4 ms·
The limitations of machine learning aside, this could actually work quite well if trained properly, because it wouldn't raise the questions that human choices d
by JoshTriplett 15y ago
The limitations of machine learning aside, this could actually work quite well if trained properly, because it wouldn't raise the questions that human choices do about bias. Given sufficient information, a machine-learning algorithm can make decisions based on factors that humans would refuse to consider or would intentionally err on the side of not considering.
- StavrosK 15y agoYes, of course. It could also disregard features completely if it has found that they make no statistical difference.
- JoshTriplett 15y agoExactly. Most importantly, it can focus exclusively on statistics, rather than guesses.
- woodson 15y agoOne problem I see with this is that these statistics involve a lot of human bias. If you train your classifier on crime statistics, it's not based on a 'ground truth' (which is never attainable in a criminal case), but an expert's, i.e. judge's/jury's, decision.
- JoshTriplett 15y agoThat would indeed produce incorrect training for a classifier of "who should we pull out of line for a search". For that, you'd want to train exclusively on the observable characteristics of the people pulled out of line for a search versus whether you found something.