Hacker News Reader

Comment by legohead | Hacker News Reader

These companies focusing so much on algorithms, when the best solution is to simply use actual humans.

I've done a fair amount of work on trying to detect and deal with spammers, fraudulent activity and such. Nothing beats a human eye.

My most successful approach has been to filter out the super obvious stuff, then send the rest to a queue that is kept an eye on by humans. A small staff can easily fight back against bullshit. Yes, YouTube is orders of magnitude above what I've dealt with, but I'm sure their smart engineers can filter out the majority of stuff.

It's painfully obvious that YouTube lacks human eyes on things. That, or they lack direction and empowerment of their employees.

View on HN

YouTube already have a huge staff of human moderators, probably in the tens of thousands. Moderating content in a consistent and fair manner is essentially impossible, largely because nobody agrees on what should be kept off the platform.

Is the image of a female nipple pornographic or empowering? Is the image of a murder victim terrorist propaganda or reportage? Is a political rant hate speech or free speech? There is no universal answer to any of these questions, least of all when we consider the huge difference in cultural attitudes across an international user base.

I can highly recommend the documentary The Internet’s Dirtiest Secrets: The Cleaners, which illustrates these paradoxes with brilliant clarity by intercutting between the moderators and the moderated.

> Nothing beats a human eye.

Well, thousands of videos sure beat a human eye.

Then you'll need lots and lots of people that watch those videos and make judgement calls, and there will be edge cases, and you will have to give people a chance to challenge those decisions, etc. For a forum with a few hundred posts a day, that's certainly an option. At Youtube's scale with hundreds of hours of video uploaded every minute of every day, I have my doubts.

Not to mention the cost of necessary therapy.

the best solution is to simply use actual humans. I've done a fair amount of work on trying to detect and deal with spammers, fraudulent activity and such. Nothing beats a human eye.

Except when those humans are captured ideologically, then start to warp an even handed policy towards their own ends.

Humans are expensive, and reviewing horrible content causes mental health issues. I don’t blame them for trying but I agree they aren’t as good and have a long way to go.

> best solution is to simply use actual humans

Gee, how many humans might it take to review the >400 hours of video uploaded every minute? (source: https://www.tubefilter.com/2015/07/26/youtube-400-hours-cont...)

400 hours/minute is 24,000 minutes of video per minute, or 241,920,000 minutes of video per week. Assuming a human can review 8 hours of video per day, 5 days a week (too high, but not by much), and that YouTube engineers can create an algorithm that automatically marks 95% of video (probably low, 95% correct labelling is easy to get; 99% is usually necessary to outdo humans), then you would need 5,040 humans.

Automatically labelling 97.5% would halve that. If employees can only review 6 hours of video per work-day, then it would increase by a third. Both, you'd need 3,360 humans.

And multiple that by let's say 50K/year fully burdened cost per employee (assuming locating them in low-cost areas) and we've now added a minimum of $125 million per year in overhead for a service that already probably doesn't make much money.

> we've now added a minimum of $125 million per year in overhead

Which is <1% of the revenue that YouTube generated last year? Still a meaningful number. But, not unmanageable.

At what profit margin, though? I've never seen actual YouTube numbers broken out in a financial statement.

>and that YouTube engineers can create an algorithm that automatically marks 95% of video (probably low, 95% correct labelling is easy to get; 99% is usually necessary to outdo humans)

I think a crucial question is whether the numbers quoted are before or after the current set of automated filtering (content id and any existing mechanisms to catch bad, illegal or offensive content.)

They may already be catching the vast majority of easy and fingerprintable things. They may be at the barrier where false positives increase at a high rate.

> you'd need 3,360 humans

Likely this number would decline rapidly over time - i.e. months vs. years - as the ML would use the human tagging to improve itself to the point of reaching the 99% threshold fairly quickly.

why are you falsely assuming it all needs to be reviewed? it only needs to be reviewed if it's flagged. each additional flag on the same vid should double the likelihood that the material is objectionable. The question is how much video is actually flagged, and what percentage of those reports are false? Even that's an easy problem to solve, because if the same accounts keep flagging clearly-not-objectionable content they can just be muted.

This is why it's great to have an ecosystem of many small players.

Reply on news.ycombinator.com