Earlier quoted context omitted.
> That's the problem with these algorithms that humans don't watch over. ... and Facebook/Twitter profit from it and even promote it. Don't be fooled.. Someone is spot checking the training data against the output at a statistically valid sampling interval.. ML in this case is the data equivalent of a limited liability corp..
Why does someone need to spot check the training data against the output? I'd assumed the algorithm is trying to maximize a metric, which is best completely automated without human intervention to slow it down or provide biased opinions. Even fuzzier goals like 'not be racist' would get turned into metrics (eg. proportion of words used that are on this watchlist or similar scoring algorithm)
because one wants to make sure ones program is operating as intended?