4 ms·
So if YouTube used the same deep learning models to push people to more extremist content but without using any user signals so that everyone has the same recom
by pgodzin 4y ago
So if YouTube used the same deep learning models to push people to more extremist content but without using any user signals so that everyone has the same recommendations that's no longer promotion?
Fundamentally any decisions a site makes to filter and sort content to show, including HN ordering by vote count and mixing in new content to allow it to make the top page, is an explicit choice they are making that cannot be differentiated from "promotion"
- hrunt 4y ago> So if YouTube used the same deep learning models to push people to more extremist content but without using any user signals so that everyone has the same recommendations that's no longer promotion? I don't want to respond to part of your comment and not the other, so I'll just say: I don't know, because exactly defining the specifics is not my goal. > Fundamentally any decisions a site makes to filter and sort content to show, including HN ordering by vote count and mixing in new content to allow it to make the top page, is an explicit choice they are making that cannot be differentiated from "promotion" This is the part of your comment that actually matters to me. This statement is so definitive, and yet there are people arguing just as definitively that some ways of prioritizing content for users create liability while others do not. I think there is a difference between "elevating when it otherwise wouldn't be elevated" and "providing a moderated list", but those two states are probably separated by a grey area, not a bright line. I think how YouTube identifies content for users is distinguishable from what HN does, but I also don't know if that difference matters with regard to liability, especially with regards to Section 230, which makes no attempt to legislate HOW content is made discoverable.
- pgodzin 4y agoI'm very unclear about what "otherwise" means in "elevating when it otherwise wouldn't be elevated". There's no natural state of how content would be displayed, any choice of how to do it would result in a moderated list that elevates something that wouldn't have otherwise been elevated with a different approach. There is no distinction to me between what YouTube and HN do, and I certainly don't think the law should treat them any differently. Both should be legally protected regardless of which specific approach it takes.
- Retric 4y agoModeration is selecting based on the content. Aka not every letter to the editor gets shown it comes down to the editors views on each letter. Simple orderings like chronological are not considered moderated because the content doesn’t define the order.
- wnoise 4y ago> exactly defining the specifics is not my goal Exactly defining specifics is kind of important in having workable laws.
- llbeansandrice 4y agoIt absolutely can be differentiated. One major one is that Youtube is optimizing for engagement and increased viewing time to keep people on the website longer to see more ads and increase their profits. That intent is entirely different from HN's "promotion algorithm". The law can and does differentiate across lines like these even if they're both technically "promotion algorithms".
- pgodzin 4y agoHN optimizes for increasing engagement from people in the tech industry to show more ads for portfolio companies job postings, which increases YC's profits.