YouTube AI deletes war crimes evidence as 'extremist material' • Hacker News Board

Re: YouTube AI deletes war crimes evidence as 'extremist material'

#301

Post by nnq » Sun, Aug 13, 2017, 3:51 PM UTC

Quanttek wrote: post #273

nnq wrote: post #267

Maybe people should get their shit together and realize that true free speech include allowing videos that seek to recruit people into despicable organizations be available! Yeah, even Hitler had a right to say what he thought and it'a a good thing he had it, despite the consequences that ensued. The problem that needs to be solved is how to educate people into not being lured into those organization DESPITE having a…

Agree to disagree. This isn't really an issue of a lack of absolute free speech but instead has much more to do with AI gone bad. With any reasonable definition of free speech (i.e. limited) it would've been wrong to remove footage documenting war crimes - hell even with a highly restricted free speech. Freedom of speech is not inherenlty the highest value there possibly is. You should be able to defend something lik…

> any reasonable definition of free speech (i.e. limited)

First, "limited free speech" is not "free speech" anymore. Second, you're just not going to be able to "define" things anymore as you automate and replace with AIs more and more processes (the definition will more am more start to be "the practical implementation of the machine learning filtering algorithm and the choice of training data", anything else will be "approximations" since you won't be able to prove things about these statistical algorithms much). The choice will be either (a) full freedom + massive investment in mechanisms to manage the negative consequences of this freedom (let's start with education first, not only making it free for all at all levels, but also forcing people be given free paid time to educate themselves, not worked to dumbness 8+hrs/day and then expecting them to differentiate real-news from fake-news...) or (b) give up freedom and live in a "well managed totalitarian system" with "freedom for distractions and sex only" or some other deal like that.

> we could easily justify not outlawing murder

No, there'l a clear criteria: reversibility! If I say something incredibly hate enticing, I can be proven wrong, and I can even retract my words and say "I changed my mind" later, that should be ok. If I murder someone, that can't be undone, even if I say I changed my mind about murdering him, he's still dead and I've still proven that I'm capable of murder (imho all people are, but that's a different discussion...).

Re: YouTube AI deletes war crimes evidence as 'extremist material'

#302

Post by tripzilch » Sun, Aug 13, 2017, 4:00 PM UTC

vidarh wrote: post #185

the8472 wrote: post #183

Earlier quoted context omitted.

I think the problem is the expectation of people that someone else do the filtering for them. I.e. "I don't want to see this kind of content" leads to "someone else should remove it from all the sites I visit". Which obviously leads to conflicting requirements once you have more than one person and those people disagree on what they want to see and don't want to see. The only reasonable solution is to host everything…

The problem on these distributed platforms is not filtering what people see , but filtering what people host or allow to transit their network connections. > In a decentralized system you also skip the law requirements since you cannot enforce multiple incompatible jurisdictions at the platform level, individual users will be responsible for enforcing it on their own nodes, similar how all you can do when accidentall…

While I don't disagree with your argument per se (not sure if I quite agree either, though), note that avoiding a decentralised platform because of being "worried about the consequences" is not necessarily the same thing as worrying that "the content (..) is content they don't find too offensive".

The first includes both legal and moral considerations, the second only moral ones.

My consideration of whether to share, part of the time, some slice of my home Internet connection bandwidth as a Tor exit node is almost entirely a legal one (I admit that time/effort may play a role too). I'd consider the moral aspect too, but I wouldn't have to think long to decide that (for me personally) the trade-offs are worth it (I could explain why and how, but I don't want to derail the discussion in that direction).

In fact I'd argue this goes for anyone, in some sense. Even if their underlying reasons align with the legal considerations (and thus not run one), it's a moral judgement. (in the worst case, there exist people who equate moral judgement with legality)

Re: YouTube AI deletes war crimes evidence as 'extremist material'

#303

Post by AmIFirstToThink » Sun, Aug 13, 2017, 4:15 PM UTC

fastball wrote: post #289

AmIFirstToThink wrote: post #248

Earlier quoted context omitted.

