Hacker News Reader

Comment by kodablah | Hacker News Reader

I think the argument is framed as wanting a human during take down dispute, not pre-approval. There's no need to treat this as an absolute. Why are human contact and reasonable expediency considered mutually exclusive? There is a lot of room in the middle. Can allow human dispute resolution for non-free accounts (i.e. paid or earning). Can limit upload frequency site wide while dispute/abuse queues are backed up (knowing YouTube would be incentivized to prevent backup). Can have non-profit reviewer company that charge very minimal amounts for human review. Etc.

Let's be clear, the lack of humans anywhere in the pipeline for uploaders is because they don't have to, not because it is impossible. Having said that, I really hope we do not see government intervention here.

View on HN

Agree wholeheartedly: there's definitely a spectrum between what Google are doing atm and full human moderation. I am a strong believer in using AI to augment humans, not replace them. Nudging Google towards putting more humans in the pipeline is what I am hoping posts like this are achieving.

When was the last time Google listened to the public? In spirit, not because there was a backlash? It’s a techno conglomerate with no human face. Google was cool and interesting in 00’s, but nowadays all they create is fodder for those sweet ad dollars.

> Let's be clear, the lack of humans anywhere in the pipeline for uploaders is because they don't have to, not because it is impossible.

The real problem here is that people are envisioning human interaction as something that can produce better outcomes, but the cost of that may not be economical.

You can hire a generic human to look over a flagged video for 15 seconds and render a verdict, but the verdict is going to be bad. That's not enough time to make an accurate evaluation.

Suppose you have an epidemiologist on YouTube making a claim about virus propagation. It gets flagged because there is a CNN article reporting that the CDC said something contradictory. The epidemiologist argues that the CNN article is a misinterpretation and if you look at the original CDC publication, that's not really what they said.

At this point for YouTube to get it right they've basically got to hire their own M.D. to study the CDC publication in enough detail to understand the context. That is prohibitively expensive. But the alternative is often censoring an epidemiologist who is trying to correct misinformation being disseminated by CNN.

And science is a moving target. The infamous example was the CDC in 2020 telling people not to wear masks. But there was a point in time when scientists were publishing studies on the health benefits of tobacco consumption. Challenging the established orthodoxy is inherently necessary for progress.

The underlying problem here is that YouTube isn't competent to make these determinations. They're attempting to do something they have no capacity to succeed in.

So true. Automation cannot infer nuance. It's ok to strike down obvious spam and troll but the but this pretense that private actors should spend their resources to cultivate a sane public space in just a broad-spectrum reflex reaction to micro-targeted disinformation. Nonsense.

What stops the YouTube reviewer from calling or emailing the CDC to render a verdict on which interpretation of their public information is accurate?

When it comes to government agencies, they are contactable (usually with a simple email). You might not get an immediate response, but a few hours to a few days later you can expect to hear back.

Why would youtube be trying to make a determination like that? Individuals are the arbiters of truth, not organizations. Ministries of truth are not desirable.

What's with the assumption here that there aren't humans involved in this decision? The linked post literally quotes Google saying:

"Hi Sun Knudsen, Our team has reviewed your content, and, unfortunately, we think it violates our harmful and dangerous policy."

Google has always had humans in the loop. Not for every decision but certainly the more impactful ones that affect user generated content. When I worked there, there were even humans reviewing requests for account un-suspensions (yes, really! they didn't do a great job though, needless to say).

Also, it's quite hard to understand what sort of algorithm could result in this sort of suspension. What training data would give it hundreds of thousands of examples of infosec discussions labelled as harmful? But it's very easy to see how a large army of low paid humans given vague advice like "don't allow harmful or dangerous content" could decide that linking to hacking information is harmful.

The problem here is not AI. The problem here is Google allowing vague and low quality thinking amongst the ranks of its management, who have decided to give in to political activists who are determined to describe anything they're ideologically opposed to as "harmful", "dangerous", "unsafe" etc. When management give in to those radicals and tell their human and AI moderators to suspend "harmful" content, they inevitably end up capturing all kinds of stuff that isn't political in nature, simply due to the doublespeak nature of the original rules.

"Our team" is quite often a lie. Why do you assume that a human was involved? Especially when it comes to Google, who notoriously skimps on human support staff.

They lie both ways too, whenever there is a public kerfuffle they always blame the algorithm / the AI.

It's a bad idea to give liars the benefit of the doubt

Because I worked there and back then, when Google claimed content had been manually reviewed, it had been. It might have been by a low wage worker in India or Brazil and they might have spent 20 seconds on it, but there was a person in the loop.

That's why I turned this around. Why do you assume they're lying, without evidence?

Why wouldn't we assume they are lying? Interacting with their dispute process doesn't have any human qualities, so what would make us think that humans are involved?

They are a massive monopoly and just the other day it was reported that they manipulated their own ad markets for their own profit. They constantly inaccurately accuse people of malicious intent and copyright infringement, often massively hindering peoples' careers and generally providing zero recourse. To not assume lying and at best self-serving intent given Alphabet/Google's position in the world would be foolish.

