Yep. In some cases, Google has addressed this by Balkanizing their rulesets. Sometimes semi-literally; the territorial boundaries Google Maps shows are contingent upon the request's point of origin in some disputed regions [https://www.npr.org/sections/thetwo-way/2014/04/12/302337754...]

Using such an approach on the global channels like YT comments isn't an option (unless they choose to Balkanize them, and let only Chinese commenters talk to Chinese people).

Google isn't the only firm that does this; Twitter will selectively filter some tweets based on whether it's legal to show that content in countries that have outlawed fascist agitation (which makes for an interesting side-effect; doing a request for a tweet stream against a US proxy server and a German proxy server and delta-ing the results can give you a clear signal on "Twitter thinks this person is a Nazi propagandist." I'm wondering when someone will put an API in front of that and make it a service... "DoesTwitterThinkThatGuyIsANazi.com" anyone? ;) ).

Sure, but that only works if countries' laws find that acceptable. If China says that you have to delete something from US users if you want to do business in China, they have every legal right to do so: As the sovereign ruler of the land, they decide who does business there, and they can set any requirement they see fit.

So Balkanized rulesets might work for some countries but not others. China is well known for exercising their market size to make companies jump for them, the US is as well.

>As the sovereign ruler of the land, they decide who does business there, and they can set any requirement they see fit.

At the potential cost of a trade war of course.

Of course. But nobody is going to war over censored YouTube comments or delisted search results, and the Chinese government is happy to push that boundary.

> Using such an approach on the global channels like YT comments isn't an option (unless they choose to Balkanize them, and let only Chinese commenters talk to Chinese people).

That doesn't seem unreasonable. They could hide filtered comments from chinese users. This would allow the rest of the world to be unaffected by chinese regulation and "allowed" speech between chinese and foreigners.

But this is a case of Chinese speech in the Chinese language. If I used the English phrase “communist bandits”, YT presumably wouldn’t care. YT is applying Chinese regulations to Chinese comments written in Chinese.

People outside china might be writing in chinese. Should the CCP be able to regulate speech of all chinese-speaking people globally? Should the UK be able to regulate all english speech globally?

Right and exactly. My guess is that this phrase has actually been trained into the ML algorithm as a "no-go" by the community (not by direct action from Google) because it's primarily used in contexts where Chinese nationals or national-sympathetics flag it. If the phrase isn't generally used in contexts that don't get flagged, the algorithm would learn quickly that "nobody" wants to see it.

"by the community" is the smoke and mirrors here. There is not a single unified community.

No, there isn't, which is why the ML system can be fooled (especially by phrases that only have meaning for a subset of users; most comments on YT aren't in Chinese, so any isolated Chinese phrase sticks out like a sore thumb in terms of Bayesian priors and first-impression effects on the ML algorithm will be strong. The system will interpret "Most users never see this phrase but it gets flagged often when users do see it" as "This is objectionable content and should be downed automatically").

> People outside china might be writing in chinese. Should the CCP be able to regulate speech of all chinese-speaking people globally?

I can see why they would want to do that and I could see why YT would go along with their demands to do so. I didn’t say I agreed with it.