

I have long wondered why these companies have only one model? Why not many models, letting the user choose? Why not let users tune what they want to see? It seems like a better user experience and safer from this type of (IMO valid) concerns. I’d want nothing to do with picking a users feed if I were them.

Timon3 5mo ago
Because better alignment with user interests would hurt the companies own interests. Their goal is to maximize profits, so even if they have multiple models available, they would choose the one predicted to maximize profits generated by you.