• Question: What reward motivates computers to try and get the right answer?

    Asked by Ira on 9 Jan 2020.
    • Photo: Maja Popovic

      Maja Popovic answered on 9 Jan 2020:

      Now I have to say something plainly: computers are *stupid*.
      They don’t get any reward, and they are not doing anything for a reward or motivation.
      They do only what people tell them to do, whatever people tell them to do.
      And people are telling them what to do in the form of computer programs (this is what the computers understand).

      So, if the program is well designed for the answer you want, you will get the right answer whenever you ask for it.
      If it is not, however, then either the answer will be wrong, or the computer will not know what to do.

    • Photo: Sreejita Ghosh

      Sreejita Ghosh answered on 9 Jan 2020:

      I do agree with Maja.
      The part in the design which makes a computer ‘learn’ is something called a ‘cost function’. We use an aspect of Maths called optimisation (now computer scientists and Mathematicians have slightly different definitions of optimisation and I am using the Computer scientist definition) to ensure that the cost function value is as low as possible. Whenever the computer makes a wrong choice the cost function value increases and the optimizer makes the computer ‘re-learn’. We also set limit on how many times the computer program can be made to learn (when it gives a wrong answer), before “giving up on it”.

Comments