Hacker News Reader

Comment by rayalez | Hacker News Reader

That makes sense. The problem is that at the end of the day the decision will have to be made either by algorithms or by people. Algorithms are less nuanced, hard to write, and will often fail, people are expensive and biased.

With the amount of videos youtube has to process, and without an almost human-level AI, I don't think there's a perfect solution.

Algorithms flagging videos for review, and then humans making decisions based on some basic set of principles seems right.

Also, to let advertisers decide where to spend their money there's targeting tools...

View on HN

The solution is obvious - tag/categorize the videos, according to very strict rules, let the advertisers decide which tags to serve. Allow the publishers to dispute the tags.

For instance, "nudity" can be defined as "visible genitals, female nipples" in the US.

"Controversial" tag should not even exist, it makes no sense, any non-trivial subject is controversial.

And honestly, YouTube can benefit from a more precise tagging/category system, it's currently extremely crude.

Strict rules for tagging would just run into the same problems as strict rules for demonetization. LGBT groups won't be any happier to see their content publicly branded "sexually explicit" than just demonetized.

By strict rules I also mean objective and unbiased.

Penis/vagina -> nudity (even if it's medical or educational).

"Fuck" -> cursing (even if it's about the etymology of the word).

Advocating murder of group X -> ban (even if you're quoting a religious text).

If these rules are strict and clear, we have a chance of creating a legal-like system. Otherwise it's subjective chaos with everyone pissed off.

So a lecture quoting from the Old Testament, or an atheist quoting its worst parts, would get banned under your regime? Because, at least for the first three examples, you are explicitly including meta-discussions of the subject.

It's impossible to create an "objective" standard. Because what you want to rate is meaning, and meaning is subjective, based on context and intent.

One of history's most famous photos is of a fully nude girl, maybe 8 or 9 years old. It makes no sense to stick any labels to such an image without considering the context it was created in, and the cultural significance it has.

You didn't read carefully enough. I said "advocating murder". So no, a lecture or a simple quotation wouldn't result in anything.

No, I read with plenty of care. Which is why I asked you to clarify why meta-discussions are included for the first three examples, but not the last.

I'd also posit that it's impossible to define a clear distinction between a "lecture quoting a text advocating murder" and actually advocating murder.

Just as it is impossible to just say penis->nudity, without losing the vast difference in meaning and effect from a photo of Michelangelo's David to hardcore pornography.

That's because, fundamentally, there just is no such clear delineation. Give me any two photos including a penis, and I'll give you one that is more offensive than one, yet less offensive than the other.

The tech community loves to ignore the long history of these problems, and denigrates everything that can't be expressed as a smart contract as "biased" or "subjective".

But that's ok... In less than 200 years we'll have a great algorithm who will finally render the definitive test of what's porn and what's art: "I'll know it when I see it".

> I'd also posit that it's impossible to define a clear distinction between a "lecture quoting a text advocating murder" and actually advocating murder.

A steel-manning of context should be enough to draw a distinction. Sadly, most contexts today are being redefined by folks who are seeking or making porn for those who need to get off on their own moral outrage.

"Thou shalt not suffer a witch to live."

"And he that blasphemeth the name of the LORD, he shall surely be put to death, and all the congregation shall certainly stone him."

"If a damsel that is a virgin be betrothed unto an husband, and a man find her in the city, and lie with her; Then ye shall bring them both out unto the gate of that city, and ye shall stone them with stones that they die"

Are you under the impression that the legal system is entirely objective?

> Advocating murder of group X -> ban (even if you're quoting a religious text).

An alarming number of social justice advocates have been equating words with violence. Canadian Bill C-16 (now law) is particularly troubling because under the Ontario Human Rights Code it makes debate over gender neutral pronouns a punishable offense. It seems to me that the folks at YouTube are, necessarily, responding with the most "conservative" policies to avoid "controversy" (i.e. boycotts).

C-16 just adds gender identity to the list of identities you're already not permitted to discriminate or advocate violence against (race, color, ethnicity, religion, age, sex, disability). It does not prohibit debate over pronouns. Go read it: http://www.parl.ca/DocumentViewer/en/42-1/bill/C-16/royal-as...

That's true - but the Canadian Dept of Justice indicated (and then removed the link from their website) that they would enforce it in accordance with the Ontario Human Rights Code. Go read it: http://www.ohrc.on.ca/en/code_grounds/gender_identity

> Advocating murder of group X -> ban (even if you're quoting a religious text).

I would say that this should not be banned. Let all who profess these beliefs be out there in the open so that they are easier to find.

So just mentioning genocide is advocating?

Could you give a specific example of what you meant?

Unfortunately, it is impossible to do this -- even strict objective and unbiased rules will piss people off.

The classic example I can think of, that has already generated headlines, concerns nudity policies in social media (Facebook in particular) that have flagged breastfeeding mothers for "nudity" (and thus ticked people off). A "by the letter" policy risks much more of this sort of thing... flagging something as "bad" that much of the social norm says isn't really a big deal.

One other thing I would worry about with "objective" rules is what culture determines the "watch list". The cultural norms for what is taboo actually does vary to some degree from culture to culture. Youtube is used heavily in so many countries, with diverse cultures. How can a single policy encompass the social norms of countries as diverse as, say, heavy Youtube using countries such as India, Japan, the UK, Germany, and the US? (And this isn't even accounting for the diversity of culture within these countries.)

As an example, the swastika, which is extremely taboo in Germany and triggers "hate speech" flags in much of the rest of the West, is (usually styled differently but still) a religious symbol in India. You could create a generic flag category for "swastika", but chances are this will eventually flag some, say, Hindu content from India and tick those people off.

I suppose you could structure the rules to account for as many cultural norms as possible, but that can get very complicated fast.

It's ironic that all your examples are so impractical and yet you confidently used them to illustrate how tagging is so easy.

I think a better solution is to have some core principles that Youtube's censorship team abides by. If the video in question doesn't violate all the principles (or X number of them), then it remains up.

Of course that is difficult and actually involves discussions, which Youtube doesn't seem interested in having.

There is zero censorship involved here. This isn't a leak about deciding what gets taken down, its about what is able to be monetized. The leak in fact specifically mentions this choice was used to avoid censorship by removing video.

Afaik they already thought of that solution because YouTube videos support tagging/categories but it's the video uploader who sets those tags.

The problem is that once you introduce such a system, and leave all the control over the tags to the uploader, people start gaming it for better search rankings/more views. So you are back to the same old problem of "who checks if the content actually matches the tags".

I think in bufferoverflows solution the tags come from algorithms and user reports not the publisher.

Where would this fit in?

https://www.youtube.com/watch?v=9R5w-PIzlUk

It's a clip-art style "coloring video" drawing a baby with a lot of syringes, which then each empty their colored liquid into the baby. Technically, perfectly innocent. In context of a positive flood of videos with an undertone ranging from brain damaging to abusive targeted at little kids it doesn't seem that innocent to me. The sum isn't larger than the whole, but it's larger than each individual piece seen in a vacuum. However, where to draw the line? I would draw the line at "if it targets little kids, demonetize it", not even because of the content of such videos, simply because I think marketing to little kids is immoral no matter in what context, but I know that's not going to fly. But there could be a category for it, and then people could check out which advertisers advertise to little kids and let that inform their wallet voting.

For me, YT is just totally ruined anyway. Yeah I still watch videos on it, 99% of the video material on the web is on YT, but even using an ad-blocker what used to feel lame now feels positively fucked up considering how horribly bad YT has handled this before it blew up as well as after advertisers started retracting ads. I want to turn my back on it for good, and if we can't find a way to self-host and directly pay what we watch, I'll find a way to live without video on the web.

Since the number of categories is limited, even if large, there will always be weird videos that don't strictly fit into any category. It's a slippery slope argument, it doesn't mean we shouldn't organize information better.

This particular video can be under

Art > Drawing > Coloring

I don't see anything damaging about it. Yes, it's weird, but so what?

For context, see https://www.theverge.com/2017/11/9/16629788/youtube-kids-dis...

> “We are shocked and appalled to see that our adverts have appeared alongside such exploitative and inappropriate content,” said a Mars spokesperson in a statement. “We have taken the decision to immediately suspend all our online advertising on YouTube and Google globally. Until we have confidence that appropriate safeguards are in place, we will not advertise on YouTube and Google.”

http://www.tubefilter.com/2017/11/24/advertisers-suspend-you...

> I don't see anything damaging about it. Yes, it's weird, but so what?

So you don't, probably not having looked at thousands of video descriptions and thumbnails on hundreds of channels featuring bondage, pregnant children, adults and children impregnated after having been drugged, dolls in bath tubs filled with things, objects and persons under car wheels and feet, drinking urine and eating poop, dominance and submission, binge eating of candy or just objects, objects being removed from a body or lumps removed by getting a syringe, babies faking their death, adults and kids with pacifiers, maggots, people being eaten, limbs being removed, unresolved tension, and dozens of other concepts [you see, that stands for something, there is just no space to expand all placeholders, just like when I said "flood" or "targeted at little kids"] repeated ad nauseam, across live action, claymation, 2D and 3D "art", all underlaid with the same handful of music and audio samples, produced on all continents except Africa maybe. So what?

Like many forms of abuse, e.g. mobbing or sexual harassment, each individual act can be explained away, and people who don't really look into things just see the one-off "weird thing". So what?

Then there's all the stuff that's neither here nor there, like making people jealous by drawing a heart on someone's belly, or a million "finger songs", or "Jony Jony Yes Papa". That's how babies learn colors. It's just copycats that experiment with the medium while not straying from the script that doesn't exist.

And of course, it's just the brains of toddlers, those aren't sponges or anything, and quite obviously, if it's not traumatizing to you, there can be no damage. We know that even "just" too much lack of healthy interaction can stunt development, but what's millions of hours of low-effort, "weird" content gonna do?

Anyways, I'm not here to tell you how you live your life, I'm just stating how I live mine. If you're not at work and not easily grossed out, maybe enjoy this 0.01% slice: https://i.imgur.com/MziRRQw.jpg -- but that's still just images, that's without the deceptive music and channel descriptions that advertise themselves as great entertainment for kids. And you can always find something there or in the ElsaGate subreddit that can serve as lightning rod, to say oh, this is overreacting, let's dismiss it all out of hand.

Your entire argument is "won't somebody think of the children".

Argument for what? For me personally being turned off from YT for good? Heh. If you think thinking of those who can't defend themselves yet is some sort of no-no, then that is your "entire argument". In the end, I don't care what simplification you use as your personal lightning rod, it's your life. If you were a friend, you would have been for the longest time - as it is, no biggie.

You do realize that sometimes it's actually acceptable to be concerned for the well-being of children?

Not even once. Not even in the context of something called "Youtube for Kids". Pearl-clutching consideration for others: not okay. Pearl-clutching shock over people having consideration for others: totally on.

Okay so I've watched quite a few of those particular videos now to make sense of them... And I've come to the conclusion that the reaction to the syringes is cultural misunderstanding. My hunch is that the purpose of the syringes is to get children comfortable with the vaccination process. There are number of other videos where the syringes are being demonstrated on animals to make the animals feel better after getting sick. In this case, it seems like US Americans are projecting their particular sensibilities onto videos made by some people in an entirely different part of the world. Yes, they're weird to me too, and definitely not high quality, but I've actually started to appreciate some of those videos as being well intentioned propaganda, as opposed to some of the other videos that are literally showing children rape scenes and violence. It's important that we don't just lump everything that makes us uncomfortable into the same category.

I found that channel by following chains of featured channels from hardcore ElsaGate channels. You'll also find such coloring videos on channels that have blatantly messed up content, too.

> it seems like US Americans are projecting their particular sensibilities onto videos made by some people in an entirely different part of the world

What part of the world would that be? Just saying "oh, there's probably a culture that has these different sensibilities" is heard a lot around ElsaGate, without ever actually referencing a specific culture. And what is an "entirely different" part of the world? Made of antimatter?

> It's important that we don't just lump everything that makes us uncomfortable into the same category.

None of this makes me feel "uncomfortable" though, and I'm not American either, so don't lump my reaction and your reaction together as "our" reaction. Also, just because the fringe is not the center, doesn't mean they're not connected: don't lump noticing the connection together with lumping things together.

"Disturbing" or "weird" or "uncomfortable" or "nice" or "awesome" and so on are not very descriptive words. A cold breeze might make me feel uncomfortable, so might a video of a puppy getting hurt, but that doesn't describe these things. It's making it about the people calling it out, and I for one am not buying.

> The solution is obvious - tag/categorize the videos, according to very strict rules, let the advertisers decide which tags to serve. Allow the publishers to dispute the tags

Tagging would use the same ML model that youtube is current running for demonetization, so there will be error as well.

Problem here lies with the expectation, it aims with recall not precision. In other words, false negatives really hits Youtube's image, twice already this year. People only needs to find a handful problematic videos with misplaced ads and claim Youtube had some major problem, which might not be case behind the scene. Yet each time it becomes major media parade on bashing the company and result in an exodus of advertisers.

If the goal is to eliminate false positives, which means less tolerance and stronger censorship, it will hurt a lot of innocent Youtubers, regardless. The crisis lies however with Youtube's model, if they cannot do a good job, no matter by algorithm or by human, managing the balance between quality and quantity of videos on their platform while keeping the advertisers happy and assured, it is just a matter of time, Youtube would ultimately degraded into our era's little television.

Aren't algorithms also biased in theory since they are created by people?

While algorithms can be biased intentionally or unintentionally, it is not true that ALL algorithms HAVE to be biased because they are created by people.

In machine learning it might be easier to unintentionally create biased algorithms. On the other hand there are domains where I would trust algorithms to be less biased than people.

The fact this is often glossed over causes many problems in society. https://99percentinvisible.org/episode/the-age-of-the-algori... makes some interesting and good arguments imo.

> Algorithms are less nuanced, hard to write, and will often fail, people are expensive and biased.

Algorithms have a fantastic property of repeatability which means that it is

(a) possible to adjust them over time

(b) possible to identify what exactly happened

That's exactly what is needed.

> Algorithms [allow to] identify what exactly happened

"This video was demonetized because at the 1:15 mark, there is a patch of light dirt that our neural network classified as naked skin with a 70% confidence."

Oh, you meant you were going to do image recognition without any sort of machine learning?

Your alternative is the idiot on his last day at Twitter who in his post Trump account blocking interview basically admitted that it was the totality of things that happened that day to him and it being his last day and that Trumps account popped up in front of him which caused him to make the arbitrary decision to block it.

Please remove humans from making decisions that need to be uniform. So yes, algorithm it is.

You're implicitly making the assumption that an algorithm is unbiased and less likely to be fallible - and unlikely to be susceptible to bias or manipulation (at least, compared to a human). There is a considerable amount of academic and popular literature showing that these assumptions are false, at least at a technical level.

The problem of vetting YouTube videos isn't necessarily something that can be solved by an algorithm, at the end of the day. Lots of the simple cases, sure - but such an algorithm won't work on even the most moderately challenging content. After all, moderating complex content such as video doesn't just require a well-trained neural network to recognise objects portrayed in images and sounds in audio... deciding whether a video is appropriate or not for monetisation is something that requires an explicit understanding of the social and cultural context which the video is going to be viewed in - which is both highly subjective and difficult to express in an algorithm.

NB;

- Neural networks (and other AI techniques) can be tricked and manipulated (see [0,1]; examples of adversarial images)

- Machine learning algorithms can induce or confirm bias through inadequate training data and poor assumptions. (see [2]; a book about bias in machine learning algorithms)

[0]: https://arxiv.org/abs/1510.05328

[1]: http://www.evolvingai.org/files/DNNsEasilyFooled_cvpr15.pdf

[2]: https://weaponsofmathdestructionbook.com

> You're implicitly making the assumption that an algorithm is unbiased and less likely to be fallible - and unlikely to be susceptible to bias or manipulation (at least, compared to a human). There is a considerable amount of academic and popular literature showing that these assumptions are false, at least at a technical level.

No, I'm arguing that algorithms are transparent, which means that they can be adjusted.

