22 Mar Exactly why cannot respondents (or software) have a look at a Black individual and a White person and inquire, a€?what type was homosexual?
The AUC = .91 doesn't imply that 91% of homosexual guys in a given society are identified, or the classification results are correct 91percent of that time. The results of the classifier is dependent on the desired trade-off between precision (age.g., the tiny fraction of homosexual everyone the type of labeled as homosexual) and remember (elizabeth.g., the tiny fraction of gay folks in the population correctly identified as gay). Targeting large precision decreases recollection, and vice versa.
Relating to their writers, who say these people were a€?really disturbeda€? by their own results, the accuracy of an AI system can attain 91 % for homosexual men and 83 percent for homosexual girls
They go on to bring a technical, and that I think deceptive example. Men and women should comprehend your computers got usually choosing between a couple, certainly one of whom had been defined as homosexual and the more perhaps not. They have increased amount potential for obtaining that choice appropriate.