Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

YouTube’s LGBTQ Problem Is a Business-Model Problem

onezero.medium.com

Post reply on HN
Print view View original

21–30 of 180 posts

Re: YouTube’s LGBTQ Problem Is a Business-Model Problem

#21

Post by mushufasa »

jakelazaroff wrote: post #12
mushufasa wrote: post #9

Earlier quoted context omitted.

If this is the case, that's tragic in the dramatic sense -- LGBT groups fighting algorithms that are protecting them, because they can't appreciate the unseen danger prevented. I also imagine the 'algorithms' are a moving target, which makes me think this should be settled quickly if not dismissed, as Google tries to correct this. But certainly a lawsuit should bump this up the engineering priority list.

> If this is the case, that's tragic in the dramatic sense -- LGBT groups fighting algorithms that are protecting them, because they can't appreciate the unseen danger prevented. I don't think the article or GP are saying this. The algorithms, while ostensibly well-intentioned, are actually having the opposite effect — stifling the speech of the LGBTQ+ community while allowing their harassers to operate mostly unchec…

If they block 40% of real hate speech but also 5% of false positives (lgbt commenting on hate speech) I would say the algorithms are working overall.

I have no way to know the real numbers but I am assuming Google monitors overall effectiveness of algorithms and only use what works.

Re: YouTube’s LGBTQ Problem Is a Business-Model Problem

#22

Post by jakelazaroff »

Mirioron wrote: post #16

The thing is that a lot of us saw this coming. Whenever people complain about how YouTube needs to remove X type of content, they also need to realize that if YouTube actually starts policing that type of content then there will be false positives. Since YouTube will do it algorithmically, the chance that there will be collateral damage is much higher and YouTube takes a long time to reverse decisions they've already…

The point is that they shouldn't do it algorithmically. They need to have guidelines for what type of content is acceptable, and hire people to enforce them.

Re: YouTube’s LGBTQ Problem Is a Business-Model Problem

#23

Post by ppseafield »

Animats wrote: post #15

Arguing that there's a right to get ad revenue from your videos is a big stretch. Has that ever come up before in the history of free speech? Classic issues are more like "can we make the newspaper run our ad if we pay the regular rates?" Google recently won a lawsuit which complained YouTube was putting conservative content in "restricted" mode.[1] Others complain Google is not doing enough to avoid shoving too many…

It's not just about "not getting ad revenue". It's about accounts getting suspended, homophobic ads running on pro-lgbt videos, and a total lack of enforcement on their harassment policies.

https://twitter.com/gaywonk/status/1134264395717103617?lang=...

Re: YouTube’s LGBTQ Problem Is a Business-Model Problem

#24

Post by monocasa »

mushufasa wrote: post #21
jakelazaroff wrote: post #12

Earlier quoted context omitted.

> If this is the case, that's tragic in the dramatic sense -- LGBT groups fighting algorithms that are protecting them, because they can't appreciate the unseen danger prevented. I don't think the article or GP are saying this. The algorithms, while ostensibly well-intentioned, are actually having the opposite effect — stifling the speech of the LGBTQ+ community while allowing their harassers to operate mostly unchec…

If they block 40% of real hate speech but also 5% of false positives (lgbt commenting on hate speech) I would say the algorithms are working overall. I have no way to know the real numbers but I am assuming Google monitors overall effectiveness of algorithms and only use what works.

If they're having trouble classifying, they might not have good data on effectiveness.

Re: YouTube’s LGBTQ Problem Is a Business-Model Problem

#25

Post by jakelazaroff »

mushufasa wrote: post #21
jakelazaroff wrote: post #12

Earlier quoted context omitted.

> If this is the case, that's tragic in the dramatic sense -- LGBT groups fighting algorithms that are protecting them, because they can't appreciate the unseen danger prevented. I don't think the article or GP are saying this. The algorithms, while ostensibly well-intentioned, are actually having the opposite effect — stifling the speech of the LGBTQ+ community while allowing their harassers to operate mostly unchec…

If they block 40% of real hate speech but also 5% of false positives (lgbt commenting on hate speech) I would say the algorithms are working overall. I have no way to know the real numbers but I am assuming Google monitors overall effectiveness of algorithms and only use what works.