The advertisers would find it suitable to chase a bubble e.g. greens, nationalists, globalist, feminist, religions etc. I think a good strong case can be made to advertisers that their ad will only be served to people opting into a certain bubble. Or reverse of that i.e. show my Ads to all people except those who are in this bubble. Inclusion list and exclusion list.

That only works if all your "bubbles" are palatable bubbles. What advertiser is going to want to advertise in front of ISIS/Neo-Nazi videos?

Why, the store that sells Nazi paraphernalia on Amazon, of course. People who sell NAZI books.

Jokes aside, why does it matter that Advertisers are not choosing certain 'unpalatable' bubbles to not be associated with those channels, that's perfectly fine.

You seem to be concerned about extremist videos making money, I don't care about that at all. I just want all videos to be available unless legal system demands its removal after due process. Present sanitized content by default, present all content if explicitly requested, where an action from user says they want to see offending content.

Advertisers should be allowed to chose the channels that they advertise on. Some may choose to advertise on default channel before the content is flagged. Let the buyer make the decision. Why is YouTube giving in to certain vocal Ad buyers to decide for entire Ad market?

Re: YouTube AI deletes war crimes evidence as 'extremist material'

#304

Post by tenpies » Sun, Aug 13, 2017, 5:09 PM UTC

mnm1 wrote: post #23

Such AI coupled with the inflexible policies of companies like Google and Amazon is already starting to be a problem and will only get worse as it's deployed more broadly. Accounts are closed without recourse for invalid reasons and their owners treated like violators. Short of a law requiring explanations and an appeal process, I don't see this situation getting better ever. Yet another reason not to trust these com…

> Such AI coupled with the inflexible policies of companies like Google and Amazon

I would argue that there is a key difference in customer support which makes me much more confident in Amazon than Google.

Google has non-existent customer support for the public and virtually non-existent customer support for paid customers. If something goes wrong with your Google product your best bet - even as a paid customer - is to contact someone you know at Google. Going through the official channels is a waste of time.

Amazon, on the other hand, will bend over backwards to make sure you're satisfied - even if it loses money in that transaction. Refund decisions are mostly automated at this point, although human support for both vendors and buyers is there if needed.

Re: YouTube AI deletes war crimes evidence as 'extremist material'

#305

Post by tripzilch » Sun, Aug 13, 2017, 5:56 PM UTC

Spivak wrote: post #139

aiyodev wrote: post #126

Earlier quoted context omitted.

Not to pick on you because this seems to be a popular opinion but I would just like to point out the insanity of your what you just wrote. "as long as they don't abuse copyright" Would we delete videos of the liberation of concentration camps if there was Nickelback music playing in the background? This just demonstrates how successful media companies have been in distorting the true purpose of copyright laws: to pro…

Like, Nickelback is in the original footage or someone overalyed the track on top of the footage? I would imagine if this ever really happened it would be fine in the former, and muted in the latter. You're forgetting that 'promote...' means give the creator control of that content for the purpose of limiting access and making money. Promote in the sense that it becomes possible to actually sell artistic works like c…

> Like, Nickelback is in the original footage or someone overalyed the track on top of the footage? I would imagine if this ever really happened it would be fine in the former, and muted in the latter.

While I agree that would be the most sensible course of action, if this in reality happened right now, the whole video would be deleted in both cases, automatically (assuming it's clear enough to trigger detection in the first example).

Re: YouTube AI deletes war crimes evidence as 'extremist material'

#306

Post by giancarlostoro » Sun, Aug 13, 2017, 6:42 PM UTC

aiyodev wrote: post #126

giancarlostoro wrote: post #45

Earlier quoted context omitted.

That is a little insane... Makes me wish there was some sort of website that archived specific YouTube videos marked as historical or criminal evidence or some sort of qualification, as long as they don't abuse copyright just to make sure if places like YouTube delete them they can remain. Or a service that uploads to multiple video streaming sources at once (though I imagine these might violate the TOS of YouTube fo…

Not to pick on you because this seems to be a popular opinion but I would just like to point out the insanity of your what you just wrote. "as long as they don't abuse copyright" Would we delete videos of the liberation of concentration camps if there was Nickelback music playing in the background? This just demonstrates how successful media companies have been in distorting the true purpose of copyright laws: to pro…

I only mentioned the copyright thing, because whatever site archives YouTube videos shouldn't focus on archiving all of YouTube just specific segments of it, though it might not be a bad idea to archive a good chunk of YouTube to avoid losing internet history / real world history. It'd be highly unlikely to be a factor if these videos were archived at politically targeted / history preserving archive sites though.