On top of that, is it really that different to use inaccurate AI or to use humans who are intentionally not given the time or tools to do human-quality work?

The problem is obviously with the overall quality of moderation, and the level of power exerted by platforms over users. Whether a non-skilled human reviewer or an AI is handling the broken process hardly matters given the results.

Choosing to die on that hill of technicality seems like a silly choice. People say "human" but they really mean "a person who diligently considered the content, probably watched the video more than once, understood what they were watching, and possibly has a chance to debate with another human". Or, you know, a reasonable definition for "human in the loop".

It seems you're defining "human in the loop" as "Google makes what I consider to be the correct decision in this case." I believe the point that thu2111 is trying to make is that there _are_ humans in the loop, and having them is no insurance against unilateral decisions that you disagree with or even just plain stupid decisions.

The point is that involving more humans in the moderation decisions at YouTube does not guarantee nor is there any evidence to suggest, that it would result in better moderation.

Somebody with the technical and industry knowledge to know that talking about security vulnerabilities like this shouldn't be removed can probably find a better job that YouTube video moderation.

I would accept that Google will sometimes make the wrong decision, but it's unacceptable that the care given toward making that decision stops at a low-wage worker in another country spending 20 seconds on the decision.

There's a very wide chasm between "an AI decided" or "someone spent 20 seconds on it", and "we hired an expert in every possible field and they thoroughly research the facts at hand before making a decision".

The latter is obviously not feasible, but there are many options in between that will give markedly better outcomes than the status quo while not making YT moderation economically infeasible.

There are plenty of cases where I disagree with the outcome of companies I interact with. When I spend 30 minutes chatting with an agent via text with Amazon or airlines, I might leave the conversation disappointed, but I never doubt their conscientiousness. I highly doubt United beat me at the Turing Test.

Google makes what I consider to be the correct decision in this case.

I'll settle for "Google makes a well-informed decision and will communicate the weighted pros and cons that led to that decision."

> What's with the assumption here that there aren't humans involved in this decision? The linked post literally quotes Google saying [...]

The post literally quotes Google as saying that for the _last_ time this happened. Then the post literally states they appealed and never heard back. There is nothing concerning a human for this instance, though I suspect once the uploader reaches out (were it not for the popularity this is gaining), one could expect similar dismissal. That there are barely humans involved, and in a poor way when they are, doesn't mean there are never humans involved, true. But for many, the bare minimum of assistance is indistiguishable from no assistance.

Appealing and not hearing back doesn't mean the original decision was entirely automated, it means they ignore appeals. Plenty of people would love nothing more than to argue with Google employees all day about every decision the firm makes - just think about webmasters and SEO alone - so the fact that YouTube actually has such a process is already kind of remarkable.

You seem to be arguing that because their dispute process everywhere else is reprehensible, a slightly-less-reprehensible process should be celebrated.

It shouldn't, and we need to hold Google and others to a much higher standard here. I'm not sure how we do that, but the status quo is garbage.

It's easy in theory: you create a competitor that differentiates through the quality of its content-provider service quality and how much money is spent on dispute resolution.

But I am skeptical it would be successful. It might be, but it would probably require a much larger percentage take of video revenues, which would discourage most YT content creators from going there because in fact, relative to overall YT volumes, these suspensions are rare. That's why we're discussing individual cases instead of studies with percentage figures in them.

It feels a bit weird to be playing the devil's advocate in this case because frankly the state of modern Google saddens me. It seems not much like the firm we built back in the day. They obviously shouldn't be doing this kind of thing and in the past they didn't, which is why I place the finger of blame not on AI (which may not even be at fault here, and is a general tool anyway) but rather on bad ideologies and ideas permeating the management classes in America.

In this case I blame favouritism and feminism. YouTube content moderation is run by Wojcicki, a woman who got in to Google by renting Larry and Sergey a garage, by being the sister of Sergey's wife. She's not an engineer. Thanks to the prevailing ideology an ambitious feminist woman in an executive position at a west coast tech firm is utterly untouchable, and must be given whatever she wants. So despite being the exec in charge of Google Video when it was utterly beaten by YouTube, she has ended up running YouTube. A clearer case of failing upwards can't be imagined. It's her personality that dictates the post-2016 swing towards promoting woke "authoritative content" on YT and the general culture of unaccountability for women that prevents anyone else pushing back, even as the bad PR racks up.

Knowledge is empowering, and one can also argue that lack of it is considered harmful. Not being able to link to resources that can help you understand how something works can lead to (in the big picture) that only criminals will gain access to that knowledge.

Louis Rossman had his YouTube account temporarily suspended for posting a video entitled “An angry clinton” which was a video of his pet cat Clinton meowing. Human moderation is needed in cases like this. There is obviously no analysis of the video content, the algorithm looks for keywords in the video title. The keywords used by the algorithm are centered around protecting Democratic politicians from criticism. There is no advanced AI going on, the algorithm is simple pattern matching.

Reply on news.ycombinator.com