[Rant] Can AI read your emotions? (No, but ...)

we need to talk about your face or face recognition in general a tweet has been making the rounds saying facial recognition is able to analyze in real time the emotions and feelings just that and it showed a video of a apparent real-time system looking at people's faces and determining what their emotions are now there is a predictable reaction of machine learning twitter with respect to anything to do with facial recognition and that reaction is no the biggest reaction is no this is impossible ai will never be able to infer your emotions by looking at your face this is the data is not there anything like this i just think that is really really surprising honestly now look facial recognition technology isn't exactly the most popular subject it's not going to win any nobel peace prizes anytime soon is this technology dystopian looking yes is it dangerous in the wrong hands yes does it work as advertised very probably no is it easy to be tricked absolutely yes however saying that it is impossible for an ai to look at your face and infer your emotional state that is wondering me you do this every day you look at people's faces and then you infer something about their internal state people splitting hairs here about the word analyze to analyze the emotions and feelings well if you want to split words i would say inferring is a lot heavier than analyzing your face has literally evolved to convey your internal state other people have a trouble with saying well you can fake your face not all facial expressions can be faked a lot of what you tell with your face is involuntary and there is in principle not a reason why a machine cannot pick up on these cues now this is not to say that this particular system works well it probably does not it is extremely hard to do this to look at a face and get how that person is feeling through all the deception that might be there is an extremely hard task but there's nothing supernatural about it we do this we're a machine ergo a machine can in principle do this the most criticism i see right here is that well the machine only analyzes facial expressions they have nothing to do with your emotions and feelings what is that of course this has something to do with your emotions and feelings have you ever thought to yourself huh that person looks kind of sad today have you ever gone to someone and said you know you look a little bit down is everything okay no never and you certainly didn't infer this from their face hey doctor i have a problem well what's your problem well i banged my foot and now it hurts and it has a dent in it bleeds and it's swollen and everything is bad about my foot because i hit it and it might be bro well don't say it's broken because the external symptoms will never tell us anything about the internal state of a system i'm sorry have you ever heard that an ai can diagnose lung cancer by looking at a chest x-ray well no all we can say is just that the ai detects a little bit of a spot there is no correlation at all this is no indication of the internal state of the cancer shut up twitter makes it such that everyone immediately is extreme on the one side and extreme on the other side instead of saying the data to train the system is very hard to get the systems itself aren't as good they don't understand context that this happens in or nuances that's very different from saying that no this is impossible the most ridiculous is when people come out and compare this to phrenology or literally call it phrenology you know phrenology the science of what bump on your head means something about your personality or intelligence like my face has literally evolved to tell you something about my internal emotions none of the bumps on my head have evolved to communicate about my intelligence it is a predictable reaction for some reason anywhere where facial recognition technology is used

used there is a crowd of people coming out saying phrenology faces are a real thing emotions there is a real connection between your facial expression and your emotions it is more complicated than these machines right now can assess it might require more context more data better algorithms and even things we don't have yet but this definitely exists it is not a pseudoscience not everything that has to do with face recognition is a pseudoscience it might be dangerous yet it's real so in conclusion i guess my message here is that yes this is probably an over promise of what a. i can do and it could easily be used for bad purposes on the other hand this is not a pseudoscience this is not impossible and research in this direction might actually lead to something good imagine an ai that is better than a human at recognizing emotions from someone's face assuming that is possible we could avoid a lot of conflict maybe do a lot of good work in suicide prevention and ultimately communicate with the ais as we would with other humans apart from all the bad thing that we can do with facial recognition technology ultimately its technology can be used for good and for bad and for evil i'll end with the holy trifecta of broader impact statements technology good technology bad technology biased peace out you