The mistake a lot of companies are making is they view ML as a replacement for humans. It's not. At least, it's nowhere near that now. You should view ML as an

The mistake a lot of companies are making is they view ML as a replacement for humans. It's not. At least, it's nowhere near that now. You should view ML as an enhancement for humans. So the problem here is that no ML system without essentially being an AGI is capable of fully comprehending context, in this case use of "black" and "white" in Chess that has absolutely no racial context. So the mistake here is (likely) automated action against the channel. That should never happen. At least, not for this particular signal. What should happen is the channel should be flagged for review by a human. Any human would appreciate the context and take no action. This is the "enhancement" part. With no automated assistance, a human might be capable of moderating (made up numbers) 100 channels. That doesn't scale at Youtube scale. ML systems might enhance that human's moderating ability to 10,000 channels. Even better, by flagging for review, you're actually producing training data for your ML. The P-R for your automated disciplinary system needs to be incredibly good before you take humans out of the equation. False positive bans or blocks on channels is bad for the individual channel. It can be devastating in fact. The corporate view is that "one channel doesn't matter". And they're right... to a point. At some point it undermines confidence in the platform and then it does matter.