Re:Google workers discussed tweaking search after travel ban
by _dps on Friday September 21, 2018 @03:56AM
(#18037964)
These are different uses of the word "bias". In the statistical sense there is an objectively true right answer and "bias" refers to the difference between a model's average answer and the average true answer. It has little to do with the lay use of the word to mean the application of an inappropriate subjective value judgement in a situation with no one true answer (unless you take an extremely anthropomorphic view of statistical formulas).
Re:Google workers discussed tweaking search after travel ban
by humanrebar on Thursday September 20, 2018 @10:27PM
(#18036789)
The difference is how much practical diversity we'll have in models. With enough effective choice, the severity of biased models goes down, at least somewhat. Models biased in different directions can provide a variety of perspectives.
This is like being mad at a company for using L1 regularization