> The problem of vetting YouTube videos isn't necessarily something that can be solved by an algorithm, at the end of the day. Lots of the simple cases, sure - but such an algorithm won't work on even the most moderately challenging content.

of course they would here:

while (!decision_pass_the_spot_check()) { tweak_algorithm() }

> deciding whether a video is appropriate or not for monetisation is something that requires an explicit understanding of the social and cultural context which the video is going to be viewed in - which is both highly subjective and difficult to express in an algorithm.

Certain cultures think it is acceptable to stone women. In their societal context it is OK. We do not, however, care that it is OK in their societal context. So we make sure that in our algorithm stoning women ==> demonetized. Now we move onto the next problem. Did someone flag a video that we processed that had women stoned that was not demonetized? We open a ticket with the people responsible for that model, with the video that was not demonetized and have them figure out why it did not happen.

That's how you solve a complex problem. You break them into simpler ones and when you see an issue you address it. Claiming that the problem is just too complex for algorithms is avoidance.

Your argument that algorithms are transparent is equally flawed. There is an entire field of research dedicated to introspection of layers in convolutional neural networks! A system that is complex and nuanced enough to be able to deal with a problem such as auto-moderating YouTube videos will be phenomenally difficult to inspect (and nearly impossible to "adjust" in the way that you seem to be suggesting).

To clarify; I am not saying that it will never be possible to solve this sort of problem with an algorithm - just that doing so would require solving the small inconvenience of general artificial intelligence first... :-).

It certainly isn't beyond the realm of possibility in the (very distant) future, but to suggest that it's possible right now (and that it would be better than a human!) is to over-egg the pudding somewhat.

It is not flawed at all. If you cannot provide a validation for your algorithm decisions when challenged then you algorithm belongs in a research lab or on your laptop, not in production.

Okay, then your argument is valid... in a different layer of reality. Meanwhile, everyone and their dog are rolling out blackbox neural net classifiers to production

And i cannot wait until one of these companies ends up in court trying to explain why its engineers cannot explain how a decision was made.

Great interview article posted yesterday on this exact topic with an actual practitioner:

https://news.ycombinator.com/item?id=15901628

>It is not flawed at all. If you cannot provide a validation for your algorithm decisions when challenged then you algorithm belongs in a research lab or on your laptop, not in production.

Alright, then you cannot use an algorithm for image or video recognition. So how do you solve this problem?

(b) possible to identify what exactly happened (c) blind to the context and inputs outside of the algorithm

Neither of these are really true for neural networks: it's hard to pinpoint the exact feature or hidden layer which results in a particular classification. Training also requires datasets which are often proprietary, and stochastic approaches mean you can get different networks from the same dataset.

You may get a different network from a same dataset.

You will get a different result if you let two humans classify the data.

Not "may" but will - it's very easy to render your network essentially unauditable (you don't have thousands of petabytes' worth of storage available? you can't fab your own ML coprocessor? too bad!)

You bring up Trump's Twitter suspension in the other subthread, but in that case the failure was transient, and the person responsible and their reasoning were quickly identified; YouTube has demonetized videos with little to no explanation or recourse.

Uniformity is absolutely a goal where people's livelihoods are affected, but ML algorithms can't guarantee it by themselves: they make it even more important to also maintain accountability and transparency.

> Not "may" but will - it's very easy to render your network essentially unauditable (you don't have thousands of petabytes' worth of storage available? you can't fab your own ML coprocessor? too bad!)

Google can. So that does not apply.

> You bring up Trump's Twitter suspension in the other subthread, but in that case the failure was transient, and the person responsible and their reasoning were quickly identified; YouTube has demonetized videos with little to no explanation or recourse.

There's no recourse or explanation because it is not youtube's business model regardless of what/who makes a decision to demonetize them. Adding arbitrary human behavior to those decisions only makes it worse.

Google can. So that does not apply.

If nobody outside Google can test the algorithm, by definition it isn't auditable.

Adding arbitrary human behavior to those decisions only makes it worse.

Neural networks are derived from human behavior - they don't magically divine the spirit of what you want to do, they're an approximation based on training data someone has to put together.

Reply on news.ycombinator.com