Tuesday, February 22, 2005

Reward Based Motivation

MelbournePhilosopher

Reward based motivation is something that gets discussed over at Philosophy, etc quite a lot. The author of that blog holds against the idea of all motivation essentially "trickling down" from any single motive force, Grand Narrative, or Categorical Imperative. I more or less buy into his ideas about human motivation being more of a web of motivations rather than like a tree.

I have also been discussing AGI (Artificial General Intelligence) over on the SL4 list, where the same problem arises. How do you build an Artificial Intelligence? How can an AI develop opinions and ideas without a "reward" from some imperative in order to let it know that it is doing the "right" thing. What comes first, goal or intelligence? Is there really a difference between ends and means? Some challenges :

If you were writing rules for an intelligence, how would you do the following?

1.) Implement a structure that will give an AGI a way to know when it has formed proper ideas - i.e. knowledge of when to re-enforce past behaviour etc. To what extent CAN an AGI be a blank slate (a la cognitive science), and to what extent must it be directed?
2.) How might you include a premise or idea (such as morality) that you would like to have formed in an AGI? How do you keep an idea invariant?
3.) Should an AGI be able to modify its' "supergoals" - i.e. the thing by which the AGI estimates the value of its other goals? Would you write something which had its sub-goals feeding back up the chain?
4.) What about human reasoning and thinking do you regard as an advantage, and ditto for a disadvantage?

Cheers,
-MP

1 Comments:

Blogger David said...

Hi, I was just out blog surfing for detailed info on self esteem when I ended up on your page. Obviously I ended up a little off base, but I am certainly glad I did. If you wouldn't mind, I would like to post your link on my "favorites" page. Should you ever need it, there is valuable information on my site about self esteem.

10/02/2005 10:45:00 AM  

Post a Comment

Links to this post:

Create a Link

<< Home