By using this site, you agree to our Privacy Policy and our Terms of Use. Close
GProgrammer said:
vivster said:

Try feeding it with pictures of homosexual people that contradict "typical" facial features of homosexuals and see the accuracy plummet. Since there are absolutely no hard physical clue for someone's sexuality this algorithm's gaydar is just as good as anyone else's.

your sentence 1 contradictions your sentence 2

In what way?

Feeding it more varied pictures  will bring the success rate down to the level of a normal human to about 60% I'd say. And when I say hard clue I mean a clue that will let you detect with 100% certainty the sexuality of a person. Since nothing like that exist we can only go by soft clues. Which means the decision basis and the guesses of an algorithm will be the exact same as a human's if fed with the same varied material. The machine in this example was obviously fed with extremely biased samples. Deliberately or not but this experiment is very much flawed.



If you demand respect or gratitude for your volunteer work, you're doing volunteering wrong.