It's not just ML, but the people that provide the labeling for the ML.
Google pays some large number of people to do search and grade the various results they get to see if the answers are good, which then helps feed back ML.
Heck, according to this article[0], google has been paying people to evaluate their search results since 2004.
It doesn't feed back into the ML directly, according to Google. Instead they use it to evaluate changes to search algorithms. If they get an increase in thumbs up back from the Quality Raters then their changes were positive. If not, they figure out why.
The original 2012 FTC investigation of Google anti-trust activity showed how they might have abused this process. Interesting read, no matter which side you take: http://graphics.wsj.com/google-ftc-report/
Google pays some large number of people to do search and grade the various results they get to see if the answers are good, which then helps feed back ML.
Heck, according to this article[0], google has been paying people to evaluate their search results since 2004.
[0] https://searchengineland.com/interview-google-search-quality...