The obsession with "machine learning" is actually making systems dumber. Google Search and Gmail spam filters are getting worse with each passing week, and I am almost certain the increasing reliance on ML is to blame.
I chalk it up to a cost benefit calculation. Google clearly isn't trying to eliminate all spam in search. It's not their goal. They are not trying to optimize for the user experience. They're trying to optimize revenue.
They're trying to keep spam out of my inbox, and the spam rate has been increasing for me (and other HN commenters who frequently talk about it)
The competing explanation is that "machine learning" is actually making spam generator systems smarter, so spam gets harder to detect.
The type of spam I see in my inbox is anything but smart. But I agree that this has always been a cat and mouse game.