Hacker News Reader

Comment by flyingv | Hacker News Reader

Author here. Sounds like you and I agree that algo-generated recommendations of content are different from actual content itself. But the point of the article (and title) is that the plaintiffs in this case argue that each time YT generates recommendations, they're making new content that YT can be liable for.

I'm mostly just recapping what was said by each side at oral argument and the potential dramatic consequences if SCOTUS finds fully for Gonzalez -- not sure I follow where you're seeing bias.

View on HN

I said that recommendations themselves are materially differwnt from hosted content and therefore are not protected by section 230.

How would you respond to Google's argument that all content has to be ordered in _some_ manner to be displayed?

If they don't support ordering ascending and descending by date, number of views, and title alphabetically, that's their fault.

But do those need to be basically the only sorts?

Draw the line on number of parameters needed to produce the ordering. An A-Z or chronological ordering is based off 1 parameter whereas a recommendation algorithm uses dozens. A HN-style algorithm needs maybe 2-3 parameters. The regulator can make a value judgement that up to a certain number of parameters is just hosting a directory of third party content, whereas above that you’re promoting it and have to accept some liability for the content.

A quantity-driven test would mean that an algorithm that says "sort by number of mentions of ISIS" is safe, but an algorithm that says "sort by a score comprised of newness, number of upvotes, number of comments, and geographical proximity" would fail... It would also doom search engines, whose results pages are generated by multi-variate algorithms.

In principle I'm fine with all of that. I suspect there's not much mainstream appeal for a YouTube that only recommends ISIS videos, so you're not pushing ISIS videos to someone who wasn't already looking for them. I suspect such a site would fall foul of law enforcement on other grounds (such a rule change wouldn't be allowing anything that wasn't allowed before). I don't see why the regulator would need to decide on the same limit for pull-based search results (I've told the site what I'm looking for) rather than push-based recommendations (the site is pushing something else alongside what I was looking for).

I think your distinction on push vs pull is a good one that Congress could consider incorporating if it chooses to revisit 230, though that approach is probably too far from the current text for SCOTUS to be willing to read it in when it rules on Gonzalez.

However, something worth noting: wherever 230 lands, there's not some regulator using discretion in enforcing it (like a prosecutor deciding when to charge someone with murder) -- this is a law that gives tech platforms a defense from the private lawsuits that could otherwise put them out of business (e.g., suits by indviduals/businesses for defamation because the algo ended up recommending a "John Smith is a lying fraud" video).

No thanks. I don't want incompetent and biased government regulators dictating what software features private companies are allowed to build. If you don't like YouTube recommendations then just ignore them. No one is prying your eyes open and forcing you to watch.

No one is forcing you to smoke either, but there’d be armies of marketing teams with huge budgets encouraging you and your kids try if it were unregulated. I can recommend Fifth Risk by Michael Lewis if you want the long answer.

it doesn't have to be "displayed" at all. It doesnt even have to be searchable. Once they start selecting things to show you, they stop being a host and start being a publisher.

Yes, it would be nice to be able to watch a WWII history documentary on YouTube and not have the recommendations instantly fill with Hitler/Nazi videos.

Reply on news.ycombinator.com