Hacker News Reader

Comment by bufferoverflow | Hacker News Reader

Automatic detection of curse words and nudity does work, as far as I know.

The regular categorization should be left to the publishers, but it should be much wider and multilevel, not the current choice of, what, 10 categories?

Yes, other things like hate speech are harder to detect, but, as I said, it's extremely subjective, and should not be dealt with in a general way, but rather broken down into subcategories, like "religion criticism", "radical feminism", "men's rights", etc, whatever ruffles people's feathers these days.

View on HN

> Automatic detection of curse words and nudity does work, as far as I know.

I don't know about nudity, but I can assure you that automatic detection of curse words doesn't work. Enable automatic subtitles and watch a few videos... do you think that's working?

Also, youtube is not english-only, and automated processing of anything other than english is even worse!

Recognizing a few curse words is a much simpler task than general speech recognition. Nobody says it has to be 100% correct, even humans aren't.

Yes, other things like hate speech are harder to detect

Exactly.

but, as I said, it's extremely subjective, and should not be dealt with in a general way, but rather broken down into subcategories, like "religion criticism", "radical feminism", "men's rights", etc, whatever ruffles people's feathers these days.

Unless you have some reason to think that a non-insignificant number of advertisers would be OK with ads on a subset of those categories but not all, that's just pointless busywork.

These are huge groups. Consider religious criticism - we have Dawkins, Sam Harris, Dillahunty, Thunderf00t, PZ Myers, Aron Ra, Hitchens, etc. All with hundreds of videos. Tens to hundreds of millions of views for each of these names.

If you demonetize them all because religious people consider criticism hate, you are throwing out tons of potential ad revenue.

It's not much more work, just create more granular multilevel categories.

Are all those videos being demonetized now?

My point is that implementing a complex system that forces hard decisions on both reviewers and advertisers would only be worth doing if it would actually change the things much. Since we don't have the data to know, any position we take is simply uninformed.

Without even considering the ethical concerns around directly funding that sort of thing.

  that's just pointless busywork.

Not at all - Youtube could use a complex, multi-layered system to deflect blame from themselves.

"Oh, we didn't demonetise your video, we just give advertisers tools to choose what sort of content they want their ads to show alongside. And your video is rated 'gender insensitivity level 1' because you said 'motherboard' instead of 'mainboard'. It's the advertisers' fault you made much less money on this one."

If Youtube make a clear line, they have to take a stance on whether saying 'motherboard' should be allowed or not, and whatever choice they make some people will protest. If instead they have eight categories of offensive content and twenty thresholds in each, they can outsource the decision on whether saying 'motherboard' is allowed to anonymous advertisers' marketing departments - YT can avoid taking a stance at all.

Reply on news.ycombinator.com