Results 1 to 7 of 7

Thread: VI using AI

  1. #1
    jughead's Avatar
    Join Date
    Oct 2008
    Location
    NC
    TIM
    IEI
    Posts
    886
    Mentioned
    3 Post(s)
    Tagged
    0 Thread(s)

    Default VI using AI

    So now that we have deep learning AI it should be pretty simple for us to plug in the types into a AI database that could learn and type anyone from there on out. That is all. Video>static pictures.

  2. #2

    Join Date
    Jun 2008
    Posts
    10,557
    Mentioned
    953 Post(s)
    Tagged
    3 Thread(s)

    Default

    If there will appear algorithms which model nonverbal of concrete people - then sure, it can be used for VI.
    There is a principle abbility, at least. Today they analyse similarity of faces, what is close. Just needs more details and more data to process - sequences of pictures.

    The more accurate method mb brain scan, - which would show which conscious patterns dominate at people of different types. Also mb this may check IR effects too - which people we like more and such to additional check the real type.
    Types examples: video bloggers, actors

  3. #3
    ouronis's Avatar
    Join Date
    Aug 2014
    TIM
    ref to ptr to self
    Posts
    1,353
    Mentioned
    61 Post(s)
    Tagged
    2 Thread(s)

    Default

    Not simple because you need adequate data - labels that mean something and pictures with one person with mostly uniform distances, angles. The less data you have the more uniform the samples need to be or you will learn a bunch of useless stuff

  4. #4
    What's the purpose of SEI? Tallmo's Avatar
    Join Date
    May 2017
    Location
    Finland
    TIM
    SEI
    Posts
    2,049
    Mentioned
    146 Post(s)
    Tagged
    0 Thread(s)

    Default

    I assume AI can see the difference between men and women. How did it learn that? By being fed millions of pictures of people, or is there an explicit algorithm?

    Let's say we interview 10 million people and type them by hand accurately. Then we put the pictures into a computer together with the corresponding types. Then the computer suddenly knows how to type people based on that information. And we don't even know exactly how it learned it. It's fascinating if it would work.

  5. #5
    nefnaf's Avatar
    Join Date
    Dec 2013
    TIM
    LII-Ne
    Posts
    207
    Mentioned
    8 Post(s)
    Tagged
    0 Thread(s)

    Default

    The human brain is specifically adapted to extract the maximum possible information from human faces and gestures. We can write computer program that can best any human in Chess or Go, easy enough. But this type of project would require an ungodly amount of high-quality training data, that could only be provided by groups of well-trained socionists. And even if the project got that far, there is no guarantee that computer system would even be any good. What you describe as "pretty simple" is actually a pipe dream within a dream.

  6. #6
    xerxe xerxe's Avatar
    Join Date
    Dec 2007
    Location
    Ministry of Love
    Posts
    6,215
    Mentioned
    87 Post(s)
    Tagged
    0 Thread(s)

    Default

    Quote Originally Posted by jughead View Post
    So now that we have deep learning AI it should be pretty simple for us to plug in the types into a AI database that could learn and type anyone from there on out. That is all. Video>static pictures.
    You need to train the AI using tens of thousands of images of different people posing in the same way. Each of these people must already be correctly typed.

    > We don't have tens of thousands of samples.

    > Even if we did, we'd never agree on their types.

  7. #7
    ouronis's Avatar
    Join Date
    Aug 2014
    TIM
    ref to ptr to self
    Posts
    1,353
    Mentioned
    61 Post(s)
    Tagged
    2 Thread(s)

    Default

    Quote Originally Posted by xerxe View Post
    You need to train the AI using tens of thousands of images of different people posing in the same way. Each of these people must already be correctly typed.

    > We don't have tens of thousands of samples.

    > Even if we did, we'd never agree on their types.
    Yes

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •