AI moderation is no match for hate speech in Ethiopian languages
restofworld.org
restofworld.org
This is demonstrably false. GPT4 can easily expound upon, rephrase, simplify, clarify and correct things it has said when prompted to or provided new information. That cannot be done without some form of conceptual understanding of what was written.
The issue here, which is both obvious and covered in the article, is that they aren't using AI trained on the target languages. They are translating (imperfectly) to English first and then trying to moderate that. But, frankly, that is a difficult task for a human. Idioms, slang, coded speech (analogous to whinnie the poo=Xi) are all going to slip through the cracks even with a human.
As internet penetration rises, more and more regional language communities are forming, but the economics of moderation are such that it's difficult to hire moderators to handle hundreds of different languages.
For example, with the Rakhine crisis, how many Arakan speaking moderators can you realistically find and hire? And this is just one example of hundreds.
1. Moderation jobs tend to require a 4 year degree, which is hard to find in a region with a 75% literacy rate
2. Requires incorporation in that country. Try opening a FB office or contractor in Myanmar during an active Civil War
3. (Speculative) Hard social faultlines exist. In Rakhine, society was communalized into Buddhist ethnic Arakan - Muslim ethnic Rohingya and Bengali - Hindu Bengali and Arakan. All ethnic and religious groups spoke Arakan, but all were diametrically opposed to each other due to the communalized society that developed.
This is just one example from the Rohingya Genocide, but there are sadly dozens all over the developing world.
For example, when dealing with the Tigray war like in the article above, you'd need to moderate for the 100+ languages spoke in Ethiopia, Eritrea, Somalia, South Sudan, and Kenya (all regions with ethnic fault lines that extend into Ethiopia).
This is a problem that can only realistically be solved by AI (as it's an intent polarity problem and a Machine Translation problem), but these are less sourced languages with limited corpora or knowledge.
All of this is known btw. This is why Google, Amazon, and Meta hire hundreds of linguists in India and Chinese companies like Tencent, Baidu, and Ant hired similarly within China to work on MT and Polarity in South Asia and China respectively, but a similar effort hasn't occurred in Africa and most of ASEAN yet.
I think this is a misreading of their position. I don't think any social media company would say it's bad or unhelpful to research new strategies which might enable better moderation of Amharic content. But if we reject the idea that there is a solution that works acceptably for languages in general, and insist on social media companies auditing every individual language, that raises obvious and uncomfortable questions about support for smaller languages. Should a social media company which hasn't yet been able to verify the health of its Swazi moderation tools prohibit you from posting in Swazi?
— Paul Graham [1]
Also, how do you not know the people who post stuff are the same people who kill people?
So yeah, you're both resposible. Hitman most likely more, but ordering a hit is also a murder.
But in case where one has killed by his own volition (not being under direct order), the motivation is irrelevant.
I don't understand this perspective. Johannes Gutenberg, inventor of the printing press, shouldn't be responsible for what folks print. Neither should Sir Tim Berners-Lee, inventor of the world wide web, be responsible for what people use the web for. Humans needs to accept personal reasonability and not blame their tools.
Technology is an amplifier for human behavior. If some humans use technology for nefarious rather than constructive purposes, then that is a problem with those humans - not the technology.
I do not believe a publisher is responsible for messages of hate anymore than they are for messages of love. Even if Don Black created Stormfront specifically to host hateful user content, this does not make him responsible because it's ultimately the individual who creates their account and posts their hate. If proper parenting, education, and culture were in place, Stormfront would have zero users signing up.
To bring the point home: I do not believe Facebook should be responsible if users post inflammatory content. Humans have been waging war and spewing hate from the beginning. To imply that Facebook - an over the hill social network - is responsible is laughably idiotic. The individual must accept responsibility and the culture that produced them must do better.
Does that mean that each individual who killed someone isn't personally responsible? Hell no. They're responsible. But someone who puts out the call to kill people and expects - yes, knows for sure - that one in 100,000 of their followers will pick up the gun, is equally responsible.
I bet when you see an Israeli settler kill a Palestinian kid, you hold the Israeli government accountable. I bet when you see a white cop shoot a black kid in the US, you hold the police responsible. I bet when you heard the Jews were murdered in the gas chambers, or all the Sioux died of smallpox, you didn't hold the individual soldiers accountable: You blamed the government that sent them to do the dirty deed.
Governments have free speech, just like corporations and just like individuals. But except for very rare cases where an individual is executed under law by the government, it's always an individual who kills another individual. Yet no killer gets his ideas in a vacuum.
Most people are too weak to resist the call to dehumanize others. That call to dehumanize and kill is usually sent by groups - religious groups, political groups, or in the modern day Facebook groups. To claim that the airwaves are free and no one holds responsibility except the individual who did the killing is like saying that individual Nazis acted alone to kill Jews. It is absurd on its face.
Just look at the situation on the ground. There is a vast mass of people who will murder each other in every country if they feel they have sanction from the powers higher up. (There's a whole wildly popular movie series about this in the US called "The Purge" which is popular because it appeals to this impulse).
For any moderator or company-moderator, or radio station or other controller of the flow of information to permit and profit from the overabundance of dissemination of calls for genocide, is to be complicit in genocide. Whether or not they pull the trigger, they are responsible for manipulating popular thinking. The mob does not operate on the morality of the individual; and manipulating the mob is the job of governments and corporations. One cannot wash their hands after simply standing on a platform 400 ft in the air and saying that someone should die and then that person miraculously being killed.
The people who pull the trigger may have a harder time living with themselves, when they realize they were just pawns. But it's the people who give them the sanction who are equally, if not ultimately responsible.
Do not confer onto the rest of the human race the values you think you would uphold under the circumstances you imagine, where you might act on your conscience to save lives contrarily to the way everyone around you was acting. That's not how things have ever worked in real life.
Trying to curb culture through censorship is lazy. Cultures change on their own, for better or worse, in the absence of censorship. Tyrants love censorship because it superficially offers an immediate solution while further empowering them.
Facebook didn't just pop into existence as some magical technology or culture, it simply is a corporation that rides on top of infrastructure that it doesn't have any control over. Its only purpose is to take what one person says and show it to other people it thinks want to hear that message. It's not novel in any way, but it is a publisher and as such, should be beholden to the same rules of truth - and responsible for the messages it rebroadcasts.
It is not censorship when a party sues another party for libel. It's not censorship when promoters of genocide are tried for aiding and abetting genocide. You don't understand what censorship is.
The Facebook staff do not sit around picking and choosing which messages they'll broadcast, nor should they. If you find a message disagreeable, then be upset with the one who wrote it. Do not shoot the messenger [1].
I would like real change to come about as a result of a well educated society, not by silencing speech, blaming technology or the messenger.
Respectfully, your contrived examples of genocide and Nazi's read as ramblings. Please logically and concisely make your points without invoking Godwin's law [2] or the spiritual equivalent. If you disagree with my perspective, then that's fine. We'll have to agree to disagree.
Their examples aren't contrived, you just lack the fortitude to admit your views have no grounding in reality.
Every person with Internet access should have equal access to its wealth of information. Why do some people spend their time on Facebook becoming "deranged anger addicts" instead of educating themselves on Khan Academy?
Also, assuming what I think ("I bet when you see ...") and attacking me personally (questioning my "fortitude") is not a good look. If your points are sound, then let them stand on their own without the personal attacks.
Unless I'm mistaken, I think we both want the same end result, but there is a difference in our approach: I'd rather address the root cause (bottom-up) instead of the symptoms working back to the root (top-down).
> At one stage they weighted angry responses to a post as being worth five times more than a like ...
Do not blame Facebook for this. Ask yourself why they did this. Hint: you've already said the answer, but did not commit.
A college friend reaching out after five years? Engagement. Your stacker finding you after years of evading them? Engagement. A daughter wishing her mother a happy birthday? Engagement. Matt Walsh telling his audience that all trans people are inherently vile sexual predators? Engagement.
Even if you can reason with individuals, that means nothing on someone like Matt Walshes scale. Even if 98% of people disagree with him, that is still a successful result for him as he has millions of devoted followers who will champion his causes.
We very obviously do not want the same results, you are suggesting solutions you know don't work.
I understand that addressing root causes is difficult which is why quick and easy solutions are tempting. However, addressing the symptoms doesn't solve the problem. I would rather address root causes because in doing so I believe the symptoms will be addressed. It's fine to disagree with my problem solving approach, but we'll have to agree to disagree.
Facebook is alleged to have, and has been sued for, facilitating the messaging that aided and drastically accelerated organized mass murder [in direct contravention of their own stated guidelines]. My comparisons to both Nazi and Rwandan messaging examples was in was not an example of Godwin's Law; it is at the heart of the legal cases against Facebook.
Please take a look at this article[0], which makes the connections explicitly clear, explains why Facebook's moderation may be liable under international law, and provides links to cases in which people have been held guilty of war crimes for inciting genocide on the radio, not actually murdering people themselves.
[0] https://www.abc.net.au/religion/facebook-complicity-in-the-r...
Playing the blame game is politics 101 so of course laws will reflect that.
It is clear to me that @Devasta and yourself are more interested in addressing symptoms rather than identifying root causes. I would rather address root causes because in doing so I believe the symptoms will be addressed. We will have to agree to disagree.
Facebook is supposed to retain specialists that can interpret and act on something said in niche languages on their platform, in bad faith? Failing to do so becomes facilitation of genocide?
They can't just act on it without verification; that leads to abuse reports being filed in bad faith. We saw it with DMCA, we see it again with angry ex-Redditors. You lose control of your business when you let the customer run the till, but advocacy for letting lunatics run the asylum seems to be the expectation these days.
Facebook is just a digital bathroom wall. I don't know what half the graffiti on most walls say, but failing to paint over it doesn't make the owner culpable for gang violence. The graffiti is illegal. It was illegally painted on his wall. If he can't deal with it right now, or ever, it doesn't make him an honorary member of MS-13.
The extension of this is people moving to obfuscation and encryption. Must Facebook now hire an army of cryptographers/stenographers to crack every bespoke cipher and find every hidden message encoded on their platform? Facebook may have deep pockets, but these expectations don't scale for anybody else in the same position. They'd be better tooled than the NSA, and I don't think anybody actually wants that.
It sucks, but at least attribute blame where it's deserved-- to the people posting and acting on threats.