Slashdot Mirror
Researcher Trying To Teach Computer What Women He's Attracted To
jfruh writes: Harm de Vries, a post-doctoral researcher at the Université de Montréal, is trying to build an algorithm that will sort through pictures on Tinder and OKCupid and pick out women he'll find attractive. "Tinder kept giving me pictures of girls I wasn't attracted to," he said in a phone interview. "So I wondered if I could use deep learning." His program, built using deep learning techniques, has about a 68 percent success rate, which isn't that bad. (A human friend to whom de Vries described his preferences managed 76 percent.)
I'm sure if it was a woman looking to filter men this project would be lauded as 'empowering.'
actually the idea is to like the looks first, then talk to her. then reject the ones you dislike based on personality
so you get physical attraction and personality compatibility. both
if you just talk to girls and find someone you're compatible with on a personality level but you don't want to have sex with because they're physically unattractive to you, the term for that is "friend" (as an aside, many, many men posting on this site probably know this zone well)
what you call shallow is actually called mate selection. if you can't get an erection, you're not going to procreate. mate selection is not friend selection. friend selection is a different topic. to not realize the difference is... shallow, ironically
intellectual property law is philosophically incoherent. it is your moral duty to ignore it or sabotage it
I'm sure if it was a woman looking to filter men this project would be lauded as 'empowering.'
Tinder and similar apps are already doing this, both for men and women, and nobody but passive-aggressive gamergaters are bringing the misogyny discussion into this. Calling it shallow is fair, both for men and women, but is something a significant portion of both genders do.
Neither the summary nor the linked article provide the necessary statistics to tell us how well this algorithm actually works. We're told it has a 68% success rate, which presumably means that 68% of the time it gives the same answer as de Vries (the human subject/programmer).
The problem is, we're not told anything about the sensitivity or specificity of the technique. What is the rate of false positives? False negatives?
Let's say that de Vries typically finds 1 out of 3 (33%) of the profile pictures "attractive". His computer could score 67% accuracy just by rejecting every single picture. (Such an algorithm would have zero sensitivity, but perfect specificity, and a terrible false negative rate. The "reject-everything" algorithm also scores better the more picky de Vries gets.)
This sort of story is only interesting if it includes specific information about where and how his algorithm fails (and succeeds).
~Idarubicin