Okay, but why do you think the numbers look anything like that? Conversely, if they block 5% of hate speech but are 40% false positives, then the algorithms are clearly not working. We have no reason to assume they're effective, and in fact this lawsuit contends that they're actually harmful.

Re: YouTube’s LGBTQ Problem Is a Business-Model Problem

#26

Post by jakelazaroff »

big_chungus wrote: post #19
RcouF1uZ4gsC wrote: post #8

One of the big problems with machine learning algorithms at this time is that they have a hard time figuring out intent. A lot of the terms used by the LGBTQ community, have been used by homophobic people as slurs. Yes it is great that the community is reclaiming those words, but current ML tech has no way of distinguishing when those terms are being used by the community in a respectful manner from where they are us…

> doing this really changes the economics of the platform Not to mention that moderating this kind of stuff can seriously mess you up: https://www.theverge.com/2019/6/19/18681845/facebook-moderat... You end up looking at not just mean words but CP, seriously gory stuff, fringe sexual deviancy, and more. Do we really want to do this to moderators?

If YouTube can't protect its users, and the content people are posting is too distressing even for moderators, maybe YouTube shouldn't exist?

Re: YouTube’s LGBTQ Problem Is a Business-Model Problem

#27

Post by mushufasa »

RcouF1uZ4gsC wrote: post #18
mushufasa wrote: post #14

Earlier quoted context omitted.

What about crowdsourcing efforts? Why are user flags insufficient? Why can't there be a balance between full automation and human review, such as an appeal system or 'whitelist' program? Are the economics really the obstacle, or is it more of a corporate stance towards not adjudicating subjective controversies to avoid liability under the Communications Decency Act Section 230?

>What about crowdsourcing efforts? Why are user flags insufficient? Because this will devolve into different groups flagging their opponents' videos. This will especially hurt the voices of minorities and other vulnerable populations.

What about attaching a 'reputation' score to the flagger? E.g. something like:

- if there are sufficient flags, trigger a moderator review

- flaggers who correctly flagged content get +1 score or -1 score if not

that would make it harder for people to consistently flag content maliciously. Like age-old moderation systems such as calling 911 or police. If you abuse calling 911, you can get fined.

I imagine this has been tried before, and am genuinely curious why this is worse than flawed ML.

Re: YouTube’s LGBTQ Problem Is a Business-Model Problem

#28

Post by monocasa »

jakelazaroff wrote: post #22
Mirioron wrote: post #16

The thing is that a lot of us saw this coming. Whenever people complain about how YouTube needs to remove X type of content, they also need to realize that if YouTube actually starts policing that type of content then there will be false positives. Since YouTube will do it algorithmically, the chance that there will be collateral damage is much higher and YouTube takes a long time to reverse decisions they've already…

The point is that they shouldn't do it algorithmically. They need to have guidelines for what type of content is acceptable, and hire people to enforce them.

Or both. Have a scanner, but also have a meaningful appeals process, and you don't have a demonitized gap just because you were falsely flagged.

Re: YouTube’s LGBTQ Problem Is a Business-Model Problem

#29

Post by mushufasa »

monocasa wrote: post #24
mushufasa wrote: post #21

Earlier quoted context omitted.

If they block 40% of real hate speech but also 5% of false positives (lgbt commenting on hate speech) I would say the algorithms are working overall. I have no way to know the real numbers but I am assuming Google monitors overall effectiveness of algorithms and only use what works.

If they're having trouble classifying, they might not have good data on effectiveness.

You would have human review of a random sample to test effectiveness. In fact, this is usually how the ML algorithms are constructed in the first place, e.g. supervised learning. So yes I think they should have this.
Post reply on HN