YouTube reportedly discouraged employees from reporting toxic content | Hacker News Reader

YouTube reportedly discouraged employees from reporting toxic content

theverge.com

20 pointsvezycash10 comments

Seems like an alarmist headline but not really very alarming when you dig into the details. First of all, this is from "more than 20" staffers. Out of a company of 2,800 employees and who knows how many content moderators on contract.

> One proposal offered a way to keep content that was “close to the line” of violating policies on the platform but remove it from the recommended tab. YouTube rejected that suggestion in 2016, a former engineer said, and instead continued to recommend videos regardless of how controversial they were.

This does not seem like a bad thing. Why should videos that do not violate site policies be removed from recommended? This sort of doesn't-break-the-rules-but-we-just-don't-like-it kind of content discrimination is rife for abuse, and would probably cause an even bigger uproar around people's content getting hidden for being "close to the line" but not across it.

> Employees outside of the moderation team were also reportedly discouraged from searching YouTube for toxic videos, for lawyers said the company would have a bigger liability if there was proof that staffers knew and acknowledged those videos existed.

This one is a bit strange. Why would employees outside the moderation team seeing a toxic video result in liability? Employees outside the moderation team presumably aren't equipped to tell which content is an isn't violating the rules, so I guess this rule might be to curb people trying to go out and find supposedly prohibited content on their own.

> Why should videos that do not violate site policies be removed from recommended?

The recommendation system seems out of control and I keep getting recommendations that I have not only never shown an interest in, but have actively expressed disinterest in when removing it. If they can't fix that, I'll welcome potentially biased human curators to clean it up from the vile crap that occasionally pops up there.

This is recommendations working as intended. If all people got was recommendations to what they already watch them they're no chance for new creators to get recommended.

Also, I wrote a note quick program to use selenium to traverse YouTube recommendations. I didn't find any weird rabbit holes or anything. I can release the source code if you want, though I'm pretty sure similar tools already exist.

Speaking as someone who has scraped Google properties for a living in the past: your test was not reliable. Google is very good at detecting inhuman traffic, and they frequently silently degrade results rather than throw up a CAPTCHA, to give competitors bad information and not tip them off that their scraping is being detected (which would prompt the scrapers to better conceal themselves)

I don't think so. It opened an incognito window over VPN. It then watched 50-100% of the video (randomly determined) before clicking a recommendation. I repeated the tests manually and got generally similar patterns of recommendations. I guess I could improve randomization of mouse movements, but I don't think they's much incentive for YouTube to detect automated browsers to control recommendations. Is your line of thinking that they say "hey, if we detect an automated browser we'll recommended a bunch of family friendly videos but when it's real users we'll recommended the creepy stuff?" That's pretty moustache-twirlingly conniving and I don't see a big incentive for YouTube to do this.

> This is recommendations working as intended. If all people got was recommendations to what they already watch them they're no chance for new creators to get recommended.

The problem wasn't that I got recommendations for videos that I hadn't watched, it's that I got recommendations for videos on topics that I had shown no interest at all in, and even worse, expressly denied interest in. Maybe after I've told the system that I'm not interested in watching a feminist getting rekt by Jordan Peterson for the tenth time the recommendation system can figure out that I don't care for the topic at all instead of just continuing to recommend Jordan Peterson videos from the seemingly thousands of other channels that post them.

There's something incredibly weird about YouTube's algorithm, Jordan Peterson and Hacker News readers. That's a very common complaint. I've watched a few videos with him in and now my recommendations are also super-saturated with his vids.

I actually think he's OK, albeit not anywhere near as strong a speaker as his followers like to think. And to be honest I did enjoy watching Cathy Newman getting rekt by him when that video first came out. So I guess I contributed in some microscopic way to your current pain. But even for me, YouTube acts like I watch the dude every single day - out of the 8 videos it's currently recommending to me, 2 of them involve him. This is way out of proportion.

I suspect something about content involving him pushes lots of entirely legitimate buttons inside the recommendations engine. The organic growth in his popularity has been astronomical, and video comments are flooded with men saying watching his vids helped them turn their life around.

So if I had to guess I'd say watching one Peterson video is highly, highly correlated with watching tons of them. I suspect people who only watch one vid for a few minutes and then forget about him are a much smaller group proportionally than most YouTubers. I also suspect that this behaviour highly correlates with the same demographics that read HN i.e. educated western men who probably don't watch lots of celebrity music videos. It also wouldn't surprise me if JP videos are way more virally shared than normal, as he inspires a somewhat evangelical fervour.

That said, it's weird the recommendation engine can't pick up on being explicitly told "not interested". It makes me think the engine doesn't know the videos are JP related explicitly and it's picking up on other characteristics, so clicking "not interested" on these videos just looks kind of random and useless ("please don't show me this kind of popular video that strongly appeals to people like me" ... well ok mr youtuber but what else am i meant to do?)

It is becoming increasingly clear that the tech giants are unfit to deal with their responsability within society. And this is purely because of the the incentives they have given the decisions they face. Should they do the right thing and stop downright wrong and inflamatory misinformation from spreading, or should they even recommend more of that to keep users watching in an endless cycle?

Note this is not about political bias, as some are mentioning. Inflammatory videos based on downright lies are not a serious political position. Or if they are, you should also accept Islamist extremist content and at least bring some ideas how to keep a civilized society with content like that circulating freely.

I think this is a good thing. Considering the political bias that seems to be common in these types of companies it's better for society if they censor fewer things.

Agreed. Companies like Google (Alphabet), FB, Twitter, etc. are trying to be our moral compass, when they themselves are completely biased. Just stop. Silicon Valley needs to stop dictating the world.

Reply on news.ycombinator.com