38

This is deep !

Comments
  • 1
    Well you need to account for mental disorder when training a model. As I remember that was what Joker had.
  • 10
    I think the bigger issue is that machine learning will learn to use "how it looks like" instead of "what it looks like".

    But seriously, neither humans nor machines will be able to detect depression reliably.

    That's why it's so incredibly important to just tell people, straight up: "I have been feeling quite empty lately, depressed. I wanted to let you know, not because I expect you to fix it, but because you are a big part of my life, and sadly so is this state that I'm currently in".

    Really. If you feel depressed, the most important step, more important even than therapy, is to inform family and friends.
  • 3
    And neither can you identify issues by looking at someone's face.
    This is dumb.
  • 14
    Machine learning is great! Look at the results I got:

    [This post] Cringe: 100%

    So accurate!!
  • 1
    @PrivateGER I agree, but look at the works of Paul Ekman concerning FACS(Facial Action Coding System) it is some really interesting stuff concerning determining the state of mind based on microexpressions and body language.
  • 0
    @AleCx04 You have to monitor people for days or weeks to accuratley detect those things + every human has different reactions and AI propably dosent know your previous behaviour so it cant compare anything
  • 1
    @Gregozor2121 I also do not disagree with this! My point was mainly that studies exist for this sort of thing(being able to detect certain patterns in human faces), not if it was applicable for AI to work through it. Even then I do not see it as a far fetched idea in terms of actual application. Ekman himself stated in some of his works that this is a long process(there was a tv show for it called Lie to me that made it look like if it was something instant! right there and then) wouldn't you say that using a machine learning models to make annotations of this in a case by case basis would be really interesting? As in, every patient would be its own dataset, and the learning algorithm would be of the supervised learning type, one would see if the algorithm was giving the correct results over a particular model and just fine tune as it goes unless it was properly in accordance to the expert's viewings.
    At least to me this sounds as a fascinating academic experiment.
  • 1
    The issue with this is merely that the music sucks ass.
  • 0
    @AleCx04 Yeah. Ai just cant fix the lack of data, it isnt magic, less data == shittier result.
    Even normal agos struggle with that.
Add Comment