Your exploration is going away from my initial question, why do goals get
differing priorities.
But I'll bite: We can consider a situation to be a list of elements as follows:
(prototype Situation :Items { } )
Your question: How with the AGI program differentiate the overgeneralizations
that it will tend to develop as it responds to various situations.
Let's develop a framework with which to answer it.
Let's assume an action has preconditions and post conditions.
(prototype Action :Preconditions { } :Steps { } :Postconditions { } )
For now, we'll ignore the steps and postconditions and focus on the
preconditions.
So now we can say that a precondition represents a situation just in case the
precondition items are the same as those in a situation.
(given [Situation ^ ?S :Items ?items] [Action ^ ?A (same
@:Preconditions ?items) ] do (print 'the preconditions of
action ' ?A ' match the situation ' ?S) )
The same goes for the postconditions of action S
Suppose we have a set of situations. And a set of actions.
We can say that the items in the precondition of an action are more concrete
than the items in a situation if the the precondition contains more items than
the situation.
[Situation ^ situation_1 :Items { quick brown fox } ] [Action ^ Action_1
:Preconditions { quick brown fox jumped } ... ]
We can say that the items in the precondition of an action are more abstract
than the items in a situation if the the precondition contains fewer items than
the situation.
[Situation ^ situation_1 :Items { quick brown fox } ] [Action ^ Action_1
:Preconditions { quick brown } ... ]
We can say that the items in the precondition of an action are more specific
than the items in a situation if the the precondition contains items that are
lower in a taxonomy than those in the situation.
[Situation ^ situation_1 :Items { quick brown fox } ] [Action ^ Action_1
:Preconditions { quick dark-brown fox } ... ][Type :Subtype dark-brown
:Supertype brown ]
We can say that the items in the precondition of an action are more general
than the items in a situation if the the precondition contains items that are
higeher in a taxonomy than those in the situation.
[Situation ^ situation_1 :Items { quick brown fox } ] [Action ^ Action_1
:Preconditions { quick brown animal } ... ][Type :Subtype fox :Supertype animal
]
So can you please refine your question at this point, so that it may be
answered.
How will the AGI program differentiate the overgeneralizations that it will
tend to develop as it responds to various situations.
Cheers,
~PM
Date: Tue, 30 Sep 2014 20:53:01 -0400
Subject: Re: [agi] Are all goals created equal?
From: [email protected]
To: [email protected]
On Sun, Sep 28, 2014 at 12:17 AM, Piaget Modeler via AGI <[email protected]>
wrote:
You create an AGI and endow it with an initial set of actions and needs. You
prioritize the needs so that, for example, power is more important than signal
strength. Beyond that, you plug it in and let it run. The AGI begins to create
goals and subgoals in order to satisfy its needs.
But how does it prioritize these goals and subgoals? We can distinguishurgent
from important goals by creating separate attributes for urgencyand priority.
But why would one goal get a higher priority than another, aside frominheriting
its priority from a basic need? I think the terms that you are using to
characterize the problem are over-generalized. I guess that when I complain
about the necessity of examining questions like this with more detailed
characterizations that people who understand what I am trying to say simply
dismiss it as obvious. The AGI program would be designed to use (and to learn
to use) actions under appropriate conditions and every programmer 'gets' what
conditional actions are.. But perhaps the problem that I see is that we need to
write these programs so that they will use broad generalizations (just as,
according to my point of view, you wrote your message using broad
generalizations.) So then the question of how the AGI program will
differentiate the over-generalizations that it will tend to develop as it
responds to various situations becomes the more essential question. So, for
example, trying to get a better understanding of a particular situation and
using that to further shape the characterization of that 'kind' of situation
may become a prioritized goal just because it may be (or seem like) a necessary
step that the program needs to take to better understand the situation. To add
a little to this view of how an AGI might work notice that a situation may be
characterized as being like different -kinds- of situations. This basic insight
is something that I never read in other people's comments - Is that because it
is so obvious? A situation is comprised of the 'components' of the situation
and each of these may lead to the development of different kinds of situations
which the particular situation may be likened to. And the situation may belong
to other categories that are based on the projection of some other kind of idea
onto it.Jim Bromer
On Sun, Sep 28, 2014 at 12:17 AM, Piaget Modeler via AGI <[email protected]>
wrote:
You create an AGI and endow it with an initial set of actions and needs.
You prioritize the needs so that, for example, power is more important than
signal strength. Beyond that, you plug it in and let it run.
The AGI begins to create goals and subgoals in order to satisfy its needs.
But how does it prioritize these goals and subgoals? We can distinguishurgent
from important goals by creating separate attributes for urgencyand priority.
But why would one goal get a higher priority than another, aside frominheriting
its priority from a basic need?
Perhaps I lack imagination.
Your thoughts?
~PM
AGI | Archives
| Modify
Your Subscription
AGI | Archives
| Modify
Your Subscription
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com
