Hacker News Reader

Comment by yahna | Hacker News Reader

> Pichai is beset on all sides by people who expect Google to move mountains, reinvent itself

TIL not showing fake news or having your advertisers associated with ISIS videos is "moving mountains".

You have a ready made fool proof automated solution to detecting an ISIS video or fake news ready to go? Because at the scale of youtube you can't do it with manual checking. If you do then Google would probably pay for it.

If not then, yes, these two things qualify as moving mountains.

Yeah, the fundamental question here is if we really want, given the current state of technology, to have these massive-scale automated operations.

I'm fairly ambivalent about many aspects there. Fake news is nothing new, it's at least as old as pubs and people in them spreading false rumors, but what's new is the expectation that there'd be a channel that would somehow protect people from it. But I do find the desire to disclaim responsibility for helping propagate malware, fraud, and certain types of content to be very distasteful.

They aren't disclaiming responsibility though. That's the thing. They are building the best tools to manage it that they can. They might be disclaiming the ability to do it perfectly but they are most certainly trying.

While you're considering zaphar's post and mentally designing your "get rid of ISIS videos" system, remember that YouTube gets 400 hours of video uploaded every minute

So, you’d only need an average of 18,000 people on duty around the clock to do review of everything uploaded to YouTube.

At $15/hour pay, that's only about $2.4 billion/year for reviewers; all told, with supervision and other overhead, maybe on the order of $7.5 billion/yr.

It's probably not impossible for Google, and it wouldn't even make Google unprofitable. But it would make YouTube unprofitable.

It's also (probably) not a scalable solution. Consider two additional points:

1) 400/minute is only the most recent public number I could find a source for. It appears to be from 2015.

2) In 2013 that number was 100/hour and in 2011 it was 48/hour. The growth of user-created video content isn't exactly slowing down (vlogs, etc.)

Don't forget that your 15/hr employee is going to have to detect fake news despite their own personal biases. These are the same people that reshare fake news when on Facebook. You probably need some specialized training and more than one person to vet that news.

15 dollars an hour? Those jobs would go to a place where the wage was 1.50 an hour or worse .15 an hour.

You probably wouldn’t need to review every video – videos from YouTube partners such as Disney or VEVO partners could probably be exempt in the first place.

Then you probably could let an AI pre-sort most of the videos, and have humans only skip through those videos to verify the AIs decision, instead of having to analyze them all at 1x speed.

You’d probably only need a tenth of the manpower in that situation. And YouTube might even still be profitable with that.

TIL there are still people who think Google brute-forces anything.

The post you are responding to does not indicate anything about my belief in what Google currently does (well, except that it makes the amount of money it's financial reports say it makes.)

Don't forget: Google etc already gets a ton of criticism for their use of manual screening as it is - it burns people out, gives them 'PTSD', is exploitative, etc. So to account for turnover, psychiatric healthcare benefits, extra salary (remember overhead is as much as salary), and figure you would have to double that cost to eliminate criticism (not that it would but let's say it did). That actually would be larger than Google's total profits.

Weird, the porn seems to be PERFECTLY censored and the ads are laser-targeted. Google is a world leader in ML and AI. You really think this is so difficult for them?

We have different YouTube probably. I can easily search and find fetish porn videos. The fact that we don't see it recommended videos doesn't mean it not exists.

> Google is a world leader in ML and AI. You really think this is so difficult for them?

Yes

> Ads are laser targeted?

You clearly don't work in ads, we don't manage to have ads that render 100% of the time. Everyone thinks we have super awesome targeting that tracks your every move but most of that is marketing bullshit from people are trying to work for advertisers. Here's a huge amount are targeted do are you in this area, are you in the 18-24 male demo, and did you visit website XYZ in the past 30 days.

Reply on news.ycombinator.com