Re: YouTube AI deletes war crimes evidence as 'extremist material'

#307

Post by giancarlostoro » Sun, Aug 13, 2017, 6:46 PM UTC

LoSboccacc wrote: post #250

giancarlostoro wrote: post #45

Earlier quoted context omitted.

That is a little insane... Makes me wish there was some sort of website that archived specific YouTube videos marked as historical or criminal evidence or some sort of qualification, as long as they don't abuse copyright just to make sure if places like YouTube delete them they can remain. Or a service that uploads to multiple video streaming sources at once (though I imagine these might violate the TOS of YouTube fo…

> Makes me wish there was some sort of website Can't people just use liveleaks for that kind of videos?

I guess we need more education about online video hosting is also a factor, most people from other sides of the world will know mainly the most popular websites and the smaller ones (by comparison) might not be even heard of by these people.

Re: YouTube AI deletes war crimes evidence as 'extremist material'

#308

Post by pas » Sun, Aug 13, 2017, 6:54 PM UTC

ss248 wrote: post #269

nnq wrote: post #267

Maybe people should get their shit together and realize that true free speech include allowing videos that seek to recruit people into despicable organizations be available! Yeah, even Hitler had a right to say what he thought and it'a a good thing he had it, despite the consequences that ensued. The problem that needs to be solved is how to educate people into not being lured into those organization DESPITE having a…

You are right, but i don't think that's actually possible. It's not easy to educate general population. How would you teach critical thinking, if most people just don't want to? Current solution is to just babysit general population by essentially censoring information.

Most people as in they don't want their kids to learn it, or they themselves don't want to learn it?

Because kids don't want to go to school usually, yet they do. And luckily with a good curricula the school system could teach critical thinking with a significant efficacy. (And it could be added to common core)

Re: YouTube AI deletes war crimes evidence as 'extremist material'

#309

Post by giancarlostoro » Sun, Aug 13, 2017, 7:11 PM UTC

pmlnr wrote: post #261

giancarlostoro wrote: post #45

Earlier quoted context omitted.

That is a little insane... Makes me wish there was some sort of website that archived specific YouTube videos marked as historical or criminal evidence or some sort of qualification, as long as they don't abuse copyright just to make sure if places like YouTube delete them they can remain. Or a service that uploads to multiple video streaming sources at once (though I imagine these might violate the TOS of YouTube fo…

> Makes me wish there was some sort of website that archived specific YouTube videos marked as historical or criminal evidence or some sort of qualification Make one. Add it as warrior project to archive team; save them and re-upload them if they are deleted. Unfortunately we all need to participate, because https://medium.com/message/never-trust-a-corporation-to-do-a...

Fair enough, been looking more into the projects from the Archive Team. They at least had one project, will have to see how functional it may or may not be.

https://github.com/ArchiveTeam/riptube

Re: YouTube AI deletes war crimes evidence as 'extremist material'

#310

Post by mnm1 » Sun, Aug 13, 2017, 7:29 PM UTC

tenpies wrote: post #304

mnm1 wrote: post #23

Such AI coupled with the inflexible policies of companies like Google and Amazon is already starting to be a problem and will only get worse as it's deployed more broadly. Accounts are closed without recourse for invalid reasons and their owners treated like violators. Short of a law requiring explanations and an appeal process, I don't see this situation getting better ever. Yet another reason not to trust these com…

> Such AI coupled with the inflexible policies of companies like Google and Amazon I would argue that there is a key difference in customer support which makes me much more confident in Amazon than Google. Google has non-existent customer support for the public and virtually non-existent customer support for paid customers. If something goes wrong with your Google product your best bet - even as a paid customer - is…

I specifically listed Amazon because they close accounts with no appeals and no reason. Just because you've had a good experience, doesn't mean that's universally so.