[pjc50]

https://www.dailymail.co.uk/news/article-8360073/More-60-peo... (caution Daily Mail; mildly NSFW sidebar)

"internal ⁦@Facebook⁩ research that found over 60% of people who joined groups sharing extremist content did so at Facebook’s recommendation."

So you're right: youtube has the megaphone. I wonder what the proportion of people watching extremist/disinformation content on youtube because of suggestions is? In some ways it's worse than TV, because if you publish shit on TV you get people writing to the regulator to complain (qv Janet Jackson superbowl nipple ridiculousness). On youtube you may never know what your fellow citizens are watching until they say "of course the world is ruled by lizards, here's the video that proves it".

I wonder if people would accept the compromise that youtube would host this content but force it to "unlisted". After all, the recommendations are their speech, not yours.

[BrandonMarc]

... and Facebook's recommendation would be based on their prior activity. If their interests on Facebook (what they like, follow, etc) were mostly cat videos, Facebook wouldn't be recommending extremist groups.

In the meantime, there's a heck of a gulf between whether or not Facebook lets a group be recommended, and actively censoring content dissenting to the chosen narrative.

[pbourke]

> If their interests on Facebook (what they like, follow, etc) were mostly cat videos, Facebook wouldn't be recommending extremist groups.

The whole point of recommendation algorithms is to find missing edges in the graph, so it can easily lead you to misinformation in 1 or 2 hops.

Think of it this way: the misinformation content is highly valuable - it generates a lot of engagement. There is always a “potential energy” (people like you also liked...) between low-value content and high-value content that the platforms are attempting to convert to “kinetic energy” (engagement - views, clicks, comments) in order to monetize it. The goal is to find the shortest path to the high value content.

[intended]

Proof required, from my personal experience moderating a political forum, and from that of other mods, the issue is the flooding of our information networks with Information prions and virii targeting our limbic systems. Social media is currently heavily polluted.

[curryst]

> If their interests on Facebook (what they like, follow, etc) were mostly cat videos, Facebook wouldn't be recommending extremist groups.

You're right, it's not that obvious, it's far more sinister. Cat videos are unlikely to end with you being recommended extremist groups, because there likely isn't much engagement from cat video viewers and extremist groups.

People who are deeply unsatisfied with life, however, might engage if they see it as a way out of their dissatisfaction, inadvertently training the recommendation algorithm to promote extremist content to dissatisfied people. That strikes me as at least plausible, though I don't know if the data is out there to find out what people are recommended what content under what criteria.

> In the meantime, there's a heck of a gulf between whether or not Facebook lets a group be recommended, and actively censoring content dissenting to the chosen narrative.

I disagree with this part. I don't have numbers handy for Facebook, but YouTube gets 500 hours of video uploaded every minute. It is physically impossible for you to see everything that gets uploaded to YouTube. Even if they stopped accepting uploads right now, you'd probably still die before you saw a significant portion of the content available.

Removing something from recommendations is, in most cases, tantamount to censoring it. If there are 500 hours uploaded per minute, and we assume that each video is 15 minutes long (which is likely longer than the reality), that's 2000 videos uploaded per minute. Assuming random distribution of views (which it's not, because of the recommendations), your video has a 0.05% chance of being viewed out of the videos uploaded in the same minute as yours. If you widen that to videos uploaded in the same hour, it goes down to a 0.00083% chance. Widen it to a day and you're down to a 0.0000347% chance. You would get 1 view per 2.8M views if YouTube deleted everything before that day, and killed recommendations entirely. I don't know how typical my usage patterns are, but I only search for probably 1 out of every 25 or 50 YouTube videos I watch. If that's a typical usage pattern, then you would actually get 1 view per 75M - 150M views. If everyone in the US logged on and watched a random video, you would get ~2-4 views.

It's all theoretical napkin math, but there is a staggering amount of data in the hands of Facebook, Google, et al. I do agree that actually removing the content is more significant, but the difference between removing the content and just making it so obscure that it's hard to see unless you're looking for it is basically the same. It's like if newspapers would agree to publish your stuff, but only if you encoded it as the first letter of each line of text. They have technically published your views, they've just made it hard enough to find that the only people who see it is people who already knew it was there.

I don't know what to suggest though. This is almost an inevitable outcome of collecting this amount of content; a lot of it is going to be relegated to some esoteric corner where no one ever sees it.

[timeeater]

The 60% number sounds big, but how many people actually joined groups with extremist content? Without that context, the 60% doesn't say much.

It also doesn't say why people joined those groups. Maybe they are just curious to see what the crazy people are up to.