This is what Google and Facebook would like us to believe, but I'm not convinced. What it really comes down to is that they'd very much like to stay in the realm of automation, especially Google who are notorious for keeping humans out of the loop at any cost. See the long history of problems with their automated "developer relations" on Android. And the problem is that the disincentives for continuing to fail at these halfway-automated tasks are simply not strong enough to force their hand.
So they can just keep "doing their best" with automation, even though automation itself is far from the best they could do.
> it’s now so big that it must use automation to solve these problems or else swim in slightly shallower pools full of money hiring people
I think something like 300 hours of video are uploaded every minute to YouTube. How do you propose people review that in real time? Just to keep up in real time you need (300*60) 18000 people watching the videos and not falling behind, which means you likely need at least 5x that number.
Cases like this, and the OP, are common-sense to any human being. Google doesn't especially want them to be resolved in this nonsensical way, they just don't care enough to direct the necessary resources to make sure a human touch is applied to them.
Also, YouTube doesn't want to settle disputes because that makes them liable to be sued if they happen to be wrong.
So basically, how PornHub works. They cannot risk not reviewing videos because the risk is so great, so they do. And they somehow manage to thrive in their space.
There are tons of levers to pull to reduce volume without reducing profit (e.g. replace "nyancat for 24h" with a loop option), and they don't have to review everything.
If we limit review to copyright claims only, the hourly cost may be higher per person, but the review volume would be much lower.
And people really really do.
And for every hackernews poster (including myself) who says things like "I'll gladly pay X$/month for Gmail/Google/Facebook/Youtube just so that my data/privacy remained my own", there is a THOUSAND or more voices that would not.
How many people would they really have to hire? YouTube alone made $5 billion in 3 months' ad revenue: https://www.theverge.com/2020/2/3/21121207/youtube-google-al...
That's 720,000 hours per day. If you figure people work 8-hour days, that's 90,000 people. Assuming they never take vacations, don't get weekends off, and really do work 8-hour shifts with no breaks. None of which seems remotely plausible. If we assume 5-day workweeks, 6 hours of actual "productive" time per day, and 2 weeks annual vacation, closer to 175,000 people.
YouTube revenue in the corresponding timeframe is somewhere between $9.5 billion and $14 billion. How much of that is profit and how much is already being spent on infrastructure costs and whatnot is unclear. But 175k people at $50k/year, say, is $8.75 billion/year. Which is getting pretty darn close to those total revenue numbers.
Note that that's $50k expense to the employer (i.e. including benefits, employer-side taxes, office space, etc, etc, not just salaries), which is likely to be an underestimate if these people are in the US.
There's the question of whether every minute of video needs to be watched by humans, of course. But at that point we're back into automation-land to some extent.
Not even close. We're talking a small % of videos that receive copyright takedown requests, which is already a small % of all videos uploaded.
The number of cases that actually need manual moderation would be many orders of magnitude smaller.
Imagine a system that works something like the following:
- Users get access to manual dispute resolution by proving their real identity
- Persons with enough obvious violations (i.e. outright piracy with no possible argument of fair use) are barred from the dispute process
- All disputes get sent to human moderators and handled within 24 hours
- There are multiple levels of appeals, starting at appeals to unskilled laborers and ending at appeals to experienced copyright attorneys
- Going to higher levels of appeals requires paying in a nominal fee that is returned if the claim is overturned
- The final level of appeals can decide the case is too unclear and should be handled by the courts as per the DMCA
- Claimants are penalized some nominal fee for overturned claims
How many moderators would that need? It'd be 1,000 at most. Majority of them would be outsourced laborers costing around $20-30k, with a few experienced attorneys costing $200-300k for the final stage of appeals. Total cost $20-30M. That's a lot of money, but still only 0.5% of their actual revenue. Such a system would surely create far more value both to the company and to the world.
Imagine if DMCA appeals required a human review on channels above a certain size, for example.